bump: version 0.48.6 → 0.49.0

feat(news): add Nextcloud News app integration

.dockerignore

@@ -5,3 +5,5 @@
 !uv.lock
 
 !nextcloud_mcp_server/**/*.py
+!nextcloud_mcp_server/**/*.html
+!nextcloud_mcp_server/auth/static/*

.github/workflows/bump-version.yml

@@ -15,17 +15,17 @@ jobs:
       packages: write
     steps:
       - name: Check out
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
         with:
           fetch-depth: 0
           token: "${{ secrets.PERSONAL_ACCESS_TOKEN }}"
       - name: Create bump and changelog
-        uses: commitizen-tools/commitizen-action@5b0848cd060263e24602d1eba03710e056ef7711 # 0.24.0
+        uses: commitizen-tools/commitizen-action@bb4f1df6601e2a1a891506581b0c53acdc88e07d # 0.26.0
         with:
           github_token: ${{ secrets.PERSONAL_ACCESS_TOKEN }}
           changelog_increment_filename: body.md
       - name: Release
-        uses: softprops/action-gh-release@5be0e66d93ac7ed76da52eca8bb058f665c3a5fe # v2.4.2
+        uses: softprops/action-gh-release@a06a81a03ee405af7f2048a818ed3f03bbf83c7b # v2.5.0
         with:
           body_path: "body.md"
           tag_name: v${{ env.REVISION }}

.github/workflows/claude-code-review.yml

@@ -0,0 +1,57 @@
+name: Claude Code Review
+
+on:
+  pull_request:
+    types: [opened, synchronize]
+    # Optional: Only run on specific file changes
+    # paths:
+    #   - "src/**/*.ts"
+    #   - "src/**/*.tsx"
+    #   - "src/**/*.js"
+    #   - "src/**/*.jsx"
+
+jobs:
+  claude-review:
+    # Optional: Filter by PR author
+    # if: |
+    #   github.event.pull_request.user.login == 'external-contributor' ||
+    #   github.event.pull_request.user.login == 'new-developer' ||
+    #   github.event.pull_request.author_association == 'FIRST_TIME_CONTRIBUTOR'
+
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: read
+      issues: read
+      id-token: write
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
+        with:
+          fetch-depth: 1
+
+      - name: Run Claude Code Review
+        id: claude-review
+        uses: anthropics/claude-code-action@6337623ebba10cf8c8214b507993f8062fd4ccfb # v1
+        with:
+          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          prompt: |
+            REPO: ${{ github.repository }}
+            PR NUMBER: ${{ github.event.pull_request.number }}
+
+            Please review this pull request and provide feedback on:
+            - Code quality and best practices
+            - Potential bugs or issues
+            - Performance considerations
+            - Security concerns
+            - Test coverage
+
+            Use the repository's CLAUDE.md for guidance on style and conventions. Be constructive and helpful in your feedback.
+
+            Use `gh pr comment` with your Bash tool to leave your review as a comment on the PR.
+
+          # See https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md
+          # or https://docs.claude.com/en/docs/claude-code/cli-reference for available options
+          claude_args: '--allowed-tools "Bash(gh issue view:*),Bash(gh search:*),Bash(gh issue list:*),Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*),Bash(gh pr list:*)"'
+

.github/workflows/claude.yml

@@ -0,0 +1,50 @@
+name: Claude Code
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  issues:
+    types: [opened, assigned]
+  pull_request_review:
+    types: [submitted]
+
+jobs:
+  claude:
+    if: |
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude')) ||
+      (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@claude')) ||
+      (github.event_name == 'issues' && (contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')))
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: read
+      issues: read
+      id-token: write
+      actions: read # Required for Claude to read CI results on PRs
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
+        with:
+          fetch-depth: 1
+
+      - name: Run Claude Code
+        id: claude
+        uses: anthropics/claude-code-action@6337623ebba10cf8c8214b507993f8062fd4ccfb # v1
+        with:
+          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+
+          # This is an optional setting that allows Claude to read CI results on PRs
+          additional_permissions: |
+            actions: read
+
+          # Optional: Give a custom prompt to Claude. If this is not specified, Claude will perform the instructions specified in the comment that tagged it.
+          # prompt: 'Update the pull request description to include a summary of changes.'
+
+          # Optional: Add claude_args to customize behavior and configuration
+          # See https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md
+          # or https://docs.claude.com/en/docs/claude-code/cli-reference for available options
+          # claude_args: '--allowed-tools Bash(gh pr:*)'
+

.github/workflows/docker-build-publish.yml

@@ -12,11 +12,11 @@ jobs:
       packages: write
     steps:
       - name: Checkout repository
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
 
       - name: Docker meta
         id: meta
-        uses: docker/metadata-action@318604b99e75e41977312d83839a89be02ca4893 # v5
+        uses: docker/metadata-action@c299e40c65443455700f0fdfc63efafe5b349051 # v5
         with:
           # list of Docker images to use as base name for tags
           images: |

.github/workflows/helm-release.yml

@@ -14,7 +14,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
         with:
           fetch-depth: 0
 

.github/workflows/rag-evaluation.yml

@@ -0,0 +1,105 @@
+name: RAG Evaluation
+
+on:
+  workflow_dispatch:
+    inputs:
+      manual_path:
+        description: 'Path to Nextcloud User Manual PDF in Nextcloud'
+        required: false
+        default: 'Nextcloud Manual.pdf'
+      embedding_model:
+        description: 'OpenAI embedding model'
+        required: false
+        default: 'openai/text-embedding-3-small'
+      generation_model:
+        description: 'OpenAI generation model'
+        required: false
+        default: 'openai/gpt-4o-mini'
+
+jobs:
+  rag-evaluation:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    permissions:
+      models: read
+
+    steps:
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+
+      - name: Run docker compose with vector sync
+        uses: hoverkraft-tech/compose-action@248470ecc5ed40d8ed3d4480d8260d77179ef579 # v2.4.2
+        with:
+          compose-file: |
+            ./docker-compose.yml
+            ./docker-compose.ci.yml
+          up-flags: "--build"
+        env:
+          # Environment variables passed to docker-compose.ci.yml
+          OPENAI_API_KEY: ${{ secrets.GITHUB_TOKEN }}
+          OPENAI_BASE_URL: "https://models.github.ai/inference"
+          OPENAI_EMBEDDING_MODEL: ${{ inputs.embedding_model }}
+          OPENAI_GENERATION_MODEL: ${{ inputs.generation_model }}
+          VECTOR_SYNC_SCAN_INTERVAL: "5"
+
+      - name: Install the latest version of uv
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
+
+      - name: Wait for Nextcloud to be ready
+        run: |
+          echo "Waiting for Nextcloud..."
+          max_attempts=60
+          attempt=0
+          until curl -o /dev/null -s -w "%{http_code}\n" http://localhost:8080/ocs/v2.php/apps/serverinfo/api/v1/info | grep -q "401"; do
+            attempt=$((attempt + 1))
+            if [ $attempt -ge $max_attempts ]; then
+              echo "Service did not become ready in time."
+              exit 1
+            fi
+            echo "Attempt $attempt/$max_attempts: Service not ready, sleeping for 5 seconds..."
+            sleep 5
+          done
+          echo "Nextcloud is ready."
+
+      - name: Wait for MCP server to be ready
+        run: |
+          echo "Waiting for MCP server..."
+          max_attempts=30
+          attempt=0
+          until curl -o /dev/null -s -w "%{http_code}\n" http://localhost:8000/health/live | grep -q "200"; do
+            attempt=$((attempt + 1))
+            if [ $attempt -ge $max_attempts ]; then
+              echo "MCP server did not become ready in time."
+              exit 1
+            fi
+            echo "Attempt $attempt/$max_attempts: MCP not ready, sleeping for 2 seconds..."
+            sleep 2
+          done
+          echo "MCP server is ready."
+
+      - name: Run RAG evaluation tests
+        env:
+          NEXTCLOUD_HOST: "http://localhost:8080"
+          NEXTCLOUD_USERNAME: "admin"
+          NEXTCLOUD_PASSWORD: "admin"
+          RAG_MANUAL_PATH: ${{ inputs.manual_path }}
+          OPENAI_API_KEY: ${{ secrets.GITHUB_TOKEN }}
+          OPENAI_BASE_URL: "https://models.github.ai/inference"
+          OPENAI_EMBEDDING_MODEL: ${{ inputs.embedding_model }}
+          OPENAI_GENERATION_MODEL: ${{ inputs.generation_model }}
+        run: |
+          uv run pytest tests/integration/test_rag.py -v --log-cli-level=INFO --provider openai
+
+      - name: Capture MCP container logs
+        if: always()
+        run: |
+          echo "=== MCP Container Logs ==="
+          docker compose logs mcp --tail=500
+
+      - name: Upload test results
+        if: always()
+        uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4 # v5
+        with:
+          name: rag-evaluation-results
+          path: |
+            pytest-results.xml
+          retention-days: 30

.github/workflows/release.yml

@@ -18,9 +18,9 @@ jobs:
       contents: read
     steps:
       - name: Checkout
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
       - name: Install uv
-        uses: astral-sh/setup-uv@5a7eac68fb9809dea845d802897dc5c723910fa3 # v7.1.3
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
       - name: Install Python 3.11
         run: uv python install 3.11
       - name: Build

.github/workflows/test.yml

@@ -9,9 +9,9 @@ jobs:
   linting:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
       - name: Install the latest version of uv
-        uses: astral-sh/setup-uv@5a7eac68fb9809dea845d802897dc5c723910fa3 # v7.1.3
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
       - name: Check format
         run: |
           uv run --frozen ruff format --diff
@@ -27,7 +27,7 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-      - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
         with:
           submodules: 'true'
 
@@ -35,7 +35,7 @@ jobs:
       ###### Required to build OIDC App ######
 
       - name: Set up php 8.4
-        uses: shivammathur/setup-php@bf6b4fbd49ca58e4608c9c89fba0b8d90bd2a39f # v2
+        uses: shivammathur/setup-php@44454db4f0199b8b9685a5d763dc37cbf79108e1 # v2
         with:
           php-version: 8.4
           coverage: none
@@ -49,14 +49,14 @@ jobs:
 
 
       - name: Run docker compose
-        uses: hoverkraft-tech/compose-action@3846bcd61da338e9eaaf83e7ed0234a12b099b72 # v2.4.1
+        uses: hoverkraft-tech/compose-action@248470ecc5ed40d8ed3d4480d8260d77179ef579 # v2.4.2
         with:
           compose-file: "./docker-compose.yml"
           #compose-flags: "--profile qdrant"
           up-flags: "--build"
 
       - name: Install the latest version of uv
-        uses: astral-sh/setup-uv@5a7eac68fb9809dea845d802897dc5c723910fa3 # v7.1.3
+        uses: astral-sh/setup-uv@1e862dfacbd1d6d858c55d9b792c756523627244 # v7.1.4
 
       - name: Install Playwright dependencies
         run: |
@@ -85,4 +85,4 @@ jobs:
           NEXTCLOUD_USERNAME: "admin"
           NEXTCLOUD_PASSWORD: "admin"
         run: |
-          uv run pytest -v --log-cli-level=WARN -m smoke
+          uv run pytest -v --log-cli-level=WARN -m unit -m smoke

.gitmodules

@@ -1,6 +1,6 @@
-[submodule "oidc"]
-	path = third_party/oidc
-	url = https://github.com/cbcoutinho/oidc
 [submodule "third_party/oidc"]
 	path = third_party/oidc
 	url = https://github.com/cbcoutinho/oidc
+[submodule "third_party/notes"]
+	path = third_party/notes
+	url = https://github.com/cbcoutinho/notes

CHANGELOG.md

@@ -1,3 +1,257 @@
+## v0.49.0 (2025-12-08)
+
+### Feat
+
+- **news**: add Nextcloud News app integration
+
+### Fix
+
+- resolve all type checking errors (8 errors fixed)
+
+### Refactor
+
+- **news**: simplify vector sync to fetch all items
+
+### Perf
+
+- **news**: use direct API endpoint for get_item()
+
+## v0.48.6 (2025-12-03)
+
+### Fix
+
+- **deps**: update dependency mcp to >=1.23,<1.24
+
+## v0.48.5 (2025-11-28)
+
+### Fix
+
+- **deps**: update dependency pillow to v12
+
+## v0.48.4 (2025-11-23)
+
+### Fix
+
+- Add rate limit retry logic to OpenAI provider
+
+## v0.48.3 (2025-11-23)
+
+### Fix
+
+- Increase MCP sampling timeout to 5 minutes for slower LLMs
+
+## v0.48.2 (2025-11-23)
+
+### Fix
+
+- Share vector sync state with FastMCP session lifespan via module singleton
+- Share vector sync state with FastMCP session lifespan via module singleton
+
+## v0.48.1 (2025-11-23)
+
+### Fix
+
+- Use WebDAV for tag creation and add LLM-as-a-judge for RAG tests
+
+### Refactor
+
+- Move background tasks to server lifespan and deprecate SSE transport
+
+## v0.48.0 (2025-11-23)
+
+### Feat
+
+- Add tag management methods to WebDAV client
+
+## v0.47.0 (2025-11-23)
+
+### Feat
+
+- Add OpenAI provider support for embeddings and generation
+
+## v0.46.2 (2025-11-22)
+
+### Fix
+
+- **smithery**: Enable JSON response format for scanner compatibility
+
+## v0.46.1 (2025-11-22)
+
+### Perf
+
+- Optimize vector viz search performance
+
+## v0.46.0 (2025-11-22)
+
+### Feat
+
+- Add Smithery CLI deployment support
+- Implement ADR-016 Smithery stateless deployment mode
+
+### Fix
+
+- **smithery**: Add JSON Schema metadata to mcp-config endpoint
+- **smithery**: Use container runtime pattern for config discovery
+- Add Smithery lifespan and auth mode detection
+
+## v0.45.0 (2025-11-22)
+
+### Feat
+
+- Add context expansion to semantic search with chunk overlap removal
+- Use Ollama native batch API in embed_batch()
+- Implement Qdrant placeholder state management
+- Switch files to use numeric IDs with file_path resolution
+- Implement per-chunk vector visualization with context expansion
+
+### Fix
+
+- Use alpha_composite for proper RGBA highlight blending
+- Remove pymupdf.layout.activate() to fix page_chunks behavior
+- Centralize PDF processing and generate separate images per chunk
+- Set is_placeholder=False in processor to fix search filtering
+- Increase placeholder staleness threshold to 5x scan interval
+- Add placeholder staleness check to prevent duplicate processing
+- Use empty SparseVector instead of None for placeholders
+- Return empty array instead of null for query_coords when no results
+- Align PDF text extraction between indexing and context expansion
+- Update models and viz to use int-only doc_id
+- Reconstruct full content for notes to match indexed offsets
+- Add async/await, PDF metadata, and type safety fixes
+
+### Refactor
+
+- Simplify PDF text extraction with single to_markdown call
+
+### Perf
+
+- Optimize PDF processing with parallel extraction and single-render highlights
+
+## v0.44.1 (2025-11-21)
+
+### Fix
+
+- **deps**: update dependency mcp to >=1.22,<1.23
+
+## v0.44.0 (2025-11-19)
+
+### Feat
+
+- Improve vector visualization with static assets and fixes
+- Redesign UI to match Nextcloud ecosystem aesthetic
+
+### Fix
+
+- Improve 3D plot rendering with explicit dimensions and window resize support
+- Preserve 3D plot camera and improve documentation
+- Preserve 3D plot camera position and fix CSS loading
+
+## v0.43.0 (2025-11-18)
+
+### Feat
+
+- Replace custom document chunker with LangChain MarkdownTextSplitter
+
+## v0.42.0 (2025-11-17)
+
+### Feat
+
+- **viz**: Add dual-score display and improve UI controls
+
+## v0.41.0 (2025-11-17)
+
+### Feat
+
+- add configurable fusion algorithms for BM25 hybrid search
+- add chunk position tracking to vector indexing and search
+- add vector viz template and chunk context endpoint
+
+### Fix
+
+- prevent infinite loop in DocumentChunker with position tracking
+- Relax SearchResult validation to support DBSF fusion scores > 1.0
+
+## v0.40.0 (2025-11-16)
+
+### Feat
+
+- add unified provider architecture with Amazon Bedrock support
+
+### Fix
+
+- suppress Starlette middleware type warnings in ty checker
+
+## v0.39.0 (2025-11-16)
+
+### Feat
+
+- Implement BM25 hybrid search with native Qdrant RRF fusion
+
+### Fix
+
+- Handle named vectors in visualization and semantic search
+- Update vizApp to use bm25_hybrid algorithm and remove deprecated weights
+- Update viz routes to use BM25 hybrid search after refactor
+
+## v0.38.0 (2025-11-16)
+
+### Feat
+
+- add concurrent uploads and --force flag to upload command
+- implement RAG evaluation framework with CLI tooling
+
+### Fix
+
+- download qrels from BEIR ZIP instead of HuggingFace
+
+### Refactor
+
+- migrate asyncio to anyio for consistent structured concurrency
+- replace httpx client with NextcloudClient in upload command
+
+### Perf
+
+- Eliminate double-fetching in semantic search sampling
+- fix vector viz search performance and visual encoding
+- make note deletion concurrent in upload --force
+
+## v0.37.0 (2025-11-16)
+
+### Feat
+
+- Add OpenTelemetry tracing to @instrument_tool decorator
+
+## v0.36.0 (2025-11-15)
+
+### BREAKING CHANGE
+
+- Search algorithms now require Qdrant to be populated.
+Vector sync must be enabled and documents indexed for search to work.
+
+### Feat
+
+- Normalize hybrid search RRF scores to 0-1 range
+- Enhance vector visualization UI and parallelize search verification
+- Add Vector Viz tab to app home page
+- Add vector visualization pane with multi-select document types
+- Implement custom PCA to remove sklearn dependency
+- Add multi-document Protocol with cross-app search support
+- Update nc_semantic_search tool with algorithm selection
+- Implement unified search algorithm module
+
+### Fix
+
+- Reorder tabs and fix viz pane session access
+
+### Refactor
+
+- Optimize Nextcloud access verification with centralized filtering
+- Make all search algorithms query Qdrant payload, not Nextcloud
+
+### Perf
+
+- Exclude vector-sync status polling from distributed tracing
+
 ## v0.35.0 (2025-11-15)
 
 ### Feat

CLAUDE.md

@@ -17,13 +17,17 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 - **Use Python 3.10+ union syntax**: `str | None` instead of `Optional[str]`
 - **Use lowercase generics**: `dict[str, Any]` instead of `Dict[str, Any]`
 - **Type all function signatures** - Parameters and return types
-- **No explicit type checker configured** - Ruff handles linting only
+- **Type checker**: `ty` is configured for static type checking
+  ```bash
+  uv run ty check -- nextcloud_mcp_server
+  ```
 
 ### Code Quality
-- **Run ruff before committing**:
+- **Run ruff and ty before committing**:
   ```bash
   uv run ruff check
   uv run ruff format
+  uv run ty check -- nextcloud_mcp_server
   ```
 - **Ruff configuration** in pyproject.toml (extends select: ["I"] for import sorting)
 
@@ -57,8 +61,60 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 - `nextcloud_mcp_server/server/` - MCP tool/resource definitions
 - `nextcloud_mcp_server/auth/` - OAuth/OIDC authentication
 - `nextcloud_mcp_server/models/` - Pydantic response models
+- `nextcloud_mcp_server/providers/` - Unified LLM provider infrastructure (embeddings + generation)
 - `tests/` - Layered test suite (unit, smoke, integration, load)
 
+### Provider Architecture (ADR-015)
+
+**Unified Provider System** for embeddings and text generation:
+
+**Location:** `nextcloud_mcp_server/providers/`
+- `base.py` - `Provider` ABC with optional capabilities
+- `registry.py` - Auto-detection and factory pattern
+- `ollama.py` - Ollama provider (embeddings + generation)
+- `anthropic.py` - Anthropic provider (generation only)
+- `bedrock.py` - Amazon Bedrock provider (embeddings + generation)
+- `simple.py` - Simple in-memory provider (embeddings only, fallback)
+
+**Usage:**
+```python
+from nextcloud_mcp_server.providers import get_provider
+
+provider = get_provider()  # Auto-detects from environment
+
+# Check capabilities
+if provider.supports_embeddings:
+    embeddings = await provider.embed_batch(texts)
+
+if provider.supports_generation:
+    text = await provider.generate("prompt", max_tokens=500)
+```
+
+**Environment Variables:**
+
+Bedrock:
+- `AWS_REGION` - AWS region (e.g., "us-east-1")
+- `BEDROCK_EMBEDDING_MODEL` - Embedding model ID (e.g., "amazon.titan-embed-text-v2:0")
+- `BEDROCK_GENERATION_MODEL` - Generation model ID (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY` - Optional, uses AWS credential chain
+
+Ollama:
+- `OLLAMA_BASE_URL` - API URL (e.g., "http://localhost:11434")
+- `OLLAMA_EMBEDDING_MODEL` - Embedding model (default: "nomic-embed-text")
+- `OLLAMA_GENERATION_MODEL` - Generation model (e.g., "llama3.2:1b")
+- `OLLAMA_VERIFY_SSL` - SSL verification (default: "true")
+
+Simple (fallback, no config needed):
+- `SIMPLE_EMBEDDING_DIMENSION` - Dimension (default: 384)
+
+**Auto-Detection Priority:** Bedrock → Ollama → Simple
+
+**Backward Compatibility:**
+- Old code using `nextcloud_mcp_server.embedding.get_embedding_service()` still works
+- `EmbeddingService` now wraps `get_provider()` internally
+
+**For Details:** See `docs/ADR-015-unified-provider-architecture.md`
+
 ## Development Commands (Quick Reference)
 
 ### Testing

Dockerfile

@@ -1,17 +1,24 @@
-FROM ghcr.io/astral-sh/uv:0.9.9-python3.11-alpine@sha256:0faa7934fac1db7f5056f159c1224d144bab864fd2677a4066d25a686ae32edd
+FROM docker.io/library/python:3.12-slim-trixie@sha256:b43ff04d5df04ad5cabb80890b7ef74e8410e3395b19af970dcd52d7a4bff921
+
+COPY --from=ghcr.io/astral-sh/uv:0.9.16@sha256:ae9ff79d095a61faf534a882ad6378e8159d2ce322691153d68d2afac7422840 /uv /uvx /bin/
 
 # Install dependencies
 # 1. git (required for caldav dependency from git)
 # 2. sqlite for development with token db
-RUN apk add --no-cache git sqlite
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y \
+    git \
+    tesseract-ocr \
+    sqlite3 && apt clean
 
 WORKDIR /app
 
 COPY . .
 
-RUN uv sync --locked --no-dev --no-editable
+RUN uv sync --locked --no-dev --no-editable --no-cache
 
 ENV PYTHONUNBUFFERED=1
 ENV VIRTUAL_ENV=/app/.venv
+ENV PATH=/app/.vnev/bin:$PATH
+ENV TESSDATA_PREFIX=/usr/share/tesseract-ocr/5/tessdata
 
 ENTRYPOINT ["/app/.venv/bin/nextcloud-mcp-server", "--host", "0.0.0.0"]

Dockerfile.smithery

@@ -0,0 +1,44 @@
+# Dockerfile for Smithery stateless deployment
+# ADR-016: Stateless mode for multi-user public Nextcloud instances
+#
+# This image excludes:
+# - Vector database dependencies (qdrant-client)
+# - Background sync workers
+# - Admin UI routes (/app)
+# - Semantic search tools
+#
+# Features included:
+# - Core Nextcloud tools (notes, calendar, contacts, files, deck, tables, cookbook)
+# - Per-session app password authentication
+# - Multi-user support via Smithery session config
+
+FROM docker.io/library/python:3.12-slim-trixie@sha256:b43ff04d5df04ad5cabb80890b7ef74e8410e3395b19af970dcd52d7a4bff921
+
+WORKDIR /app
+
+# Install uv for fast dependency management
+COPY --from=ghcr.io/astral-sh/uv:0.9.16@sha256:ae9ff79d095a61faf534a882ad6378e8159d2ce322691153d68d2afac7422840 /uv /uvx /bin/
+
+# Install dependencies
+# 1. git (required for caldav dependency from git)
+# 2. sqlite for development with token db
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y \
+    git
+
+# Copy project files
+COPY . .
+
+RUN uv sync --locked --no-dev --no-editable --no-cache
+
+# Set Smithery mode environment variables
+ENV SMITHERY_DEPLOYMENT=true
+ENV VECTOR_SYNC_ENABLED=false
+
+# Smithery sets PORT=8081 by default
+EXPOSE 8081
+
+# Health check endpoint
+HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
+    CMD uv run python -c "import httpx; httpx.get('http://localhost:${PORT:-8081}/health/live').raise_for_status()"
+
+CMD ["/app/.venv/bin/smithery-main"]

README.md

@@ -1,6 +1,11 @@
+<p align="center">
+  <img src="astrolabe.svg" alt="Nextcloud MCP Server" width="128" height="128">
+</p>
+
 # Nextcloud MCP Server
 
 [![Docker Image](https://img.shields.io/badge/docker-ghcr.io/cbcoutinho/nextcloud--mcp--server-blue)](https://github.com/cbcoutinho/nextcloud-mcp-server/pkgs/container/nextcloud-mcp-server)
+[![smithery badge](https://smithery.ai/badge/@cbcoutinho/nextcloud-mcp-server)](https://smithery.ai/server/@cbcoutinho/nextcloud-mcp-server)
 
 **A production-ready MCP server that connects AI assistants to your Nextcloud instance.**
 
@@ -13,7 +18,20 @@ This is a **dedicated standalone MCP server** designed for external MCP clients
 
 ## Quick Start
 
-Get up and running in 60 seconds using Docker:
+The fastest way to get started is via [Smithery](https://smithery.ai/server/@cbcoutinho/nextcloud-mcp-server) - no Docker or self-hosting required:
+
+1. Visit the [Smithery marketplace page](https://smithery.ai/server/@cbcoutinho/nextcloud-mcp-server)
+2. Click "Deploy" and configure:
+   - **Nextcloud URL**: Your Nextcloud instance (e.g., `https://cloud.example.com`)
+   - **Username**: Your Nextcloud username
+   - **App Password**: Generate one in Nextcloud → Settings → Security → Devices & sessions
+
+> [!NOTE]
+> Smithery runs in stateless mode without semantic search. For full features, use [Docker](#docker-self-hosted) or see [ADR-016](docs/ADR-016-smithery-stateless-deployment.md).
+
+## Docker (Self-Hosted)
+
+For full features including semantic search, run with Docker:
 
 ```bash
 # 1. Create a minimal configuration
@@ -29,10 +47,15 @@ docker run -p 127.0.0.1:8000:8000 --env-file .env --rm \
 
 # 3. Test the connection
 curl http://127.0.0.1:8000/health/ready
+
+# 4. Connect to the endpoint
+http://127.0.0.1:8000/sse
+
+# Or with --transport streamable-http
+http://127.0.0.1:8000/mcp
 ```
 
 **Next Steps:**
-- Create an app password in Nextcloud: Settings → Security → Devices & sessions
 - Connect your MCP client (Claude Desktop, IDEs, `mcp dev`, etc.)
 - See [docs/installation.md](docs/installation.md) for other deployment options (local, Kubernetes)
 
@@ -123,6 +146,7 @@ This enables natural language queries and helps discover related content across
 - **[App Documentation](docs/)** - Notes, Calendar, Contacts, WebDAV, Deck, Cookbook, Tables
 - **[Document Processing](docs/configuration.md#document-processing)** - OCR and text extraction setup
 - **[Semantic Search Architecture](docs/semantic-search-architecture.md)** - Experimental vector search (Notes only, opt-in)
+- **[Vector Sync UI Guide](docs/user-guide/vector-sync-ui.md)** - Browser interface for semantic search visualization and testing
 
 ### Advanced Topics
 - **[OAuth Architecture](docs/oauth-architecture.md)** - How OAuth works (experimental)

app-hooks/post-installation/10-install-news-app.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+set -euox pipefail
+
+php /var/www/html/occ app:enable news

app-hooks/post-installation/10-install-notes-app.sh

@@ -2,4 +2,30 @@
 
 set -euox pipefail
 
-php /var/www/html/occ app:enable notes
+echo "Installing and configuring notes app for testing..."
+
+# Check if development notes app is mounted at /opt/apps/notes
+if [ -d /opt/apps/notes ]; then
+    echo "Development notes app found at /opt/apps/notes"
+
+    # Remove any existing notes app in apps (from app store or old symlink)
+    if [ -e /var/www/html/custom_apps/notes ]; then
+        echo "Removing existing notes in apps..."
+        rm -rf /var/www/html/custom_apps/notes
+    fi
+
+    # Create symlink from apps to the mounted development version
+    # Per Nextcloud docs: apps outside server root need symlinks in server root
+    echo "Creating symlink: custom_apps/notes -> /opt/apps/notes"
+    ln -sf /opt/apps/notes /var/www/html/custom_apps/notes
+
+    echo "Enabling notes app from /opt/apps (development mode via symlink)"
+    php /var/www/html/occ app:enable notes
+elif [ -d /var/www/html/custom_apps/notes ]; then
+    echo "notes app directory found in apps (already installed)"
+    php /var/www/html/occ app:enable notes
+else
+    echo "notes app not found, installing from app store..."
+    php /var/www/html/occ app:install notes
+    php /var/www/html/occ app:enable notes
+fi

astrolabe.svg

@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="512" height="512" viewBox="0 0 512 512">
+  <rect width="512" height="512" rx="80" ry="80" fill="#0082C9"/>
+  <path d="M255.9 21.04c-11.8 0-22.2 4.08-28.6 10.01-5.6 4.98-8.6 11.41-8.6 18.11 0 5.55 2.2 11.01 5.9 15.48-16.4 4.97-30.1 13.64-39 24.53 22.1-7.67 45.7-11.86 70.3-11.86 24.6 0 48.3 4.19 70.3 11.86-8.9-10.89-22.6-19.56-39-24.53 3.9-4.47 5.9-9.93 5.9-15.48 0-6.7-3-13.13-8.5-18.11-6.4-5.93-16.9-10.01-28.7-10.01zm0 20.34c5.3 0 10.1 1.27 13.6 3.52 1.7 1.16 3.4 2.43 3.4 4.27 0 1.76-1.7 3.03-3.4 4.19-3.5 2.33-8.3 3.61-13.6 3.61-5.3 0-10.1-1.28-13.6-3.61-1.6-1.16-3.3-2.43-3.3-4.19 0-1.84 1.7-3.11 3.3-4.27 3.5-2.25 8.3-3.52 13.6-3.52zm.1 48.1c-110.8 0-200.72 90.02-200.72 200.82S145.2 491 256 491s200.7-89.9 200.7-200.7c0-110.8-89.9-200.82-200.7-200.82zm0 32.62c92.9 0 168.2 75.3 168.2 168.2 0 92.8-75.3 168.2-168.2 168.2-92.9 0-168.26-75.4-168.26-168.2 0-92.9 75.36-168.2 168.26-168.2zm-8.2 6.3c-9.6.5-19 1.9-28.3 4.1l2.3 7.8c8.4-2 17.1-3.3 26-3.8v-8.1zm16.2 0v8.1c9 .5 17.7 1.8 26 3.8l2.2-7.8c-9.1-2.2-18.6-3.6-28.2-4.1zm-60 8.5c-9 3.2-17.6 7-25.8 11.6l4.1 7.1c7.7-4.3 15.6-7.9 23.9-10.8l-2.2-7.9zm103.7 0-2 7.9c8.4 2.9 16.2 6.5 23.8 10.8l4.2-7.1c-8.2-4.6-16.9-8.4-26-11.6zm-143.3 20.3c-7.5 5.4-14.6 11.4-21.1 17.9l5.8 5.8c5.9-6.1 12.5-11.7 19.5-16.6l-4.2-7.1zm182.9 0-4 7.1c6.9 4.9 13.5 10.5 19.5 16.6l5.7-5.8c-6.5-6.5-13.7-12.5-21.2-17.9zm-91.4 11.5c-37 0-67.4 28.6-70.3 64.9l15.9 4.7c.7-29.6 24.7-53.4 54.4-53.4 30.1 0 54.4 24.4 54.4 54.3 0 15-6.2 28.7-16 38.5l.1.1c1.7 2.7 3 5.6 4.1 8.6.9 3 1.7 5.7 2.3 8.6v.4c33.8-16.7 57.2-51.5 57.2-91.7 0-3.8-.2-7.3-.6-10.9-3.2-3.3-6.3-6.4-9.8-9.5 1.5 6.5 2.3 13.4 2.3 20.4 0 28.7-13 54.7-33.5 71.8 6.3-10.6 10.1-23 10.1-36.3 0-38.9-31.7-70.5-70.6-70.5zm-91.8 14.6c-3.3 3.1-6.5 6.2-9.7 9.5-.3 3.6-.5 7.1-.5 10.9 0 7.3.7 14.2 2.1 20.9l9.1 2.7c-2.1-7.5-3.1-15.4-3.1-23.6 0-7 .7-13.9 2.1-20.4zm-31.6 4c-5.8 7.1-10.9 14.6-15.4 22.6l7.1 4c4.1-7.4 8.8-14.3 14-20.8l-5.7-5.8zm246.8 0-5.7 5.8c5.3 6.5 10 13.4 13.9 20.8l7.1-4c-4.4-8-9.5-15.5-15.3-22.6zm-269.2 37.1c-2.5 5.7-4.6 11.4-6.4 17.6l.1-.3c3.4-5 7.9-9.3 12.9-12.5l.3-.6-6.9-4.2zm291.8 0-7.2 4.2c3.2 7.3 5.7 15.1 7.6 23.1l7.9-2.1c-2.1-8.8-4.9-17.3-8.3-25.2zm-261.2 11.5c-13.4.1-25.7 9-29.7 22.5l114.8 34.2c-4.9 16.7 4.6 34.2 21.2 39.2L361.7 366c16.6 5 34.1-4.4 39.1-21l-114.6-34.4c4.9-16.5-4.7-34.1-21.3-39.1 0 0-72.4-21.5-114.8-34.3-3.1-.9-6.3-1.4-9.4-1.3zm-42.09 29.7c-.9 6.9-1.4 14-1.4 21.3 0 1.3.1 2.9.1 4.2h8.09v-4.2c0-6.5.4-12.9 1.2-19.2l-7.99-2.1zm314.59 0-7.9 2.1c.7 6.3 1.3 12.7 1.3 19.2 0 1.3 0 2.9-.2 4.2h8.2v-4.2c0-7.3-.5-14.4-1.4-21.3zm-157.3 24.7c6.3 0 11.5 5 11.5 11.3 0 6.4-5.2 11.6-11.5 11.6s-11.5-5.2-11.5-11.6c0-6.3 5.2-11.3 11.5-11.3zM98.51 307.4c1 8.2 2.89 16.4 5.09 24.3l7.9-2.1c-2.1-7.2-3.8-14.6-4.8-22.2h-8.19zm306.69 0c-1.1 7.6-2.7 15-4.8 22.2l7.8 2.1c2.2-7.9 4.1-16.1 5.2-24.3h-8.2zm-191.3 10.9c-19 13.3-31.4 35.3-31.4 60.1 0 10.4 2.3 20.4 6.2 29.7 8.8 4.9 17.9 8.8 27.6 11.7-10.8-10.7-17.5-25.2-17.5-41.4 0-19 9.3-36 23.7-46.3-3.8-4.1-6.7-8.7-8.6-13.8zM116.8 345l-7.9 2c3.1 7.6 6.8 14.7 11 21.6l6.9-4.2c-3.8-6.2-7-12.8-10-19.4zm194.8 20.5c.9 4.1 1.4 8.5 1.4 12.9 0 16.2-6.7 30.7-17.4 41.4 9.6-2.9 18.8-6.8 27.5-11.7 4-9.3 6.2-19.3 6.2-29.7 0-2.7-.2-5.2-.4-7.7l-17.3-5.2zM136 377.9l-7.1 4.1c4.7 6.2 9.7 12.1 15.3 17.3l5.7-5.5c-5.1-5-9.7-10.3-13.9-15.9zm243.9 2.3-.2.1c-2.1.3-4 .6-6.2.7h-.1c-3.6 4.5-7.3 8.8-11.5 12.8l5.8 5.5c5.5-5.2 10.5-11.1 15.2-17.3l-3-1.8zm-217.8 24-5.9 5.9c6 4.8 12.2 9.7 18.8 13.6l3.8-7.8c-5.7-2.9-11.4-6.8-16.7-11.7zm187.7 0c-5.4 4.9-11.1 8.8-16.8 11.7l3.9 7.8c6.5-3.9 12.8-8.8 18.7-13.6l-5.8-5.9zm-156.4 19.5-4.1 6.8c6.6 4 13.7 5.8 20.7 8.8l2.2-7.9c-6.5-1.9-12.7-4.8-18.8-7.7zm125.2 0c-6.2 2.9-12.5 5.8-19.1 7.7l2.3 7.9c7.2-3 14-4.8 20.7-8.8l-3.9-6.8zm-90.7 11.7-2 7.8c7.1 1 14.5 1.9 21.9 1.9v-7.7c-6.8 0-13.5-1.1-19.9-2zm55.9 0c-6.3.9-13 2-19.8 2v7.7c7.5 0 14.8-.9 22.1-1.9l-2.3-7.8z" fill="#fff"/>
+</svg>

charts/nextcloud-mcp-server/Chart.lock

@@ -1,9 +1,9 @@
 dependencies:
 - name: qdrant
   repository: https://qdrant.github.io/qdrant-helm
-  version: 1.15.5
+  version: 1.16.2
 - name: ollama
   repository: https://otwld.github.io/ollama-helm
-  version: 1.34.0
-digest: sha256:d51c97d05be2614b751c0dd7267ef7dc959eff5ebef859c5f895c5c554b7a874
-generated: "2025-11-09T17:08:02.86648061Z"
+  version: 1.35.0
+digest: sha256:bcb0779739e4710b90bb65f6a7baeaa295bd0ba9776f8a1cf8d9b69d233c8ec0
+generated: "2025-12-05T11:11:27.999374001Z"

charts/nextcloud-mcp-server/Chart.yaml

@@ -2,8 +2,8 @@ apiVersion: v2
 name: nextcloud-mcp-server
 description: A Helm chart for Nextcloud MCP Server - enables AI assistants to interact with Nextcloud
 type: application
-version: 0.35.0
-appVersion: "0.35.0"
+version: 0.49.0
+appVersion: "0.49.0"
 keywords:
   - nextcloud
   - mcp
@@ -27,10 +27,10 @@ annotations:
   grafana_dashboard_folder: "Nextcloud MCP"
 dependencies:
   - name: qdrant
-    version: "1.15.5"
+    version: "1.16.2"
     repository: https://qdrant.github.io/qdrant-helm
     condition: qdrant.networkMode.deploySubchart
   - name: ollama
-    version: "1.34.0"
+    version: "1.35.0"
     repository: https://otwld.github.io/ollama-helm
     condition: ollama.enabled

docker-compose.ci.yml

@@ -0,0 +1,25 @@
+# CI-specific overrides for RAG evaluation pipeline
+# This file is used by the rag-evaluation.yml workflow to configure the MCP
+# container with OpenAI/GitHub Models API for vector embeddings.
+#
+# Usage:
+#   docker compose -f docker-compose.yml -f docker-compose.ci.yml up
+#
+# Environment variables (set in CI workflow):
+#   OPENAI_API_KEY - API key for embeddings (GitHub Models uses GITHUB_TOKEN)
+#   OPENAI_BASE_URL - API endpoint (e.g., https://models.github.ai/inference)
+#   OPENAI_EMBEDDING_MODEL - Model name (e.g., openai/text-embedding-3-small)
+#   OPENAI_GENERATION_MODEL - Model name for generation (e.g., openai/gpt-4o-mini)
+
+services:
+  mcp:
+    environment:
+      # OpenAI provider configuration (required for CI vector sync)
+      - OPENAI_API_KEY=${OPENAI_API_KEY}
+      - OPENAI_BASE_URL=${OPENAI_BASE_URL:-https://models.github.ai/inference}
+      - OPENAI_EMBEDDING_MODEL=${OPENAI_EMBEDDING_MODEL:-openai/text-embedding-3-small}
+      - OPENAI_GENERATION_MODEL=${OPENAI_GENERATION_MODEL:-openai/gpt-4o-mini}
+      # Faster sync for CI
+      - VECTOR_SYNC_SCAN_INTERVAL=${VECTOR_SYNC_SCAN_INTERVAL:-5}
+      # Enable document processing for PDF parsing
+      - ENABLE_DOCUMENT_PROCESSING=true

docker-compose.yml

@@ -3,7 +3,7 @@ services:
   # https://hub.docker.com/_/mariadb
   db:
     # Note: Check the recommend version here: https://docs.nextcloud.com/server/latest/admin_manual/installation/system_requirements.html#server
-    image: docker.io/library/mariadb:lts@sha256:6b848cb24fbbd87429917f6c4422ac53c343e85692eb0fef86553e99e4f422f3
+    image: docker.io/library/mariadb:lts@sha256:1cac8492bd78b1ec693238dc600be173397efd7b55eabc725abc281dc855b482
     restart: always
     command: --transaction-isolation=READ-COMMITTED
     volumes:
@@ -17,11 +17,11 @@ services:
   # Note: Redis is an external service. You can find more information about the configuration here:
   # https://hub.docker.com/_/redis
   redis:
-    image: docker.io/library/redis:alpine@sha256:28c9c4d7596949a24b183eaaab6455f8e5d55ecbf72d02ff5e2c17fe72671d31
+    image: docker.io/library/redis:alpine@sha256:6cbef353e480a8a6e7f10ec545f13d7d3fa85a212cdcc5ffaf5a1c818b9d3798
     restart: always
 
   app:
-    image: docker.io/library/nextcloud:32.0.1@sha256:5b043f7ea2f609d5ff5635f475c30d303bec17775a5c3f7fa435e3818e669120
+    image: docker.io/library/nextcloud:32.0.2@sha256:8cb1dc8c26944115469dd22f4965d2ed35bab9cf8c48d2bb052c8e9f83821ded
     restart: always
     ports:
       - 0.0.0.0:8080:80
@@ -158,7 +158,7 @@ services:
       - oauth-tokens:/app/data
 
   keycloak:
-    image: quay.io/keycloak/keycloak:26.4.5@sha256:653852bfdea2be6e958b9e90a976eff1c6de34edd55f2f679bdc48ef16bc528e
+    image: quay.io/keycloak/keycloak:26.4.7@sha256:9409c59bdfb65dbffa20b11e6f18b8abb9281d480c7ca402f51ed3d5977e6007
     command:
       - "start-dev"
       - "--import-realm"
@@ -224,8 +224,28 @@ services:
       - keycloak-tokens:/app/data
       - keycloak-oauth-storage:/app/.oauth
 
+  # Smithery stateless deployment mode (ADR-016)
+  # Test with: docker compose --profile smithery up smithery
+  # Then: curl http://localhost:8081/.well-known/mcp-config
+  smithery:
+    build:
+      context: .
+      dockerfile: Dockerfile.smithery
+    restart: always
+    depends_on:
+      app:
+        condition: service_healthy
+    ports:
+      - 127.0.0.1:8081:8081
+    environment:
+      - SMITHERY_DEPLOYMENT=true
+      - VECTOR_SYNC_ENABLED=false
+      - PORT=8081
+    profiles:
+      - smithery
+
   qdrant:
-    image: qdrant/qdrant:v1.15.5@sha256:0fb8897412abc81d1c0430a899b9a81eb8328aa634e7242d1bc804c1fe8fe863
+    image: qdrant/qdrant:v1.16.2@sha256:dab6de32f7b2cc599985a7c764db3e8b062f70508fb85ca074aa856f829bf335
     restart: always
     ports:
       - 127.0.0.1:6333:6333  # REST API

docs/ADR-011-improving-semantic-search-quality.md

@@ -1,7 +1,8 @@
 # ADR-011: Improving Semantic Search Quality Through Better Chunking and Embeddings
 
-**Status**: Proposed
+**Status**: Partially Implemented (Chunking Complete, Embeddings Pending)
 **Date**: 2025-11-12
+**Implementation Date**: 2025-11-18 (Chunking)
 **Authors**: Development Team
 **Related**: ADR-003 (Vector Database Architecture), ADR-008 (MCP Sampling for RAG)
 
@@ -893,3 +894,50 @@ This ADR addresses the root causes of poor semantic search recall:
 - No new infrastructure or ongoing costs
 
 **Next Steps**: Approve ADR → Implement changes → Reindex → Validate → Production rollout
+
+## Implementation Status
+
+### Completed (2025-11-18)
+
+**✅ Semantic Markdown-Aware Chunking (Option C1 + C3 Hybrid)**
+
+Implementation details:
+- Replaced custom word-based chunking with `MarkdownTextSplitter` from LangChain
+- Optimized for Nextcloud Notes markdown content with special handling for:
+  - Headers (`#`, `##`, `###`, etc.)
+  - Code blocks (` ``` `)
+  - Lists (`-`, `*`, `1.`)
+  - Horizontal rules (`---`)
+  - Paragraphs and sentences
+- Maintained `ChunkWithPosition` interface for backward compatibility
+- Updated configuration defaults:
+  - `DOCUMENT_CHUNK_SIZE`: 512 words → 2048 characters
+  - `DOCUMENT_CHUNK_OVERLAP`: 50 words → 200 characters
+- Updated unit tests to verify position tracking and boundary preservation
+- All tests passing with markdown-aware character-based chunking
+
+**Files Modified**:
+- `nextcloud_mcp_server/vector/document_chunker.py` - LangChain integration
+- `nextcloud_mcp_server/config.py` - Character-based defaults
+- `tests/unit/test_document_chunker.py` - Updated test suite
+
+**Dependencies Added**:
+- `langchain-text-splitters>=1.0.0` (already present in `pyproject.toml`)
+
+**Migration Required**:
+- ⚠️ Full reindex required to apply new chunking strategy
+- Existing documents in vector database use old word-based chunks
+- See "Migration Strategy" section above for reindexing process
+
+### Pending
+
+**⏳ Embedding Model Upgrade (Option E1)**
+
+Still to be implemented:
+- Switch from `nomic-embed-text` (768-dim) to `mxbai-embed-large-v1` (1024-dim)
+- Implement dynamic dimension detection in `ollama_provider.py`
+- Create migration script for collection reindexing
+- Run benchmarking to validate improvement
+- Deploy to production with atomic collection swap
+
+**Estimated Timeline**: 1-2 weeks for implementation and validation

docs/ADR-014-bm25-search.md

@@ -0,0 +1,241 @@
+# ADR-014: Replace Custom Keyword Search with BM25 Hybrid Search via Qdrant
+
+**Date:** 2025-11-16
+
+**Status:** Implemented
+
+---
+
+### 1. Context
+
+Our RAG application currently employs two separate retrieval mechanisms:
+1.  **Dense (Semantic) Search:** Using vector embeddings stored in our Qdrant database to find semantically similar context.
+2.  **Keyword Search:** A custom-built fuzzy/character-based search to match-specific keywords, acronyms, and product codes that semantic search often misses.
+
+This dual-system approach has several drawbacks:
+* **Poor Relevance:** Our current keyword search is basic (e.g., `LIKE` queries or simple fuzzy matching). It is not as effective as modern full-text search algorithms like BM25.
+* **Clunky Fusion:** We lack a robust, principled method to combine the results from the two systems. This leads to disjointed logic in the application layer and suboptimal context being passed to the LLM.
+* **Architectural Complexity:** We must maintain two separate search pathways (one to Qdrant, one to the keyword search mechanism), increasing code complexity and maintenance overhead.
+
+Our vector database, **Qdrant**, natively supports **hybrid search** by combining dense vectors with BM25-based **sparse vectors** in a single collection.
+
+### 2. Decision
+
+We will **deprecate and remove** the existing custom keyword/fuzzy search functionality.
+
+We will **replace it by implementing native hybrid search within Qdrant**. This involves:
+1.  **Modifying the Qdrant Collection:** Updating our collection to support a named sparse vector index configured for BM25.
+2.  **Updating the Ingestion Pipeline:** For every document chunk, we will generate and upsert *both*:
+    * Its **dense vector** (from our existing embedding model).
+    * Its **sparse vector** (generated using a BM25-compatible model, e.g., `Qdrant/bm25` from `fastembed`).
+3.  **Refactoring Retrieval Logic:** All retrieval calls will be consolidated into a single Qdrant query using the `query_points` endpoint. This query will use the `prefetch` parameter to execute both dense and sparse searches, and Qdrant's built-in **Reciprocal Rank Fusion (RRF)** to automatically merge the results into a single, relevance-ranked list.
+4.  **Backfilling:** A one-time migration script will be created to generate and add sparse vectors for all existing documents in the Qdrant collection.
+
+---
+
+### 3. Considered Options
+
+#### Option 1: Native Qdrant Hybrid Search (Chosen)
+* Use Qdrant's built-in sparse vector and RRF capabilities.
+* **Pros:**
+    * **Consolidated Architecture:** Manages both dense and sparse indexes in one database.
+    * **No Data Sync Issues:** Updates are atomic. A single `upsert` updates both representations.
+    * **Built-in Fusion:** RRF is handled natively and efficiently by the database.
+    * **Superior Relevance:** Replaces our brittle custom search with the industry-standard BM25.
+* **Cons:**
+    * Requires a one-time data backfill which may be time-consuming.
+    * Adds a new step (sparse vector generation) to the ingestion pipeline.
+
+#### Option 2: External Full-Text Search (e.g., Elasticsearch)
+* Keep Qdrant for dense search and add a separate Elasticsearch/OpenSearch cluster for BM25.
+* **Pros:**
+    * Provides a very powerful, dedicated full-text search engine.
+* **Cons:**
+    * **High Complexity:** Introduces a new, stateful service to deploy, manage, and scale.
+    * **Data Sync Nightmare:** We would be responsible for ensuring that the document IDs and content in Qdrant and Elasticsearch are always perfectly synchronized. This is a major source of bugs.
+    * **Manual Fusion:** The application would have to query both systems and perform RRF manually.
+
+#### Option 3: Keep Current System
+* Make no changes.
+* **Pros:**
+    * No engineering effort required.
+* **Cons:**
+    * Fails to address the known relevance and architectural problems.
+    * Our RAG application's performance will remain suboptimal, especially for keyword-sensitive queries.
+
+---
+
+### 4. Rationale
+
+**Option 1 is the clear winner.** It directly solves our primary problem (poor keyword matching) by adopting the industry-standard BM25.
+
+Critically, it achieves this while **simplifying** our overall architecture, not complicating it. By leveraging features already present in our existing database (Qdrant), we avoid the massive operational and synchronization overhead of adding a second search system (Option 2).
+
+This decision consolidates our retrieval logic, eliminates the data consistency problem, and moves the complex fusion logic (RRF) from the application layer into the database, where it can be performed more efficiently.
+
+### 5. Consequences
+
+**New Work:**
+* **Ingestion:** The data ingestion pipeline must be updated to add the `fastembed` library (or similar), generate sparse vectors, and upsert them to the new named vector field in Qdrant.
+* **Retrieval:** The application's retrieval service must be refactored to use the `query_points` endpoint with `prefetch` and `fusion=models.Fusion.RRF`.
+* **Migration:** A one-time backfill script must be written and executed to add sparse vectors for all existing documents.
+* **Infrastructure:** The Qdrant collection schema must be updated (or re-created) to add the `sparse_vectors_config`.
+
+**Positive:**
+* **Improved Accuracy:** Retrieval will be significantly more accurate, handling both semantic and keyword queries robustly.
+* **Simplified Code:** The application's retrieval logic will be cleaner and simpler, with one endpoint instead of two.
+* **Reduced Maintenance:** We will remove the custom fuzzy-search code, which is brittle and difficult to maintain.
+
+**Negative:**
+* The data backfill process will require careful management to avoid downtime.
+* Ingestion time will slightly increase due to the extra step of sparse vector generation. This is considered a negligible trade-off for the gains in relevance.
+
+---
+
+### 6. Implementation Notes
+
+**Implementation completed on 2025-11-16**
+
+**Key Changes:**
+
+1. **Dependencies** (pyproject.toml:25):
+   - Added `fastembed>=0.4.2` for BM25 sparse vector embeddings
+   - Adjusted `pillow` version constraint to be compatible with fastembed
+
+2. **Qdrant Collection Schema** (nextcloud_mcp_server/vector/qdrant_client.py:113-128):
+   - Updated to named vectors: `{"dense": VectorParams(...), "sparse": SparseVectorParams(...)}`
+   - Added sparse vector configuration with BM25 index
+   - Maintains backward compatibility with existing collections (detects legacy schema)
+
+3. **BM25 Embedding Provider** (nextcloud_mcp_server/embedding/bm25_provider.py):
+   - Created `BM25SparseEmbeddingProvider` using FastEmbed's `Qdrant/bm25` model
+   - Implements `encode()` and `encode_batch()` methods
+   - Returns sparse vectors as `{indices: list[int], values: list[float]}` format
+
+4. **Document Indexing Pipeline** (nextcloud_mcp_server/vector/processor.py:229-255):
+   - Generates both dense (semantic) and sparse (BM25) embeddings for each document chunk
+   - Updates `PointStruct` to use named vectors: `vector={"dense": ..., "sparse": ...}`
+   - Maintains same chunking strategy (512 words, 50-word overlap)
+
+5. **BM25 Hybrid Search Algorithm** (nextcloud_mcp_server/search/bm25_hybrid.py):
+   - Implements `BM25HybridSearchAlgorithm` using Qdrant's native RRF fusion
+   - Uses `prefetch` parameter for parallel dense + sparse search
+   - Applies `fusion=models.Fusion.RRF` for automatic result merging
+   - Maintains same deduplication and filtering logic as semantic search
+
+6. **MCP Tool Updates** (nextcloud_mcp_server/server/semantic.py:39-68):
+   - Simplified `nc_semantic_search()` to use BM25 hybrid only
+   - Removed `algorithm`, `semantic_weight`, `keyword_weight`, `fuzzy_weight` parameters
+   - Updated default `score_threshold=0.0` for RRF scoring
+   - Returns `search_method="bm25_hybrid"` in responses
+
+7. **Legacy Algorithm Removal**:
+   - Deleted `nextcloud_mcp_server/search/keyword.py` (278 lines)
+   - Deleted `nextcloud_mcp_server/search/fuzzy.py` (220 lines)
+   - Deleted `nextcloud_mcp_server/search/hybrid.py` (238 lines - custom RRF)
+   - Updated `nextcloud_mcp_server/search/__init__.py` to export only BM25 hybrid
+
+**Migration Strategy:**
+- No migration required (vector sync feature is experimental)
+- New documents automatically indexed with both dense + sparse vectors
+- Collection re-creation on first startup with updated schema
+
+**Test Results:**
+- All unit tests passing (118 passed)
+- All integration tests passing (7 semantic search tests)
+- Code formatting verified with ruff
+
+**Benefits Realized:**
+- ✅ Consolidated architecture (single Qdrant database for both dense + sparse)
+- ✅ Native fusion algorithms (database-level, more efficient)
+- ✅ Industry-standard BM25 (replaces custom keyword search)
+- ✅ Simplified codebase (removed 736 lines of legacy code)
+- ✅ Better relevance (handles both semantic and keyword queries)
+- ✅ Configurable fusion methods (RRF and DBSF)
+
+---
+
+### 7. Fusion Algorithm Options
+
+**Update: 2025-11-16**
+
+The BM25 hybrid search now supports two fusion algorithms for combining dense (semantic) and sparse (BM25) search results:
+
+#### Reciprocal Rank Fusion (RRF)
+
+**Default fusion method.** RRF is a widely-used, well-established algorithm that combines rankings from multiple retrieval systems using the reciprocal rank formula:
+
+```
+RRF(doc) = Σ 1/(k + rank_i(doc))
+```
+
+where `k` is a constant (typically 60) and `rank_i(doc)` is the rank of the document in retrieval system `i`.
+
+**Characteristics:**
+- ✅ **General-purpose**: Works well across diverse query types and document collections
+- ✅ **Rank-based**: Focuses on relative rankings rather than absolute scores
+- ✅ **Established**: Well-tested, documented, and understood in IR literature
+- ✅ **Robust**: Less sensitive to score distribution differences between systems
+
+**When to use RRF:**
+- Default choice for most use cases
+- When you have mixed query types (semantic + keyword)
+- When retrieval systems have very different score ranges
+- When you want predictable, well-understood behavior
+
+#### Distribution-Based Score Fusion (DBSF)
+
+**Alternative fusion method.** DBSF normalizes scores from each retrieval system using distribution statistics before combining them:
+
+1. **Normalization**: For each query, calculates mean (μ) and standard deviation (σ) of scores
+2. **Outlier handling**: Uses μ ± 3σ as normalization bounds
+3. **Fusion**: Sums normalized scores across systems
+
+**Characteristics:**
+- ✅ **Score-aware**: Uses actual relevance scores, not just rankings
+- ✅ **Statistical**: Normalizes based on score distribution properties
+- ⚠️ **Experimental**: Newer algorithm, less battle-tested than RRF
+- ⚠️ **Sensitive**: May behave differently depending on score distributions
+
+**When to use DBSF:**
+- When retrieval systems have vastly different score ranges that RRF doesn't balance well
+- When you want to experiment with score-based (vs rank-based) fusion
+- When statistical normalization better matches your use case
+- For A/B testing against RRF to measure retrieval quality improvements
+
+#### Configuration
+
+Both fusion algorithms are exposed via the `fusion` parameter in MCP tools:
+
+```python
+# Use RRF (default)
+response = await nc_semantic_search(
+    query="async programming",
+    fusion="rrf"  # Can be omitted, RRF is default
+)
+
+# Use DBSF
+response = await nc_semantic_search(
+    query="async programming",
+    fusion="dbsf"
+)
+```
+
+The `nc_semantic_search_answer` tool also supports the `fusion` parameter and passes it through to the underlying search.
+
+#### Future: Configurable Weights
+
+**Current limitation**: Neither RRF nor DBSF currently support per-system weights (e.g., 0.8 for semantic, 0.2 for BM25). This is a Qdrant platform limitation tracked in [qdrant/qdrant#6067](https://github.com/qdrant/qdrant/issues/6067).
+
+When Qdrant adds weight support, the `fusion` parameter can be extended to accept weight configurations:
+
+```python
+# Hypothetical future API
+response = await nc_semantic_search(
+    query="async programming",
+    fusion="rrf",
+    fusion_weights={"dense": 0.7, "sparse": 0.3}  # Not yet implemented
+)
+```
+
+**Recommendation**: Start with RRF (default). If you encounter cases where keyword matches are under- or over-weighted, experiment with DBSF. Monitor [qdrant/qdrant#6067](https://github.com/qdrant/qdrant/issues/6067) for configurable weight support.

docs/ADR-015-unified-provider-architecture.md

@@ -0,0 +1,380 @@
+# ADR-015: Unified Provider Architecture for Embeddings and Text Generation
+
+**Status:** Accepted
+**Date:** 2025-01-16
+**Deciders:** Development Team
+**Related:** ADR-003 (Vector Database), ADR-008 (MCP Sampling), ADR-013 (RAG Evaluation)
+
+## Context
+
+Prior to this refactoring, the codebase had two separate provider systems:
+
+1. **Embedding Providers** (`nextcloud_mcp_server/embedding/`)
+   - Used `EmbeddingProvider` ABC with methods: `embed()`, `embed_batch()`, `get_dimension()`
+   - Had auto-detection via `EmbeddingService._detect_provider()`
+   - Used for semantic search and vector indexing (production)
+
+2. **LLM Providers** (`tests/rag_evaluation/llm_providers.py`)
+   - Used `LLMProvider` Protocol with method: `generate()`
+   - Had separate factory function `create_llm_provider()`
+   - Used only for RAG evaluation tests (not production)
+
+This fragmentation created several problems:
+
+### Problems with Dual Provider Systems
+
+1. **Code Duplication**
+   - Ollama configuration appeared in both `embedding/service.py` and `tests/rag_evaluation/llm_providers.py`
+   - Similar provider detection logic in multiple places
+   - Separate singleton patterns for each system
+
+2. **Limited Extensibility**
+   - Hard-coded provider detection in `EmbeddingService._detect_provider()`
+   - No support for providers that offer both capabilities (like Bedrock)
+   - Adding new providers required modifying multiple files
+
+3. **Inconsistent Patterns**
+   - BM25 provider didn't follow `EmbeddingProvider` ABC
+   - Different method names across providers (`embed` vs `encode`)
+   - ABC vs Protocol for type checking
+
+4. **Difficult Scaling**
+   - Adding Amazon Bedrock (our third provider) would exacerbate all issues
+   - No clear path for future providers (OpenAI, Cohere, etc.)
+
+### Amazon Bedrock Requirements
+
+Bedrock naturally supports **both** embeddings and text generation:
+- **Embeddings**: `amazon.titan-embed-text-v1/v2`, `cohere.embed-*`
+- **Text Generation**: `anthropic.claude-*`, `meta.llama3-*`, `amazon.titan-text-*`
+- **Unified API**: Single `invoke_model()` method via bedrock-runtime
+
+This made it the perfect opportunity to establish a unified provider architecture.
+
+## Decision
+
+We refactored the provider infrastructure to use a **unified Provider ABC** with optional capabilities:
+
+### 1. Unified Provider Interface
+
+**New Structure:**
+```
+nextcloud_mcp_server/providers/
+├── __init__.py
+├── base.py              # Provider ABC with optional capabilities
+├── registry.py          # Auto-detection and factory
+├── ollama.py            # Supports both embedding + generation
+├── anthropic.py         # Generation only
+├── bedrock.py           # Supports both embedding + generation
+└── simple.py            # Embedding only (testing fallback)
+```
+
+**Base Class (`providers/base.py`):**
+```python
+class Provider(ABC):
+    @property
+    @abstractmethod
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        pass
+
+    @property
+    @abstractmethod
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        pass
+
+    @abstractmethod
+    async def embed(self, text: str) -> list[float]:
+        """Generate embedding (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """Generate batch embeddings (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    def get_dimension(self) -> int:
+        """Get embedding dimension (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """Generate text (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def close(self) -> None:
+        """Close provider and release resources."""
+        pass
+```
+
+### 2. Provider Registry
+
+**Auto-Detection Priority** (`providers/registry.py`):
+```python
+class ProviderRegistry:
+    @staticmethod
+    def create_provider() -> Provider:
+        # 1. Bedrock (AWS_REGION or BEDROCK_*_MODEL)
+        # 2. Ollama (OLLAMA_BASE_URL)
+        # 3. Simple (fallback)
+```
+
+**Environment Variables:**
+
+**Bedrock:**
+- `AWS_REGION`: AWS region (e.g., "us-east-1")
+- `AWS_ACCESS_KEY_ID`: AWS access key (optional, uses credential chain)
+- `AWS_SECRET_ACCESS_KEY`: AWS secret key (optional)
+- `BEDROCK_EMBEDDING_MODEL`: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0")
+- `BEDROCK_GENERATION_MODEL`: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+
+**Ollama:**
+- `OLLAMA_BASE_URL`: Ollama API base URL (e.g., "http://localhost:11434")
+- `OLLAMA_EMBEDDING_MODEL`: Model for embeddings (default: "nomic-embed-text")
+- `OLLAMA_GENERATION_MODEL`: Model for text generation (e.g., "llama3.2:1b")
+- `OLLAMA_VERIFY_SSL`: Verify SSL certificates (default: "true")
+
+**Simple (no configuration, fallback):**
+- `SIMPLE_EMBEDDING_DIMENSION`: Embedding dimension (default: 384)
+
+### 3. Backward Compatibility
+
+**Old Code Continues to Work:**
+```python
+# Old way (still works)
+from nextcloud_mcp_server.embedding import get_embedding_service
+
+service = get_embedding_service()  # Returns singleton Provider
+embeddings = await service.embed_batch(texts)
+```
+
+**New Way (recommended):**
+```python
+# New way (cleaner)
+from nextcloud_mcp_server.providers import get_provider
+
+provider = get_provider()  # Returns singleton Provider
+embeddings = await provider.embed_batch(texts)
+
+# Can also use generation if provider supports it
+if provider.supports_generation:
+    text = await provider.generate("prompt")
+```
+
+**Migration Path:**
+- `embedding/service.py` now wraps `providers.get_provider()` for compatibility
+- `tests/rag_evaluation/llm_providers.py` now uses unified providers
+- Old imports still work, marked as deprecated in docstrings
+
+### 4. Amazon Bedrock Implementation
+
+**Features:**
+- Supports both embeddings and text generation
+- Model-specific request/response handling for:
+  - Titan Embed (amazon.titan-embed-text-*)
+  - Cohere Embed (cohere.embed-*)
+  - Claude (anthropic.claude-*)
+  - Llama (meta.llama3-*)
+  - Titan Text (amazon.titan-text-*)
+  - Mistral (mistral.*)
+- Uses boto3 bedrock-runtime client
+- Graceful degradation if boto3 not installed
+- Async implementation matching existing patterns
+
+**Model-Specific Handling:**
+```python
+# Bedrock embedding request (Titan)
+{"inputText": text}
+
+# Bedrock generation request (Claude)
+{
+    "anthropic_version": "bedrock-2023-05-31",
+    "max_tokens": max_tokens,
+    "temperature": 0.7,
+    "messages": [{"role": "user", "content": prompt}]
+}
+```
+
+## Consequences
+
+### Positive
+
+1. **Sustainable Provider Additions**
+   - New providers only need to implement `Provider` ABC
+   - Auto-detection via environment variables
+   - No modifications to existing code required
+
+2. **Code Consolidation**
+   - Single provider interface instead of two
+   - Unified configuration pattern
+   - Eliminated duplication
+
+3. **Better Extensibility**
+   - Providers can support one or both capabilities
+   - Clear capability detection via properties
+   - Registry pattern simplifies auto-detection
+
+4. **Improved Testing**
+   - RAG evaluation can use any provider (Ollama, Anthropic, Bedrock)
+   - Comprehensive unit tests for all providers
+   - Mocked boto3 tests for Bedrock
+
+5. **Production-Ready Bedrock Support**
+   - Full embedding and generation support
+   - Multiple model families supported
+   - AWS credential chain integration
+
+### Neutral
+
+1. **Optional Boto3 Dependency**
+   - boto3 is dev dependency only (not required for core functionality)
+   - Bedrock provider gracefully fails if boto3 not installed
+   - Users who want Bedrock must `pip install boto3`
+
+2. **Capability Properties**
+   - All providers must implement capability properties
+   - Methods raise `NotImplementedError` if capability not supported
+   - Clear error messages guide users to alternatives
+
+### Negative
+
+1. **Migration Effort**
+   - Existing code must be migrated to new imports (optional, backward compatible)
+   - Documentation needs updating
+   - Users must learn new environment variables
+
+2. **Increased Complexity**
+   - Provider base class has more methods (embedding + generation)
+   - More environment variables to configure
+   - Capability detection adds runtime checks
+
+## Implementation
+
+### Files Created
+
+**New Provider Infrastructure:**
+- `nextcloud_mcp_server/providers/__init__.py`
+- `nextcloud_mcp_server/providers/base.py`
+- `nextcloud_mcp_server/providers/registry.py`
+- `nextcloud_mcp_server/providers/ollama.py`
+- `nextcloud_mcp_server/providers/anthropic.py`
+- `nextcloud_mcp_server/providers/bedrock.py`
+- `nextcloud_mcp_server/providers/simple.py`
+
+**Tests:**
+- `tests/unit/providers/__init__.py`
+- `tests/unit/providers/test_bedrock.py` (9 unit tests)
+
+**Documentation:**
+- `docs/ADR-015-unified-provider-architecture.md` (this file)
+
+### Files Modified
+
+**Backward Compatibility:**
+- `nextcloud_mcp_server/embedding/service.py` - Now wraps `get_provider()`
+- `tests/rag_evaluation/llm_providers.py` - Uses unified providers
+
+**Dependencies:**
+- `pyproject.toml` - Added `boto3>=1.35.0` to dev dependencies
+
+### Testing Results
+
+**Unit Tests:** 127 passed (including 9 new Bedrock tests)
+**Type Checking:** All checks passed (ty)
+**Linting:** All checks passed (ruff)
+**Backward Compatibility:** Verified - existing embedding tests work
+
+## Alternatives Considered
+
+### Alternative 1: Keep Separate Provider Systems
+
+**Pros:**
+- No refactoring needed
+- Simpler short-term
+
+**Cons:**
+- Bedrock would need to be implemented twice
+- Continued code duplication
+- No long-term scalability
+
+**Decision:** Rejected - technical debt would continue to grow
+
+### Alternative 2: Separate Embedding and Generation Providers
+
+Use composition instead of unified interface:
+```python
+class CombinedProvider:
+    def __init__(self, embedding: EmbeddingProvider, generation: LLMProvider):
+        self.embedding = embedding
+        self.generation = generation
+```
+
+**Pros:**
+- Clearer separation of concerns
+- Simpler individual providers
+
+**Cons:**
+- Bedrock and Ollama naturally do both - artificial separation
+- More complex configuration (two providers to configure)
+- More boilerplate code
+
+**Decision:** Rejected - unified interface better matches provider capabilities
+
+### Alternative 3: Plugin System
+
+Dynamic provider registration via entry points:
+```python
+# setup.py
+entry_points={
+    'nextcloud_mcp.providers': [
+        'ollama = nextcloud_mcp_server.providers.ollama:OllamaProvider',
+        'bedrock = nextcloud_mcp_server.providers.bedrock:BedrockProvider',
+    ]
+}
+```
+
+**Pros:**
+- Most extensible
+- Third-party providers possible
+
+**Cons:**
+- Over-engineered for current needs
+- Added complexity
+- No immediate benefit
+
+**Decision:** Deferred - can add later if needed
+
+## Future Work
+
+1. **Additional Providers**
+   - OpenAI (embeddings + generation)
+   - Cohere (embeddings + generation)
+   - Google Vertex AI
+   - Azure OpenAI
+
+2. **Provider Features**
+   - Streaming generation support
+   - Batch API optimization (when available)
+   - Model-specific optimizations
+   - Cost tracking and metrics
+
+3. **Configuration Improvements**
+   - Provider profiles (development, production)
+   - Model aliasing (e.g., "small", "large")
+   - Fallback provider chains
+
+4. **Testing**
+   - Integration tests with real Bedrock endpoints
+   - Performance benchmarking across providers
+   - Cost comparison analysis
+
+## References
+
+- [boto3 Bedrock Runtime Documentation](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime.html)
+- [Amazon Bedrock User Guide](https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html)
+- ADR-003: Vector Database and Semantic Search
+- ADR-008: MCP Sampling for Semantic Search
+- ADR-013: RAG Evaluation Framework

docs/ADR-016-smithery-stateless-deployment.md

@@ -0,0 +1,492 @@
+# ADR-016: Smithery Stateless Deployment for Multi-User Public Nextcloud Instances
+
+**Status:** Proposed
+**Date:** 2025-01-22
+**Deciders:** Development Team
+**Related:** ADR-004 (OAuth), ADR-007 (Background Vector Sync), ADR-015 (Unified Provider)
+
+## Context
+
+[Smithery](https://smithery.ai) is a hosting platform and marketplace for MCP servers that provides:
+
+- **Discovery**: Marketplace listing for MCP servers
+- **Hosting**: Containerized deployment with auto-scaling
+- **Authentication UI**: OAuth flow presentation for users
+- **Session Configuration**: Per-user settings passed via URL parameters
+- **Observability**: Usage logs and monitoring
+
+### Current Architecture Limitations
+
+The current nextcloud-mcp-server architecture assumes a **self-hosted deployment** with:
+
+1. **Persistent Infrastructure**
+   - Qdrant vector database for semantic search
+   - Background sync worker for content indexing
+   - Refresh token storage for offline access
+
+2. **Single-Tenant Configuration**
+   - Environment variables configure one Nextcloud instance
+   - `NEXTCLOUD_HOST`, `NEXTCLOUD_USERNAME`, `NEXTCLOUD_PASSWORD`
+   - Or OAuth with a single IdP
+
+3. **Stateful Operations**
+   - Vector sync maintains index state across requests
+   - Token storage persists between sessions
+
+### Smithery Hosting Constraints
+
+Smithery-hosted containers are **stateless by design**:
+
+- No persistent storage between requests
+- No background workers or cron jobs
+- No databases (Qdrant, Redis, etc.)
+- Containers may be recycled at any time
+- Configuration passed per-session via URL parameters
+
+### Opportunity
+
+Many users have **publicly accessible Nextcloud instances** and want to:
+
+1. Try the MCP server without self-hosting infrastructure
+2. Connect multiple users to different Nextcloud instances
+3. Use basic Nextcloud tools without semantic search
+4. Benefit from Smithery's discovery and OAuth UI
+
+## Decision
+
+Implement a **stateless deployment mode** for Smithery that:
+
+1. **Disables stateful features** (vector sync, semantic search)
+2. **Creates clients per-session** from Smithery configuration
+3. **Supports multiple Nextcloud instances** via session config
+4. **Provides a useful subset of tools** that work without infrastructure
+
+### Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────────────┐
+│                    Smithery-Hosted Stateless Mode                        │
+├─────────────────────────────────────────────────────────────────────────┤
+│                                                                          │
+│  MCP Client                    Smithery                                  │
+│  (Cursor, Claude)              Infrastructure                            │
+│        │                            │                                    │
+│        │ 1. Connect                 │                                    │
+│        ├───────────────────────────►│                                    │
+│        │                            │                                    │
+│        │ 2. Config UI               │                                    │
+│        │◄───────────────────────────┤  User enters:                      │
+│        │    (Smithery presents)     │  - nextcloud_url                   │
+│        │                            │  - auth_mode (basic/oauth)         │
+│        │                            │  - credentials                     │
+│        │ 3. Tool call               │                                    │
+│        ├───────────────────────────►│                                    │
+│        │    + session config        │                                    │
+│        │                            │                                    │
+│        │                    ┌───────┴───────┐                            │
+│        │                    │  MCP Server   │                            │
+│        │                    │  Container    │                            │
+│        │                    │               │                            │
+│        │                    │ 4. Create     │                            │
+│        │                    │    client     │                            │
+│        │                    │    from       │                            │
+│        │                    │    config     │                            │
+│        │                    │      │        │                            │
+│        │                    │      ▼        │                            │
+│        │                    │ 5. Call       │                            │
+│        │                    │    Nextcloud  │───────► User's Nextcloud   │
+│        │                    │    API        │         Instance           │
+│        │                    │      │        │                            │
+│        │                    │      ▼        │                            │
+│        │ 6. Response        │ Return result │                            │
+│        │◄───────────────────┤               │                            │
+│        │                    └───────────────┘                            │
+│                                                                          │
+└─────────────────────────────────────────────────────────────────────────┘
+```
+
+### Session Configuration Schema
+
+```python
+from pydantic import BaseModel, Field
+
+class SmitheryConfigSchema(BaseModel):
+    """Configuration schema for Smithery session."""
+
+    # Required: Nextcloud instance
+    nextcloud_url: str = Field(
+        ...,
+        description="Your Nextcloud instance URL (e.g., https://cloud.example.com)"
+    )
+
+    # Authentication mode
+    auth_mode: str = Field(
+        "app_password",
+        description="Authentication method: 'app_password' or 'oauth'"
+    )
+
+    # App Password authentication (recommended for Smithery)
+    username: str | None = Field(
+        None,
+        description="Nextcloud username (required for app_password auth)"
+    )
+    app_password: str | None = Field(
+        None,
+        description="Nextcloud app password (Settings → Security → App passwords)"
+    )
+
+    # OAuth authentication (advanced)
+    # When auth_mode='oauth', Smithery handles the OAuth flow
+    # and passes the access token automatically
+```
+
+### Feature Matrix
+
+| Feature | Self-Hosted | Smithery Stateless |
+|---------|-------------|-------------------|
+| **Notes** | | |
+| List/Search notes | ✓ | ✓ |
+| Get/Create/Update notes | ✓ | ✓ |
+| Semantic search | ✓ | ✗ |
+| **Calendar** | | |
+| List calendars | ✓ | ✓ |
+| Get/Create events | ✓ | ✓ |
+| **Contacts** | | |
+| List address books | ✓ | ✓ |
+| Search/Get contacts | ✓ | ✓ |
+| **Files (WebDAV)** | | |
+| List/Download files | ✓ | ✓ |
+| Upload files | ✓ | ✓ |
+| Search files | ✓ | ✓ (keyword only) |
+| **Deck** | | |
+| List boards/cards | ✓ | ✓ |
+| Create/Update cards | ✓ | ✓ |
+| **Tables** | | |
+| List/Query tables | ✓ | ✓ |
+| Create/Update rows | ✓ | ✓ |
+| **Cookbook** | | |
+| List/Get recipes | ✓ | ✓ |
+| **Semantic Search** | | |
+| Vector search | ✓ | ✗ |
+| RAG answers | ✓ | ✗ |
+| **Background Sync** | | |
+| Auto-indexing | ✓ | ✗ |
+| Webhook sync | ✓ | ✗ |
+| **Admin UI (`/app`)** | | |
+| Vector sync status | ✓ | ✗ |
+| Vector visualization | ✓ | ✗ |
+| Webhook management | ✓ | ✗ |
+| Session management | ✓ | ✗ |
+
+### Implementation
+
+#### 1. Deployment Mode Detection
+
+```python
+# nextcloud_mcp_server/config.py
+
+class DeploymentMode(Enum):
+    SELF_HOSTED = "self_hosted"      # Full features, env-based config
+    SMITHERY_STATELESS = "smithery"  # Stateless, session-based config
+
+def get_deployment_mode() -> DeploymentMode:
+    """Detect deployment mode from environment."""
+    if os.getenv("SMITHERY_DEPLOYMENT") == "true":
+        return DeploymentMode.SMITHERY_STATELESS
+    return DeploymentMode.SELF_HOSTED
+```
+
+#### 2. Session-Based Client Factory
+
+```python
+# nextcloud_mcp_server/context.py
+
+async def get_client(ctx: Context) -> NextcloudClient:
+    """Get NextcloudClient - from session config or environment."""
+
+    mode = get_deployment_mode()
+
+    if mode == DeploymentMode.SMITHERY_STATELESS:
+        # Create client from Smithery session config
+        config = ctx.session_config
+        if not config:
+            raise McpError("Session configuration required")
+
+        return NextcloudClient(
+            base_url=config.nextcloud_url,
+            username=config.username,
+            password=config.app_password,
+        )
+    else:
+        # Existing behavior: from environment or OAuth context
+        return await _get_client_from_context(ctx)
+```
+
+#### 3. Conditional Tool Registration
+
+```python
+# nextcloud_mcp_server/app.py
+
+def create_mcp_server(mode: DeploymentMode) -> FastMCP:
+    """Create MCP server with mode-appropriate tools."""
+
+    mcp = FastMCP("Nextcloud MCP")
+
+    # Always register core tools
+    configure_notes_tools(mcp)
+    configure_calendar_tools(mcp)
+    configure_contacts_tools(mcp)
+    configure_webdav_tools(mcp)
+    configure_deck_tools(mcp)
+    configure_tables_tools(mcp)
+    configure_cookbook_tools(mcp)
+
+    # Only register stateful tools in self-hosted mode
+    if mode == DeploymentMode.SELF_HOSTED:
+        configure_semantic_tools(mcp)  # Requires Qdrant
+        register_oauth_tools(mcp)       # Requires token storage
+
+    return mcp
+```
+
+#### 4. Exclude Admin UI Routes
+
+The `/app` admin UI should **not be installed** in Smithery mode because:
+
+- **Vector sync status** - No vector sync in stateless mode
+- **Vector visualization** - No Qdrant to visualize
+- **Webhook management** - No webhook sync without background workers
+- **Session management** - No persistent sessions to manage
+
+```python
+# nextcloud_mcp_server/app.py
+
+def create_app(mode: DeploymentMode) -> Starlette:
+    """Create Starlette app with mode-appropriate routes."""
+
+    routes = [
+        Route("/health/live", health_live, methods=["GET"]),
+        Route("/health/ready", health_ready, methods=["GET"]),
+    ]
+
+    # Only mount admin UI in self-hosted mode
+    if mode == DeploymentMode.SELF_HOSTED:
+        browser_app = create_browser_app()
+        routes.append(
+            Route("/app", lambda r: RedirectResponse("/app/", status_code=307))
+        )
+        routes.append(Mount("/app", app=browser_app))
+        logger.info("Admin UI mounted at /app")
+    else:
+        logger.info("Admin UI disabled in Smithery stateless mode")
+
+    # Mount FastMCP at root
+    mcp_app = create_mcp_server(mode).streamable_http_app()
+    routes.append(Mount("/", app=mcp_app))
+
+    return Starlette(routes=routes, lifespan=starlette_lifespan)
+```
+
+**Endpoints by Mode:**
+
+| Endpoint | Self-Hosted | Smithery |
+|----------|-------------|----------|
+| `/mcp` | ✓ | ✓ |
+| `/health/live` | ✓ | ✓ |
+| `/health/ready` | ✓ | ✓ |
+| `/.well-known/mcp-config` | ✓ | ✓ |
+| `/app` | ✓ | ✗ |
+| `/app/vector-sync/status` | ✓ | ✗ |
+| `/app/vector-viz` | ✓ | ✗ |
+| `/app/webhooks` | ✓ | ✗ |
+
+#### 5. Smithery Integration Files
+
+**smithery.yaml:**
+```yaml
+runtime: "container"
+build:
+  dockerfile: "Dockerfile.smithery"
+  dockerBuildPath: "."
+startCommand:
+  type: "http"
+  configSchema:
+    type: "object"
+    required: ["nextcloud_url", "username", "app_password"]
+    properties:
+      nextcloud_url:
+        type: "string"
+        title: "Nextcloud URL"
+        description: "Your Nextcloud instance URL (e.g., https://cloud.example.com)"
+      username:
+        type: "string"
+        title: "Username"
+        description: "Your Nextcloud username"
+      app_password:
+        type: "string"
+        title: "App Password"
+        description: "Generate at Settings → Security → App passwords"
+  exampleConfig:
+    nextcloud_url: "https://cloud.example.com"
+    username: "alice"
+    app_password: "xxxxx-xxxxx-xxxxx-xxxxx-xxxxx"
+```
+
+**Dockerfile.smithery:**
+```dockerfile
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install uv
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /bin/uv
+
+# Copy project files
+COPY pyproject.toml uv.lock ./
+COPY nextcloud_mcp_server ./nextcloud_mcp_server
+
+# Install dependencies (without vector/semantic extras)
+RUN uv sync --frozen --no-dev
+
+# Set Smithery mode
+ENV SMITHERY_DEPLOYMENT=true
+ENV VECTOR_SYNC_ENABLED=false
+
+# Smithery sets PORT=8081
+EXPOSE 8081
+
+CMD ["uv", "run", "python", "-m", "nextcloud_mcp_server.smithery_main"]
+```
+
+**nextcloud_mcp_server/smithery_main.py:**
+```python
+"""Smithery-specific entrypoint for stateless deployment."""
+
+import os
+import uvicorn
+from starlette.middleware.cors import CORSMiddleware
+
+from nextcloud_mcp_server.app import create_mcp_server
+from nextcloud_mcp_server.config import DeploymentMode
+
+def main():
+    # Force stateless mode
+    os.environ["SMITHERY_DEPLOYMENT"] = "true"
+    os.environ["VECTOR_SYNC_ENABLED"] = "false"
+
+    mcp = create_mcp_server(DeploymentMode.SMITHERY_STATELESS)
+    app = mcp.streamable_http_app()
+
+    # Add CORS for browser-based clients
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["GET", "POST", "OPTIONS"],
+        allow_headers=["*"],
+        expose_headers=["mcp-session-id", "mcp-protocol-version"],
+    )
+
+    # Smithery sets PORT environment variable
+    port = int(os.environ.get("PORT", 8081))
+    uvicorn.run(app, host="0.0.0.0", port=port)
+
+if __name__ == "__main__":
+    main()
+```
+
+### Security Considerations
+
+1. **App Passwords over User Passwords**
+   - Smithery config encourages app passwords (revocable, scoped)
+   - Documentation guides users to create dedicated app passwords
+   - App passwords can be revoked without changing main password
+
+2. **HTTPS Required**
+   - `nextcloud_url` must be HTTPS for production use
+   - Validation rejects HTTP URLs in Smithery mode
+
+3. **No Credential Storage**
+   - Credentials exist only for request duration
+   - No server-side persistence of user credentials
+   - Smithery handles secure config transmission
+
+4. **Scope Limitation**
+   - Stateless mode cannot access offline_access
+   - No background operations on user's behalf
+   - Clear user expectation: tools work during session only
+
+### Migration Path
+
+Users can start with Smithery stateless mode and migrate to self-hosted:
+
+1. **Try on Smithery** → Basic tools, no setup
+2. **Self-host for semantic search** → Add Qdrant, enable vector sync
+3. **Full deployment** → Background sync, webhooks, multi-user OAuth
+
+## Consequences
+
+### Positive
+
+1. **Lower barrier to entry** - Users can try without infrastructure
+2. **Multi-user support** - Each session connects to different Nextcloud
+3. **Smithery ecosystem** - Discovery, observability, OAuth UI
+4. **Clear feature tiers** - Stateless (simple) vs self-hosted (full)
+
+### Negative
+
+1. **No semantic search** - Key differentiator unavailable on Smithery
+2. **Per-request auth** - Credentials sent with each request
+3. **No offline access** - Cannot perform background operations
+4. **Maintenance burden** - Two deployment modes to support
+
+### Neutral
+
+1. **Feature subset** - May encourage users to self-host for full features
+2. **Documentation needs** - Clear guidance on mode differences required
+
+## Alternatives Considered
+
+### 1. External MCP Only
+
+**Approach:** Only support self-hosted external MCP registration on Smithery.
+
+**Rejected because:**
+- Higher barrier to entry for new users
+- Misses opportunity for Smithery marketplace visibility
+- Users want to try before committing to infrastructure
+
+### 2. Embedded Vector DB (SQLite-vec)
+
+**Approach:** Use SQLite with vector extensions for per-request indexing.
+
+**Rejected because:**
+- No persistence between requests anyway
+- Indexing latency too high for synchronous requests
+- Complexity without benefit in stateless context
+
+### 3. External Vector DB Service
+
+**Approach:** Connect to Pinecone/Weaviate Cloud from Smithery container.
+
+**Rejected because:**
+- Adds external dependency and cost
+- Per-user collections require complex multi-tenancy
+- Sync still impossible without background workers
+
+### 4. Hybrid: Smithery + User's Qdrant
+
+**Approach:** User provides their own Qdrant URL in session config.
+
+**Considered for future:**
+- Could enable semantic search for advanced users
+- Adds complexity to session config
+- Sync still requires external trigger (manual or webhook)
+
+## References
+
+- [Smithery Documentation](https://smithery.ai/docs)
+- [Smithery Session Configuration](https://smithery.ai/docs/build/session-config)
+- [Smithery External MCPs](https://smithery.ai/docs/build/external)
+- [MCP Streamable HTTP Transport](https://modelcontextprotocol.io/docs/concepts/transports)
+- [Nextcloud App Passwords](https://docs.nextcloud.com/server/latest/user_manual/en/session_management.html#app-passwords)

docs/bedrock-setup.md

@@ -0,0 +1,338 @@
+# Amazon Bedrock Setup Guide
+
+This guide covers how to configure the Nextcloud MCP Server to use Amazon Bedrock for embeddings and text generation.
+
+## Prerequisites
+
+1. **AWS Account** with access to Amazon Bedrock
+2. **boto3 library** installed: `pip install boto3` or `uv sync --group dev`
+3. **Model Access** - Request access to models in AWS Bedrock console
+
+## Required AWS Permissions
+
+### IAM Policy for Bedrock Access
+
+The AWS IAM user or role needs the following permissions:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockInvokeModels",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel",
+        "bedrock:InvokeModelWithResponseStream"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:*::foundation-model/*"
+      ]
+    }
+  ]
+}
+```
+
+### Minimal Permissions (Production)
+
+For production deployments, restrict to specific models:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockEmbeddings",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:us-east-1::foundation-model/amazon.titan-embed-text-v2:0"
+      ]
+    },
+    {
+      "Sid": "BedrockGeneration",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-sonnet-20240229-v1:0"
+      ]
+    }
+  ]
+}
+```
+
+### Additional Permissions (Optional)
+
+For advanced use cases:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockListModels",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:ListFoundationModels",
+        "bedrock:GetFoundationModel"
+      ],
+      "Resource": "*"
+    },
+    {
+      "Sid": "BedrockAsyncInvoke",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModelAsync",
+        "bedrock:GetAsyncInvoke",
+        "bedrock:ListAsyncInvokes"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:*::foundation-model/*"
+      ]
+    }
+  ]
+}
+```
+
+## Model Access
+
+Before using Bedrock models, you must request access in the AWS Console:
+
+1. Navigate to **Amazon Bedrock** → **Model access**
+2. Click **Manage model access**
+3. Select models you want to use:
+   - **Embeddings:** Amazon Titan Embed Text, Cohere Embed
+   - **Text Generation:** Anthropic Claude, Meta Llama, Amazon Titan Text
+4. Click **Request model access**
+5. Wait for approval (usually instant for most models)
+
+## Supported Models
+
+### Embedding Models
+
+| Provider | Model ID | Dimensions | Best For |
+|----------|----------|------------|----------|
+| Amazon Titan | `amazon.titan-embed-text-v1` | 1,536 | General purpose |
+| Amazon Titan | `amazon.titan-embed-text-v2:0` | 1,024 | Latest, improved quality |
+| Cohere | `cohere.embed-english-v3` | 1,024 | English text |
+| Cohere | `cohere.embed-multilingual-v3` | 1,024 | Multilingual |
+
+### Text Generation Models
+
+| Provider | Model ID | Context | Best For |
+|----------|----------|---------|----------|
+| Anthropic | `anthropic.claude-3-sonnet-20240229-v1:0` | 200K | Balanced performance |
+| Anthropic | `anthropic.claude-3-haiku-20240307-v1:0` | 200K | Fast, cost-effective |
+| Anthropic | `anthropic.claude-3-opus-20240229-v1:0` | 200K | Highest quality |
+| Meta | `meta.llama3-8b-instruct-v1:0` | 8K | Fast, open-source |
+| Meta | `meta.llama3-70b-instruct-v1:0` | 8K | High quality |
+| Amazon | `amazon.titan-text-express-v1` | 8K | Fast, low cost |
+| Mistral | `mistral.mistral-7b-instruct-v0:2` | 32K | Efficient |
+
+## Configuration
+
+### Environment Variables
+
+**Required:**
+```bash
+AWS_REGION=us-east-1
+```
+
+**Optional (at least one model required):**
+```bash
+# For embeddings
+BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+
+# For text generation (RAG evaluation)
+BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+```
+
+**AWS Credentials (choose one method):**
+
+**Method 1: Environment Variables**
+```bash
+AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
+AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
+```
+
+**Method 2: AWS Credentials File** (`~/.aws/credentials`)
+```ini
+[default]
+aws_access_key_id = AKIAIOSFODNN7EXAMPLE
+aws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
+```
+
+**Method 3: IAM Role** (when running on AWS EC2/ECS/Lambda)
+- No credentials needed, uses instance/task role automatically
+
+### Docker Configuration
+
+Add to your `docker-compose.yml`:
+
+```yaml
+services:
+  mcp:
+    environment:
+      - AWS_REGION=us-east-1
+      - BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+      - BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+```
+
+Or use AWS credentials file volume mount:
+
+```yaml
+services:
+  mcp:
+    volumes:
+      - ~/.aws:/root/.aws:ro
+    environment:
+      - AWS_REGION=us-east-1
+      - BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+```
+
+## Usage Examples
+
+### Embeddings Only
+
+```bash
+export AWS_REGION=us-east-1
+export BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+export AWS_ACCESS_KEY_ID=your-key
+export AWS_SECRET_ACCESS_KEY=your-secret
+
+uv run nextcloud-mcp-server
+```
+
+### Both Embeddings and Generation
+
+```bash
+export AWS_REGION=us-east-1
+export BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+export BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+
+# For RAG evaluation with Bedrock
+export RAG_EVAL_PROVIDER=bedrock
+export RAG_EVAL_BEDROCK_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+
+uv run python -m tests.rag_evaluation.evaluate
+```
+
+### Programmatic Usage
+
+```python
+from nextcloud_mcp_server.providers import BedrockProvider
+
+# Embeddings only
+provider = BedrockProvider(
+    region_name="us-east-1",
+    embedding_model="amazon.titan-embed-text-v2:0",
+)
+
+embeddings = await provider.embed_batch(["text1", "text2"])
+
+# Both capabilities
+provider = BedrockProvider(
+    region_name="us-east-1",
+    embedding_model="amazon.titan-embed-text-v2:0",
+    generation_model="anthropic.claude-3-sonnet-20240229-v1:0",
+)
+
+# Generate embeddings
+embedding = await provider.embed("query text")
+
+# Generate text
+response = await provider.generate("Write a summary", max_tokens=500)
+```
+
+## Cost Considerations
+
+### Embedding Costs (as of Jan 2025)
+
+| Model | Price per 1K tokens |
+|-------|---------------------|
+| Titan Embed Text v2 | $0.0001 |
+| Cohere Embed English v3 | $0.0001 |
+
+### Generation Costs (as of Jan 2025)
+
+| Model | Input (per 1K tokens) | Output (per 1K tokens) |
+|-------|----------------------|------------------------|
+| Claude 3 Haiku | $0.00025 | $0.00125 |
+| Claude 3 Sonnet | $0.003 | $0.015 |
+| Claude 3 Opus | $0.015 | $0.075 |
+| Llama 3 8B | $0.0003 | $0.0006 |
+| Titan Text Express | $0.0002 | $0.0006 |
+
+**Note:** Prices vary by region. Check [AWS Bedrock Pricing](https://aws.amazon.com/bedrock/pricing/) for current rates.
+
+## Troubleshooting
+
+### Error: "Executable doesn't exist" or boto3 not found
+
+**Solution:**
+```bash
+uv sync --group dev  # Installs boto3
+```
+
+### Error: "AccessDeniedException"
+
+**Causes:**
+1. IAM permissions missing
+2. Model access not requested
+3. Wrong AWS region
+
+**Solution:**
+1. Verify IAM policy includes `bedrock:InvokeModel`
+2. Request model access in Bedrock console
+3. Check model is available in your region
+
+### Error: "ResourceNotFoundException"
+
+**Cause:** Invalid model ID or model not available in region
+
+**Solution:**
+- Verify model ID matches exactly (case-sensitive)
+- Check model availability in your AWS region
+- Use `aws bedrock list-foundation-models` to see available models
+
+### Error: "ThrottlingException"
+
+**Cause:** Rate limit exceeded
+
+**Solution:**
+- Reduce request rate
+- Request quota increase via AWS Support
+- Use batch operations where possible
+
+## Security Best Practices
+
+1. **Use IAM Roles** when running on AWS infrastructure
+2. **Rotate Access Keys** regularly if using IAM users
+3. **Restrict Permissions** to only required models
+4. **Enable CloudTrail** for audit logging
+5. **Use AWS Secrets Manager** for credential management
+6. **Monitor Costs** with AWS Cost Explorer and Budgets
+
+## Regional Availability
+
+Amazon Bedrock is available in:
+- **US East (N. Virginia)**: `us-east-1` ✅ Most models
+- **US West (Oregon)**: `us-west-2` ✅ Most models
+- **Asia Pacific (Singapore)**: `ap-southeast-1`
+- **Asia Pacific (Tokyo)**: `ap-northeast-1`
+- **Europe (Frankfurt)**: `eu-central-1`
+
+**Note:** Model availability varies by region. Check the [AWS Bedrock documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html) for current availability.
+
+## References
+
+- [AWS Bedrock Documentation](https://docs.aws.amazon.com/bedrock/)
+- [AWS Bedrock Pricing](https://aws.amazon.com/bedrock/pricing/)
+- [boto3 Bedrock Runtime API](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime.html)
+- [Provider Architecture ADR](./ADR-015-unified-provider-architecture.md)

docs/observability.md

@@ -243,7 +243,7 @@ If you see cardinality warnings:
 The observability stack integrates at multiple layers:
 
 1. **HTTP Layer**: `ObservabilityMiddleware` tracks all HTTP requests
-2. **MCP Layer**: Tools use `@trace_mcp_tool` for span creation
+2. **MCP Layer**: Tools use `@instrument_tool` for automatic metrics and trace span creation
 3. **Client Layer**: `BaseNextcloudClient` tracks all API calls
 4. **OAuth Layer**: Token operations are traced and metered
 5. **Background Tasks**: Vector sync operations emit metrics/traces

docs/user-guide/vector-sync-ui.md

@@ -0,0 +1,93 @@
+# Vector Sync UI Guide
+
+This guide covers the browser-based interface for the Nextcloud MCP Server's semantic search and vector synchronization features.
+
+## Overview
+
+The Vector Sync UI (`/app`) provides an interactive interface to test semantic search queries and visualize results from your Nextcloud documents. It exposes the same retrieval capabilities that LLMs use in Retrieval-Augmented Generation (RAG) workflows, powered by Alpine.js for reactive state, htmx for dynamic updates, and Plotly.js for 3D visualization.
+
+**Supported Apps**: Notes, Files (text/PDF), Calendar (events/tasks), Contacts (CardDAV), and Deck are indexed and searchable.
+
+## Accessing the UI
+
+Navigate to `/app` after authentication:
+- **BasicAuth mode**: `http://localhost:8000/app` (uses credentials from environment)
+- **OAuth mode**: `http://localhost:8000/app` (redirects to login if not authenticated)
+
+## Tabs
+
+### Welcome Page
+
+Landing page that introduces semantic search and RAG workflows. Shows authentication status, explains how vector embeddings work, and provides feature navigation. Adapts content based on whether `VECTOR_SYNC_ENABLED=true`.
+
+### User Info
+
+Displays authentication details and session information:
+- **BasicAuth**: Username, mode badge, Nextcloud host
+- **OAuth**: Username, session ID (truncated), background access status, IdP profile, revocation option
+
+### Vector Sync Status
+
+Real-time monitoring of document indexing:
+- **Indexed Documents**: Total chunks stored in Qdrant vector database (immediately searchable)
+- **Pending Documents**: Queue awaiting embedding processing
+- **Status**: "✓ Idle" (green) when up-to-date, "⟳ Syncing" (orange) during processing
+
+Auto-refreshes every 10 seconds via htmx. Check this tab after adding content to verify indexing completion.
+
+### Vector Visualization
+
+Interactive search interface with 3D PCA plot of semantic space.
+
+**Search Controls**:
+- **Query**: Natural language search (e.g., "health benefits of coffee")
+- **Algorithm**: Semantic (Dense) for pure vector search, or BM25 Hybrid (default) combining vectors + keywords
+- **Fusion** (Hybrid only): RRF (Reciprocal Rank Fusion) or DBSF (Distribution-Based Score Fusion)
+- **Advanced**: Filter by document type, adjust score threshold (0.0-1.0), set result limit (max 100)
+
+**3D Visualization**:
+
+The plot uses Principal Component Analysis (PCA) to reduce 768-dimensional embeddings to 3D. Documents are positioned by semantic similarity with the query point shown in red. Point size and opacity indicate relevance, and the Viridis color scale shows relative scores (yellow = highest match).
+
+**Critical Fix**: Vectors are L2-normalized before PCA to match Qdrant's cosine distance, ensuring query points position accurately near similar documents. Without normalization, magnitude differences cause misleading spatial separation.
+
+**Results List**:
+
+Each result shows document title (clickable link to Nextcloud), excerpt, raw score, relative percentage, and document type. Click "Show Chunk" to view the matched text segment with surrounding context (up to 500 characters before/after).
+
+## Configuration
+
+**Required**:
+```bash
+VECTOR_SYNC_ENABLED=true
+```
+
+**Optional** (for browser-accessible links):
+```bash
+NEXTCLOUD_PUBLIC_ISSUER_URL=https://your-public-nextcloud-url.com
+```
+
+**Admin Access**: Webhooks tab only visible to Nextcloud admins (verified via Provisioning API).
+
+## Use Cases
+
+**Testing Search Queries**: Preview results before they reach LLMs in RAG workflows. Compare semantic vs. hybrid algorithms, verify relevance scores, and validate that correct documents are retrieved. Use chunk context to see exactly which text segments match and why unexpected documents appear.
+
+**Monitoring Indexing**: Track real-time progress after creating or modifying documents. Check if the queue is backing up (high pending count) or confirm the system is idle after bulk imports. Verify documents become searchable immediately after indexing completes.
+
+**Algorithm Comparison**: Pure semantic search excels at conceptual queries and synonyms. BM25 hybrid combines semantic understanding with precise keyword matching for better accuracy on specific terms. Experiment with RRF vs. DBSF fusion for different score distributions.
+
+## Troubleshooting
+
+**Vector Sync Tab Not Visible**: Set `VECTOR_SYNC_ENABLED=true` and restart the server.
+
+**No Search Results**: Check Vector Sync Status to confirm documents are indexed (not just pending). Try broader queries or lower the score threshold in Advanced options. Initial indexing may take time depending on document volume.
+
+**Links to Nextcloud Apps Not Working**: Set `NEXTCLOUD_PUBLIC_ISSUER_URL` to your browser-accessible Nextcloud URL for correct link generation.
+
+## Related Documentation
+
+- [Configuration Guide](../configuration.md) - Environment variables and settings
+- [Authentication Modes](../authentication.md) - BasicAuth vs OAuth setup
+- [Installation Guide](../installation.md) - Getting started
+- [ADR-008: MCP Sampling for RAG](../ADR-008-mcp-sampling-for-rag.md) - Technical details on RAG workflows

nextcloud_mcp_server/app.py

@@ -3,6 +3,7 @@ import os
 import time
 from collections.abc import AsyncIterator
 from contextlib import AsyncExitStack, asynccontextmanager
+from contextvars import ContextVar
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Optional
 
@@ -24,6 +25,9 @@ from starlette.middleware.authentication import AuthenticationMiddleware
 from starlette.middleware.cors import CORSMiddleware
 from starlette.responses import JSONResponse, RedirectResponse
 from starlette.routing import Mount, Route
+from starlette.staticfiles import StaticFiles
+from starlette.types import ASGIApp, Receive, Send
+from starlette.types import Scope as StarletteScope
 
 from nextcloud_mcp_server.auth import (
     InsufficientScopeError,
@@ -35,6 +39,8 @@ from nextcloud_mcp_server.auth import (
 from nextcloud_mcp_server.auth.unified_verifier import UnifiedTokenVerifier
 from nextcloud_mcp_server.client import NextcloudClient
 from nextcloud_mcp_server.config import (
+    DeploymentMode,
+    get_deployment_mode,
     get_document_processor_config,
     get_settings,
 )
@@ -54,6 +60,7 @@ from nextcloud_mcp_server.server import (
     configure_contacts_tools,
     configure_cookbook_tools,
     configure_deck_tools,
+    configure_news_tools,
     configure_notes_tools,
     configure_semantic_tools,
     configure_sharing_tools,
@@ -121,6 +128,26 @@ def initialize_document_processors():
         except Exception as e:
             logger.warning(f"Failed to register Tesseract processor: {e}")
 
+    # Register PyMuPDF processor (high priority, local, no API required)
+    if "pymupdf" in config["processors"]:
+        pymupdf_config = config["processors"]["pymupdf"]
+        try:
+            from nextcloud_mcp_server.document_processors.pymupdf import (
+                PyMuPDFProcessor,
+            )
+
+            processor = PyMuPDFProcessor(
+                extract_images=pymupdf_config.get("extract_images", True),
+                image_dir=pymupdf_config.get("image_dir"),
+            )
+            registry.register(processor, priority=15)  # Higher than unstructured
+            logger.info(
+                f"Registered PyMuPDF processor: extract_images={pymupdf_config.get('extract_images', True)}"
+            )
+            registered_count += 1
+        except Exception as e:
+            logger.warning(f"Failed to register PyMuPDF processor: {e}")
+
     # Register custom processor
     if "custom" in config["processors"]:
         custom_config = config["processors"]["custom"]
@@ -217,6 +244,25 @@ def validate_pkce_support(discovery: dict, discovery_url: str) -> None:
     click.echo(f"✓ PKCE support validated: {code_challenge_methods}")
 
 
+@dataclass
+class VectorSyncState:
+    """
+    Module-level state for vector sync background tasks.
+
+    This singleton bridges the Starlette server lifespan (where background tasks run)
+    and FastMCP session lifespans (where MCP tools need access to the streams).
+    """
+
+    document_send_stream: Optional[MemoryObjectSendStream] = None
+    document_receive_stream: Optional[MemoryObjectReceiveStream] = None
+    shutdown_event: Optional[anyio.Event] = None
+    scanner_wake_event: Optional[anyio.Event] = None
+
+
+# Module-level singleton for vector sync state
+_vector_sync_state = VectorSyncState()
+
+
 @dataclass
 class AppContext:
     """Application context for BasicAuth mode."""
@@ -243,17 +289,160 @@ class OAuthAppContext:
     )
 
 
+@dataclass
+class SmitheryAppContext:
+    """Application context for Smithery stateless mode.
+
+    ADR-016: No shared client - clients created per-request from session config.
+    """
+
+    pass  # No shared state needed - everything comes from session config
+
+
+# ADR-016: Smithery config schema for container runtime
+# This schema is served at /.well-known/mcp-config for Smithery discovery
+# See: https://smithery.ai/docs/build/session-config
+SMITHERY_CONFIG_SCHEMA = {
+    "$schema": "http://json-schema.org/draft-07/schema#",
+    "$id": "https://server.smithery.ai/nextcloud-mcp-server/.well-known/mcp-config",
+    "title": "Nextcloud MCP Server Configuration",
+    "description": "Configuration for connecting to your Nextcloud instance via app password authentication",
+    "x-query-style": "flat",  # Our schema has no nested objects, so flat style works
+    "type": "object",
+    "required": ["nextcloud_url", "username", "app_password"],
+    "properties": {
+        "nextcloud_url": {
+            "type": "string",
+            "title": "Nextcloud URL",
+            "description": "Your Nextcloud instance URL (e.g., https://cloud.example.com). Must be publicly accessible.",
+            "pattern": "^https?://.+",
+        },
+        "username": {
+            "type": "string",
+            "title": "Username",
+            "description": "Your Nextcloud username",
+            "minLength": 1,
+        },
+        "app_password": {
+            "type": "string",
+            "title": "App Password",
+            "description": "Nextcloud app password. Generate at Settings > Security > App passwords. Do NOT use your main password.",
+            "minLength": 1,
+        },
+    },
+    "additionalProperties": False,
+}
+
+
+# ADR-016: Context variable to hold Smithery session config per-request
+# This is set by SmitheryConfigMiddleware and accessed in context.py
+_smithery_session_config: ContextVar[dict[str, str] | None] = ContextVar(
+    "smithery_session_config"
+)
+_smithery_session_config.set(None)  # Set initial value
+
+
+def get_smithery_session_config() -> dict | None:
+    """Get the current Smithery session config from context variable.
+
+    Used by context.py to access config extracted from URL query parameters.
+    """
+    return _smithery_session_config.get()
+
+
+class SmitheryConfigMiddleware:
+    """Middleware to extract Smithery config from URL query parameters.
+
+    ADR-016: For container runtime, Smithery passes configuration as URL query
+    parameters to the /mcp endpoint. This middleware extracts those parameters
+    and stores them in a context variable for access in tools.
+
+    Configuration parameters:
+    - nextcloud_url: Nextcloud instance URL
+    - username: Nextcloud username
+    - app_password: Nextcloud app password
+
+    The extracted config is stored in a ContextVar and can be accessed via
+    get_smithery_session_config() in context.py.
+    """
+
+    def __init__(self, app: ASGIApp):
+        self.app = app
+
+    async def __call__(
+        self, scope: StarletteScope, receive: Receive, send: Send
+    ) -> None:
+        if scope["type"] == "http":
+            # Extract config from query parameters
+            from urllib.parse import parse_qs
+
+            query_string = scope.get("query_string", b"").decode("utf-8")
+            params = parse_qs(query_string)
+
+            # Build session config from query parameters
+            # Smithery uses dot notation for nested objects, but our schema is flat
+            session_config = {}
+            for key in ["nextcloud_url", "username", "app_password"]:
+                if key in params:
+                    # parse_qs returns lists, take first value
+                    session_config[key] = params[key][0]
+
+            # Store in context variable for access by context.py
+            if session_config:
+                _smithery_session_config.set(session_config)
+                logger.debug(
+                    f"Smithery config extracted: nextcloud_url={session_config.get('nextcloud_url')}, "
+                    f"username={session_config.get('username')}"
+                )
+
+        try:
+            await self.app(scope, receive, send)
+        finally:
+            # Clear context variable after request
+            _smithery_session_config.set(None)
+
+
+@asynccontextmanager
+async def app_lifespan_smithery(server: FastMCP) -> AsyncIterator[SmitheryAppContext]:
+    """
+    Manage application lifecycle for Smithery stateless mode.
+
+    ADR-016: Minimal lifespan with no shared state.
+    - No shared Nextcloud client (created per-request from session config)
+    - No vector sync (disabled in Smithery mode)
+    - No persistent storage (stateless deployment)
+    - No document processors (not enabled in Smithery mode)
+    """
+    logger.info("Starting MCP server in Smithery stateless mode")
+    logger.info("Clients will be created per-request from session config")
+
+    try:
+        yield SmitheryAppContext()
+    finally:
+        logger.info("Shutting down Smithery stateless mode")
+
+
 def is_oauth_mode() -> bool:
     """
     Determine if OAuth mode should be used.
 
     OAuth mode is enabled when:
     - NEXTCLOUD_USERNAME and NEXTCLOUD_PASSWORD are NOT set
+    - AND we are NOT in Smithery stateless mode
     - Or explicitly enabled via configuration
 
     Returns:
         True if OAuth mode, False if BasicAuth mode
     """
+    # ADR-016: Smithery stateless mode uses per-request BasicAuth from session config
+    # It's not OAuth mode even though env credentials aren't set
+    deployment_mode = get_deployment_mode()
+    if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+        logger.info(
+            "BasicAuth mode (Smithery stateless - credentials from session config)"
+        )
+        return False
+
     username = os.getenv("NEXTCLOUD_USERNAME")
     password = os.getenv("NEXTCLOUD_PASSWORD")
 
@@ -326,7 +515,7 @@ async def load_oauth_client_credentials(
         # and the authorization server will limit them to these allowed scopes.
         #
         # The PRM endpoint advertises the same scopes dynamically via @require_scopes decorators.
-        dcr_scopes = "openid profile email notes:read notes:write calendar:read calendar:write todo:read todo:write contacts:read contacts:write cookbook:read cookbook:write deck:read deck:write tables:read tables:write files:read files:write sharing:read sharing:write"
+        dcr_scopes = "openid profile email notes:read notes:write calendar:read calendar:write todo:read todo:write contacts:read contacts:write cookbook:read cookbook:write deck:read deck:write tables:read tables:write files:read files:write sharing:read sharing:write news:read news:write"
 
         # Add offline_access scope if refresh tokens are enabled
         enable_offline_access = os.getenv("ENABLE_OFFLINE_ACCESS", "false").lower() in (
@@ -386,15 +575,15 @@ async def load_oauth_client_credentials(
 @asynccontextmanager
 async def app_lifespan_basic(server: FastMCP) -> AsyncIterator[AppContext]:
     """
-    Manage application lifecycle for BasicAuth mode.
+    Manage application lifecycle for BasicAuth mode (FastMCP session lifespan).
 
     Creates a single Nextcloud client with basic authentication
-    that is shared across all requests.
+    that is shared across all requests within a session.
 
-    If vector sync is enabled (VECTOR_SYNC_ENABLED=true), also starts
-    background tasks for automatic document indexing (ADR-007).
+    Note: Background tasks (scanner, processor) are started at server level
+    in starlette_lifespan, not here. This lifespan runs per-session.
     """
-    logger.info("Starting MCP server in BasicAuth mode")
+    logger.info("Starting MCP session in BasicAuth mode")
     logger.info("Creating Nextcloud client with BasicAuth")
 
     client = NextcloudClient.from_env()
@@ -410,91 +599,20 @@ async def app_lifespan_basic(server: FastMCP) -> AsyncIterator[AppContext]:
     # Initialize document processors
     initialize_document_processors()
 
-    settings = get_settings()
-
-    # Check if vector sync is enabled
-    if settings.vector_sync_enabled:
-        logger.info("Vector sync enabled - starting background tasks")
-
-        # Get username from environment for BasicAuth mode
-        username = os.getenv("NEXTCLOUD_USERNAME")
-        if not username:
-            raise ValueError(
-                "NEXTCLOUD_USERNAME is required for vector sync in BasicAuth mode"
-            )
-
-        # Initialize Qdrant collection before starting background tasks
-        logger.info("Initializing Qdrant collection...")
-        from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
-
-        try:
-            await get_qdrant_client()  # Triggers collection creation if needed
-            logger.info("Qdrant collection ready")
-        except Exception as e:
-            logger.error(f"Failed to initialize Qdrant collection: {e}")
-            raise RuntimeError(
-                f"Cannot start vector sync - Qdrant initialization failed: {e}"
-            ) from e
-
-        # Initialize shared state
-        send_stream, receive_stream = anyio.create_memory_object_stream(
-            max_buffer_size=settings.vector_sync_queue_max_size
+    # Yield client context - scanner runs at server level (starlette_lifespan)
+    # Include vector sync state from module singleton (set by starlette_lifespan)
+    try:
+        yield AppContext(
+            client=client,
+            storage=storage,
+            document_send_stream=_vector_sync_state.document_send_stream,
+            document_receive_stream=_vector_sync_state.document_receive_stream,
+            shutdown_event=_vector_sync_state.shutdown_event,
+            scanner_wake_event=_vector_sync_state.scanner_wake_event,
         )
-        shutdown_event = anyio.Event()
-        scanner_wake_event = anyio.Event()
-
-        # Start background tasks using anyio TaskGroup
-        async with anyio.create_task_group() as tg:
-            # Start scanner task
-            await tg.start(
-                scanner_task,
-                send_stream,
-                shutdown_event,
-                scanner_wake_event,
-                client,
-                username,
-            )
-
-            # Start processor pool (each gets a cloned receive stream)
-            for i in range(settings.vector_sync_processor_workers):
-                await tg.start(
-                    processor_task,
-                    i,
-                    receive_stream.clone(),
-                    shutdown_event,
-                    client,
-                    username,
-                )
-
-            logger.info(
-                f"Background sync tasks started: 1 scanner + {settings.vector_sync_processor_workers} processors"
-            )
-
-            # Yield with background tasks running
-            try:
-                yield AppContext(
-                    client=client,
-                    storage=storage,
-                    document_send_stream=send_stream,
-                    document_receive_stream=receive_stream,
-                    shutdown_event=shutdown_event,
-                    scanner_wake_event=scanner_wake_event,
-                )
-            finally:
-                # Shutdown signal
-                logger.info("Shutting down background sync tasks")
-                shutdown_event.set()
-
-                # TaskGroup automatically cancels all tasks on exit
-                logger.info("Background sync tasks stopped")
-                await client.close()
-    else:
-        # No vector sync - simple lifecycle
-        try:
-            yield AppContext(client=client, storage=storage)
-        finally:
-            logger.info("Shutting down BasicAuth mode")
-            await client.close()
+    finally:
+        logger.info("Shutting down BasicAuth session")
+        await client.close()
 
 
 async def setup_oauth_config():
@@ -810,7 +928,7 @@ async def setup_oauth_config():
     )
 
 
-def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
+def get_app(transport: str = "streamable-http", enabled_apps: list[str] | None = None):
     # Initialize observability (logging will be configured by uvicorn)
     settings = get_settings()
 
@@ -837,8 +955,9 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
             "OpenTelemetry tracing disabled (set OTEL_EXPORTER_OTLP_ENDPOINT to enable)"
         )
 
-    # Determine authentication mode
+    # Determine authentication mode and deployment mode
     oauth_enabled = is_oauth_mode()
+    deployment_mode = get_deployment_mode()
 
     if oauth_enabled:
         logger.info("Configuring MCP server for OAuth mode")
@@ -899,8 +1018,17 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
             auth=auth_settings,
         )
     else:
-        logger.info("Configuring MCP server for BasicAuth mode")
-        mcp = FastMCP("Nextcloud MCP", lifespan=app_lifespan_basic)
+        # ADR-016: Use Smithery lifespan for stateless mode, BasicAuth otherwise
+        if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+            logger.info("Configuring MCP server for Smithery stateless mode")
+            # json_response=True returns plain JSON-RPC instead of SSE format,
+            # required for Smithery scanner compatibility
+            mcp = FastMCP(
+                "Nextcloud MCP", lifespan=app_lifespan_smithery, json_response=True
+            )
+        else:
+            logger.info("Configuring MCP server for BasicAuth mode")
+            mcp = FastMCP("Nextcloud MCP", lifespan=app_lifespan_basic)
 
     @mcp.resource("nc://capabilities")
     async def nc_get_capabilities():
@@ -919,6 +1047,7 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
         "contacts": configure_contacts_tools,
         "cookbook": configure_cookbook_tools,
         "deck": configure_deck_tools,
+        "news": configure_news_tools,
     }
 
     # If no specific apps are specified, enable all
@@ -936,8 +1065,12 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
             )
 
     # Register semantic search tools (cross-app feature)
+    # ADR-016: Skip in Smithery stateless mode (no vector database)
     settings = get_settings()
-    if settings.vector_sync_enabled:
+    deployment_mode = get_deployment_mode()
+    if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+        logger.info("Skipping semantic search tools (Smithery stateless mode)")
+    elif settings.vector_sync_enabled:
         logger.info("Configuring semantic search tools (vector sync enabled)")
         configure_semantic_tools(mcp)
     else:
@@ -1014,180 +1147,177 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
             "Dynamic tool filtering enabled for OAuth mode (JWT and Bearer tokens)"
         )
 
-    if transport == "sse":
-        mcp_app = mcp.sse_app()
-        starlette_lifespan = None
-    elif transport in ("http", "streamable-http"):
-        mcp_app = mcp.streamable_http_app()
+    mcp_app = mcp.streamable_http_app()
 
-        @asynccontextmanager
-        async def starlette_lifespan(app: Starlette):
-            # Set OAuth context for OAuth login routes (ADR-004)
-            if oauth_enabled:
-                # Prepare OAuth config from setup_oauth_config closure variables
-                mcp_server_url = os.getenv(
-                    "NEXTCLOUD_MCP_SERVER_URL", "http://localhost:8000"
-                )
-                nextcloud_resource_uri = os.getenv(
-                    "NEXTCLOUD_RESOURCE_URI", nextcloud_host
-                )
-                discovery_url = os.getenv(
-                    "OIDC_DISCOVERY_URL",
-                    f"{nextcloud_host}/.well-known/openid-configuration",
-                )
-                scopes = os.getenv("NEXTCLOUD_OIDC_SCOPES", "")
+    @asynccontextmanager
+    async def starlette_lifespan(app: Starlette):
+        # Set OAuth context for OAuth login routes (ADR-004)
+        if oauth_enabled:
+            # Prepare OAuth config from setup_oauth_config closure variables
+            mcp_server_url = os.getenv(
+                "NEXTCLOUD_MCP_SERVER_URL", "http://localhost:8000"
+            )
+            nextcloud_resource_uri = os.getenv("NEXTCLOUD_RESOURCE_URI", nextcloud_host)
+            discovery_url = os.getenv(
+                "OIDC_DISCOVERY_URL",
+                f"{nextcloud_host}/.well-known/openid-configuration",
+            )
+            scopes = os.getenv("NEXTCLOUD_OIDC_SCOPES", "")
 
-                oauth_context_dict = {
-                    "storage": refresh_token_storage,
-                    "oauth_client": oauth_client,
-                    "token_verifier": token_verifier,  # For querying IdP userinfo endpoint
-                    "config": {
-                        "mcp_server_url": mcp_server_url,
-                        "discovery_url": discovery_url,
-                        "client_id": client_id,  # From setup_oauth_config (DCR or static)
-                        "client_secret": client_secret,  # From setup_oauth_config (DCR or static)
-                        "scopes": scopes,
-                        "nextcloud_host": nextcloud_host,
-                        "nextcloud_resource_uri": nextcloud_resource_uri,
-                        "oauth_provider": oauth_provider,
-                    },
-                }
-                app.state.oauth_context = oauth_context_dict
+            oauth_context_dict = {
+                "storage": refresh_token_storage,
+                "oauth_client": oauth_client,
+                "token_verifier": token_verifier,  # For querying IdP userinfo endpoint
+                "config": {
+                    "mcp_server_url": mcp_server_url,
+                    "discovery_url": discovery_url,
+                    "client_id": client_id,  # From setup_oauth_config (DCR or static)
+                    "client_secret": client_secret,  # From setup_oauth_config (DCR or static)
+                    "scopes": scopes,
+                    "nextcloud_host": nextcloud_host,
+                    "nextcloud_resource_uri": nextcloud_resource_uri,
+                    "oauth_provider": oauth_provider,
+                },
+            }
+            app.state.oauth_context = oauth_context_dict
 
-                # Also set oauth_context on browser_app for session authentication
-                # browser_app is in the same function scope (defined later in create_app)
-                # We need to find it in the mounted routes
-                for route in app.routes:
-                    if isinstance(route, Mount) and route.path == "/app":
-                        route.app.state.oauth_context = oauth_context_dict
-                        logger.info(
-                            "OAuth context shared with browser_app for session auth"
-                        )
-                        break
-
-                logger.info(
-                    f"OAuth context initialized for login routes (client_id={client_id[:16]}...)"
-                )
-            else:
-                # BasicAuth mode - share storage with browser_app for webhook management
-                from nextcloud_mcp_server.auth.storage import RefreshTokenStorage
-
-                storage = RefreshTokenStorage.from_env()
-                await storage.initialize()
-
-                app.state.storage = storage
-
-                # Also share with browser_app for webhook routes
-                for route in app.routes:
-                    if isinstance(route, Mount) and route.path == "/app":
-                        route.app.state.storage = storage
-                        logger.info(
-                            "Storage shared with browser_app for webhook management"
-                        )
-                        break
-
-            # Start background vector sync tasks for BasicAuth mode (ADR-007)
-            # For streamable-http transport, FastMCP lifespan isn't automatically triggered
-            # so we manually start background tasks here if vector sync is enabled
-            import anyio as anyio_module
-
-            settings = get_settings()
-            if not oauth_enabled and settings.vector_sync_enabled:
-                logger.info("Starting background vector sync tasks for BasicAuth mode")
-
-                # Get username from environment
-                username = os.getenv("NEXTCLOUD_USERNAME")
-                if not username:
-                    raise ValueError(
-                        "NEXTCLOUD_USERNAME required for vector sync in BasicAuth mode"
+            # Also set oauth_context on browser_app for session authentication
+            # browser_app is in the same function scope (defined later in create_app)
+            # We need to find it in the mounted routes
+            for route in app.routes:
+                if isinstance(route, Mount) and route.path == "/app":
+                    route.app.state.oauth_context = oauth_context_dict
+                    logger.info(
+                        "OAuth context shared with browser_app for session auth"
                     )
+                    break
 
-                # Get Nextcloud client from MCP app context
-                # Create client since we're outside FastMCP lifespan
-                client = NextcloudClient.from_env()
+            logger.info(
+                f"OAuth context initialized for login routes (client_id={client_id[:16]}...)"
+            )
+        else:
+            # BasicAuth mode - share storage with browser_app for webhook management
+            from nextcloud_mcp_server.auth.storage import RefreshTokenStorage
 
-                # Initialize Qdrant collection before starting background tasks
-                logger.info("Initializing Qdrant collection...")
-                from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+            storage = RefreshTokenStorage.from_env()
+            await storage.initialize()
 
-                try:
-                    await get_qdrant_client()  # Triggers collection creation if needed
-                    logger.info("Qdrant collection ready")
-                except Exception as e:
-                    logger.error(f"Failed to initialize Qdrant collection: {e}")
-                    raise RuntimeError(
-                        f"Cannot start vector sync - Qdrant initialization failed: {e}"
-                    ) from e
+            app.state.storage = storage
 
-                # Initialize shared state
-                send_stream, receive_stream = anyio_module.create_memory_object_stream(
-                    max_buffer_size=settings.vector_sync_queue_max_size
+            # Also share with browser_app for webhook routes
+            for route in app.routes:
+                if isinstance(route, Mount) and route.path == "/app":
+                    route.app.state.storage = storage
+                    logger.info(
+                        "Storage shared with browser_app for webhook management"
+                    )
+                    break
+
+        # Start background vector sync tasks for BasicAuth mode (ADR-007)
+        # Scanner runs at server-level (once), not per-session
+        import anyio as anyio_module
+
+        settings = get_settings()
+        if not oauth_enabled and settings.vector_sync_enabled:
+            logger.info("Starting background vector sync tasks for BasicAuth mode")
+
+            # Get username from environment
+            username = os.getenv("NEXTCLOUD_USERNAME")
+            if not username:
+                raise ValueError(
+                    "NEXTCLOUD_USERNAME required for vector sync in BasicAuth mode"
                 )
-                shutdown_event = anyio_module.Event()
-                scanner_wake_event = anyio_module.Event()
 
-                # Store in app state for access from routes (ADR-007)
-                app.state.document_send_stream = send_stream
-                app.state.document_receive_stream = receive_stream
-                app.state.shutdown_event = shutdown_event
-                app.state.scanner_wake_event = scanner_wake_event
+            # Create client for vector sync (server-level, not per-session)
+            client = NextcloudClient.from_env()
 
-                # Also share with browser_app for /app route
-                for route in app.routes:
-                    if isinstance(route, Mount) and route.path == "/app":
-                        route.app.state.document_send_stream = send_stream
-                        route.app.state.document_receive_stream = receive_stream
-                        route.app.state.shutdown_event = shutdown_event
-                        route.app.state.scanner_wake_event = scanner_wake_event
-                        logger.info(
-                            "Vector sync state shared with browser_app for /app"
-                        )
-                        break
+            # Initialize Qdrant collection before starting background tasks
+            logger.info("Initializing Qdrant collection...")
+            from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
-                # Start background tasks using anyio TaskGroup
-                async with anyio_module.create_task_group() as tg:
-                    # Start scanner task
+            try:
+                await get_qdrant_client()  # Triggers collection creation if needed
+                logger.info("Qdrant collection ready")
+            except Exception as e:
+                logger.error(f"Failed to initialize Qdrant collection: {e}")
+                raise RuntimeError(
+                    f"Cannot start vector sync - Qdrant initialization failed: {e}"
+                ) from e
+
+            # Initialize shared state
+            send_stream, receive_stream = anyio_module.create_memory_object_stream(
+                max_buffer_size=settings.vector_sync_queue_max_size
+            )
+            shutdown_event = anyio_module.Event()
+            scanner_wake_event = anyio_module.Event()
+
+            # Store in app state for access from routes (ADR-007)
+            app.state.document_send_stream = send_stream
+            app.state.document_receive_stream = receive_stream
+            app.state.shutdown_event = shutdown_event
+            app.state.scanner_wake_event = scanner_wake_event
+
+            # Also store in module singleton for FastMCP session lifespans
+            _vector_sync_state.document_send_stream = send_stream
+            _vector_sync_state.document_receive_stream = receive_stream
+            _vector_sync_state.shutdown_event = shutdown_event
+            _vector_sync_state.scanner_wake_event = scanner_wake_event
+            logger.info("Vector sync state stored in module singleton")
+
+            # Also share with browser_app for /app route
+            for route in app.routes:
+                if isinstance(route, Mount) and route.path == "/app":
+                    route.app.state.document_send_stream = send_stream
+                    route.app.state.document_receive_stream = receive_stream
+                    route.app.state.shutdown_event = shutdown_event
+                    route.app.state.scanner_wake_event = scanner_wake_event
+                    logger.info("Vector sync state shared with browser_app for /app")
+                    break
+
+            # Start background tasks using anyio TaskGroup
+            async with anyio_module.create_task_group() as tg:
+                # Start scanner task
+                await tg.start(
+                    scanner_task,
+                    send_stream,
+                    shutdown_event,
+                    scanner_wake_event,
+                    client,
+                    username,
+                )
+
+                # Start processor pool (each gets a cloned receive stream)
+                for i in range(settings.vector_sync_processor_workers):
                     await tg.start(
-                        scanner_task,
-                        send_stream,
+                        processor_task,
+                        i,
+                        receive_stream.clone(),
                         shutdown_event,
-                        scanner_wake_event,
                         client,
                         username,
                     )
 
-                    # Start processor pool (each gets a cloned receive stream)
-                    for i in range(settings.vector_sync_processor_workers):
-                        await tg.start(
-                            processor_task,
-                            i,
-                            receive_stream.clone(),
-                            shutdown_event,
-                            client,
-                            username,
-                        )
+                logger.info(
+                    f"Background sync tasks started: 1 scanner + "
+                    f"{settings.vector_sync_processor_workers} processors"
+                )
 
-                    logger.info(
-                        f"Background sync tasks started: 1 scanner + "
-                        f"{settings.vector_sync_processor_workers} processors"
-                    )
-
-                    # Run MCP session manager and yield
-                    async with AsyncExitStack() as stack:
-                        await stack.enter_async_context(mcp.session_manager.run())
-                        try:
-                            yield
-                        finally:
-                            # Shutdown signal
-                            logger.info("Shutting down background sync tasks")
-                            shutdown_event.set()
-                            await client.close()
-                            # TaskGroup automatically cancels all tasks on exit
-            else:
-                # No vector sync - just run MCP session manager
+                # Run MCP session manager and yield
                 async with AsyncExitStack() as stack:
                     await stack.enter_async_context(mcp.session_manager.run())
-                    yield
+                    try:
+                        yield
+                    finally:
+                        # Shutdown signal
+                        logger.info("Shutting down background sync tasks")
+                        shutdown_event.set()
+                        await client.close()
+                        # TaskGroup automatically cancels all tasks on exit
+        else:
+            # No vector sync - just run MCP session manager
+            async with AsyncExitStack() as stack:
+                await stack.enter_async_context(mcp.session_manager.run())
+                yield
 
     # Health check endpoints for Kubernetes probes
     def health_live(request):
@@ -1340,6 +1470,26 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
     )
     logger.info("Test webhook endpoint enabled: /webhooks/nextcloud")
 
+    # ADR-016: Add Smithery well-known config endpoint for container runtime discovery
+    if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+
+        def smithery_mcp_config(request):
+            """Smithery MCP configuration endpoint.
+
+            Returns JSON Schema for Smithery's configuration UI.
+            This endpoint is required for Smithery container runtime discovery.
+            """
+            return JSONResponse(SMITHERY_CONFIG_SCHEMA)
+
+        routes.append(
+            Route(
+                "/.well-known/mcp-config",
+                smithery_mcp_config,
+                methods=["GET"],
+            )
+        )
+        logger.info("Smithery config endpoint enabled: /.well-known/mcp-config")
+
     # Note: Metrics endpoint is NOT exposed on main HTTP port for security reasons.
     # Metrics are served on dedicated port via setup_metrics() (default: 9090)
 
@@ -1470,71 +1620,98 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
         )
 
     # Add user info routes (available in both BasicAuth and OAuth modes)
-    # These require session authentication, so we wrap them in a separate app
-    from nextcloud_mcp_server.auth.session_backend import SessionAuthBackend
-    from nextcloud_mcp_server.auth.userinfo_routes import (
-        revoke_session,
-        user_info_html,
-        vector_sync_status_fragment,
-    )
-    from nextcloud_mcp_server.auth.viz_routes import (
-        vector_visualization_html,
-        vector_visualization_search,
-    )
-    from nextcloud_mcp_server.auth.webhook_routes import (
-        disable_webhook_preset,
-        enable_webhook_preset,
-        webhook_management_pane,
-    )
-
-    # Create a separate Starlette app for browser routes that need session auth
-    # This prevents SessionAuthBackend from interfering with FastMCP's OAuth
-    browser_routes = [
-        Route("/", user_info_html, methods=["GET"]),  # /app → webapp (HTML UI)
-        Route(
-            "/revoke", revoke_session, methods=["POST"], name="revoke_session_endpoint"
-        ),  # /app/revoke → revoke_session
-        # Vector sync status fragment (htmx polling)
-        Route(
-            "/vector-sync/status",
+    # ADR-016: Skip /app admin UI in Smithery stateless mode (no vector sync, webhooks)
+    if deployment_mode != DeploymentMode.SMITHERY_STATELESS:
+        # These require session authentication, so we wrap them in a separate app
+        from nextcloud_mcp_server.auth.session_backend import SessionAuthBackend
+        from nextcloud_mcp_server.auth.userinfo_routes import (
+            revoke_session,
+            user_info_html,
             vector_sync_status_fragment,
-            methods=["GET"],
-        ),  # /app/vector-sync/status
-        # Vector visualization routes
-        Route(
-            "/vector-viz", vector_visualization_html, methods=["GET"]
-        ),  # /app/vector-viz
-        Route(
-            "/vector-viz/search",
+        )
+        from nextcloud_mcp_server.auth.viz_routes import (
+            chunk_context_endpoint,
+            vector_visualization_html,
             vector_visualization_search,
-            methods=["GET"],
-        ),  # /app/vector-viz/search
-        # Webhook management routes (admin-only)
-        Route("/webhooks", webhook_management_pane, methods=["GET"]),  # /app/webhooks
-        Route(
-            "/webhooks/enable/{preset_id:str}", enable_webhook_preset, methods=["POST"]
-        ),
-        Route(
-            "/webhooks/disable/{preset_id:str}",
+        )
+        from nextcloud_mcp_server.auth.webhook_routes import (
             disable_webhook_preset,
-            methods=["DELETE"],
-        ),
-    ]
+            enable_webhook_preset,
+            webhook_management_pane,
+        )
 
-    browser_app = Starlette(routes=browser_routes)
-    browser_app.add_middleware(
-        AuthenticationMiddleware,
-        backend=SessionAuthBackend(oauth_enabled=oauth_enabled),
-    )
+        # Create a separate Starlette app for browser routes that need session auth
+        # This prevents SessionAuthBackend from interfering with FastMCP's OAuth
+        browser_routes = [
+            Route(
+                "/", user_info_html, methods=["GET"]
+            ),  # /app → user info with all tabs
+            Route(
+                "/revoke",
+                revoke_session,
+                methods=["POST"],
+                name="revoke_session_endpoint",
+            ),  # /app/revoke → revoke_session
+            # Vector sync status fragment (htmx polling)
+            Route(
+                "/vector-sync/status",
+                vector_sync_status_fragment,
+                methods=["GET"],
+            ),  # /app/vector-sync/status
+            # Vector visualization routes
+            Route(
+                "/vector-viz", vector_visualization_html, methods=["GET"]
+            ),  # /app/vector-viz
+            Route(
+                "/vector-viz/search",
+                vector_visualization_search,
+                methods=["GET"],
+            ),  # /app/vector-viz/search
+            Route(
+                "/chunk-context",
+                chunk_context_endpoint,
+                methods=["GET"],
+            ),  # /app/chunk-context
+            # Webhook management routes (admin-only)
+            Route(
+                "/webhooks", webhook_management_pane, methods=["GET"]
+            ),  # /app/webhooks
+            Route(
+                "/webhooks/enable/{preset_id:str}",
+                enable_webhook_preset,
+                methods=["POST"],
+            ),
+            Route(
+                "/webhooks/disable/{preset_id:str}",
+                disable_webhook_preset,
+                methods=["DELETE"],
+            ),
+        ]
 
-    # Add redirect from /app to /app/ (Starlette requires trailing slash for mounted apps)
-    routes.append(
-        Route("/app", lambda request: RedirectResponse("/app/", status_code=307))
-    )
+        # Add static files mount if directory exists
+        static_dir = os.path.join(os.path.dirname(__file__), "auth", "static")
+        if os.path.isdir(static_dir):
+            browser_routes.append(
+                Mount("/static", StaticFiles(directory=static_dir), name="static")
+            )
+            logger.info(f"Mounted static files from {static_dir}")
 
-    # Mount browser app at /app (webapp and admin routes)
-    routes.append(Mount("/app", app=browser_app))
-    logger.info("App routes with session auth: /app, /app/webhooks, /app/revoke")
+        browser_app = Starlette(routes=browser_routes)
+        browser_app.add_middleware(
+            AuthenticationMiddleware,  # type: ignore[invalid-argument-type]
+            backend=SessionAuthBackend(oauth_enabled=oauth_enabled),
+        )
+
+        # Add redirect from /app to /app/ (Starlette requires trailing slash for mounted apps)
+        routes.append(
+            Route("/app", lambda request: RedirectResponse("/app/", status_code=307))
+        )
+
+        # Mount browser app at /app (webapp and admin routes)
+        routes.append(Mount("/app", app=browser_app))
+        logger.info("App routes with session auth: /app, /app/webhooks, /app/revoke")
+    else:
+        logger.info("Admin UI (/app) disabled in Smithery stateless mode")
 
     # Mount FastMCP at root last (catch-all, handles OAuth via token_verifier)
     routes.append(Mount("/", app=mcp_app))
@@ -1613,7 +1790,7 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
 
     # Add CORS middleware to allow browser-based clients like MCP Inspector
     app.add_middleware(
-        CORSMiddleware,
+        CORSMiddleware,  # type: ignore[invalid-argument-type]
         allow_origins=["*"],  # Allow all origins for development
         allow_credentials=True,
         allow_methods=["*"],
@@ -1623,7 +1800,7 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
 
     # Add observability middleware (metrics + tracing)
     if settings.metrics_enabled or settings.otel_exporter_otlp_endpoint:
-        app.add_middleware(ObservabilityMiddleware)
+        app.add_middleware(ObservabilityMiddleware)  # type: ignore[invalid-argument-type]
         logger.info("Observability middleware enabled (metrics and/or tracing)")
 
     # Add exception handler for scope challenges (OAuth mode only)
@@ -1654,4 +1831,11 @@ def get_app(transport: str = "sse", enabled_apps: list[str] | None = None):
 
         logger.info("WWW-Authenticate scope challenge handler enabled")
 
+    # ADR-016: Apply SmitheryConfigMiddleware in Smithery stateless mode
+    # This must be the outermost middleware to extract config from URL query parameters
+    # before any other middleware processes the request
+    if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+        app = SmitheryConfigMiddleware(app)
+        logger.info("SmitheryConfigMiddleware enabled for query parameter config")
+
     return app

nextcloud_mcp_server/auth/static/vector-viz.css

@@ -0,0 +1,219 @@
+.viz-layout {
+    display: flex;
+    flex-direction: column;
+    gap: 16px;
+    height: 100%;
+    min-height: 0;
+    overflow-y: auto;
+}
+.viz-card {
+    background: var(--color-main-background);
+    border-radius: 0;
+    padding: 16px;
+    box-shadow: none;
+}
+.viz-controls-card {
+    flex: 0 0 auto;
+    border-bottom: 1px solid var(--color-border);
+    padding-bottom: 16px;
+}
+.viz-controls-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(200px, 1fr));
+    gap: 12px;
+    align-items: end;
+}
+@media (min-width: 768px) {
+    .viz-controls-grid {
+        grid-template-columns: 2fr 1.5fr 1.5fr auto auto;
+    }
+}
+.viz-control-group {
+    display: flex;
+    flex-direction: column;
+    gap: 4px;
+}
+.viz-control-group label {
+    font-weight: 500;
+    color: var(--color-main-text);
+    font-size: 13px;
+}
+.viz-control-group input[type="text"],
+.viz-control-group input[type="number"],
+.viz-control-group select {
+    width: 100%;
+    padding: 7px 10px;
+    border: 1px solid var(--color-border-dark);
+    border-radius: var(--border-radius);
+    font-size: 14px;
+    background: var(--color-main-background);
+    color: var(--color-main-text);
+}
+.viz-control-group input:focus,
+.viz-control-group select:focus {
+    outline: none;
+    border-color: var(--color-primary-element);
+}
+.viz-control-group input[type="range"] {
+    width: 100%;
+}
+.viz-control-group select[multiple] {
+    min-height: 100px;
+}
+.viz-weight-display {
+    display: inline-block;
+    min-width: 40px;
+    text-align: right;
+    color: #666;
+}
+.viz-btn {
+    background: var(--color-primary-element);
+    color: white;
+    border: none;
+    padding: 7px 16px;
+    border-radius: var(--border-radius);
+    cursor: pointer;
+    font-size: 14px;
+    font-weight: 500;
+    white-space: nowrap;
+}
+.viz-btn:hover {
+    background: #0052a3;
+}
+.viz-btn-secondary {
+    background: #6c757d;
+    color: white;
+    border: none;
+    padding: 7px 16px;
+    border-radius: var(--border-radius);
+    cursor: pointer;
+    font-size: 14px;
+    white-space: nowrap;
+}
+.viz-btn-secondary:hover {
+    background: #5a6268;
+}
+.viz-card-plot {
+    flex: 0 0 auto;
+    display: flex;
+    flex-direction: column;
+    min-height: 500px;
+    height: 600px;
+    /* Remove horizontal padding to extend to full viewport width */
+    padding-left: 0;
+    padding-right: 0;
+    margin-left: -16px;
+    margin-right: -16px;
+}
+#viz-plot-container {
+    width: 100%;
+    height: 100%;
+    position: relative;
+    overflow: visible;
+}
+#viz-plot {
+    width: 100%;
+    height: 100%;
+}
+.viz-loading {
+    text-align: center;
+    padding: 40px;
+    color: #666;
+}
+.viz-loading-overlay {
+    position: absolute;
+    inset: 0;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    background: white;
+    color: #666;
+}
+.viz-no-results {
+    text-align: center;
+    padding: 40px;
+    color: #666;
+    font-style: italic;
+}
+.viz-advanced-section {
+    margin-top: 12px;
+    padding: 12px;
+    background: var(--color-background-hover);
+    border-radius: var(--border-radius);
+    border: 1px solid var(--color-border);
+}
+.viz-info-box {
+    background: var(--color-primary-element-light);
+    border-left: 3px solid var(--color-primary-element);
+    padding: 10px 12px;
+    margin-bottom: 16px;
+    font-size: 13px;
+    color: var(--color-main-text);
+}
+.chunk-toggle-btn {
+    background: #6c757d;
+    color: white;
+    border: none;
+    padding: 4px 10px;
+    border-radius: 3px;
+    cursor: pointer;
+    font-size: 12px;
+    margin-top: 6px;
+}
+.chunk-toggle-btn:hover {
+    background: #5a6268;
+}
+.chunk-context {
+    background: var(--color-background-hover);
+    border: 1px solid var(--color-border);
+    border-radius: var(--border-radius);
+    padding: 12px;
+    margin-top: 8px;
+    font-family: 'SFMono-Regular', 'Consolas', 'Liberation Mono', 'Menlo', monospace;
+    font-size: 13px;
+    line-height: 1.6;
+    white-space: pre-wrap;
+    word-wrap: break-word;
+}
+.chunk-text {
+    color: var(--color-text-maxcontrast);
+}
+.chunk-matched {
+    background: #fff3cd;
+    border: 1px solid #ffc107;
+    padding: 2px 4px;
+    border-radius: var(--border-radius);
+    font-weight: 500;
+    color: var(--color-main-text);
+}
+.chunk-ellipsis {
+    color: var(--color-text-maxcontrast);
+    font-style: italic;
+}
+
+/* PDF highlighted image styles */
+.chunk-image-container {
+    margin-bottom: 16px;
+    border: 1px solid var(--color-border);
+    border-radius: var(--border-radius);
+    overflow: hidden;
+    background: #fff;
+}
+.chunk-image-header {
+    background: var(--color-background-dark);
+    padding: 8px 12px;
+    font-size: 12px;
+    font-weight: 500;
+    color: var(--color-text-maxcontrast);
+    border-bottom: 1px solid var(--color-border);
+    font-family: var(--font-face);
+}
+.chunk-highlighted-image {
+    display: block;
+    max-width: 100%;
+    height: auto;
+    cursor: zoom-in;
+}
+.chunk-highlighted-image:hover {
+    opacity: 0.95;
+}

nextcloud_mcp_server/auth/static/vector-viz.js

@@ -0,0 +1,253 @@
+// Initialize vizApp for vector visualization
+function vizApp() {
+    return {
+        query: '',
+        algorithm: 'bm25_hybrid',
+        fusion: 'rrf',
+        showAdvanced: false,
+        showQueryPoint: true,
+        docTypes: [''],
+        limit: 50,
+        scoreThreshold: 0.0,
+        loading: false,
+        results: [],
+        coordinates: null,
+        queryCoords: null,
+        expandedChunks: {},
+        chunkLoading: {},
+
+        init() {
+            // Set up window resize listener to resize plot
+            window.addEventListener('resize', () => {
+                if (this.coordinates && this.results.length > 0) {
+                    Plotly.Plots.resize('viz-plot');
+                }
+            });
+        },
+
+        async executeSearch() {
+            this.loading = true;
+            this.results = [];
+
+            try {
+                const params = new URLSearchParams({
+                    query: this.query,
+                    algorithm: this.algorithm,
+                    limit: this.limit,
+                    score_threshold: this.scoreThreshold,
+                });
+
+                if (this.algorithm === 'bm25_hybrid') {
+                    params.append('fusion', this.fusion);
+                }
+
+                const selectedTypes = this.docTypes.filter(t => t !== '');
+                if (selectedTypes.length > 0) {
+                    params.append('doc_types', selectedTypes.join(','));
+                }
+
+                const response = await fetch(`/app/vector-viz/search?${params}`);
+                const data = await response.json();
+
+                if (data.success) {
+                    this.results = data.results;
+                    this.coordinates = data.coordinates_3d;
+                    this.queryCoords = data.query_coords;
+                    this.renderPlot(this.coordinates, this.queryCoords, this.results);
+                } else {
+                    alert('Search failed: ' + data.error);
+                }
+            } catch (error) {
+                alert('Error: ' + error.message);
+            } finally {
+                this.loading = false;
+            }
+        },
+
+        updatePlot() {
+            // Toggle query point visibility without recreating the plot
+            // This preserves camera position naturally since layout is untouched
+            if (this.coordinates && this.queryCoords && this.results.length > 0) {
+                const plotDiv = document.getElementById('viz-plot');
+
+                // If plot exists, just toggle the query trace visibility
+                if (plotDiv && plotDiv.data && plotDiv.data.length >= 2) {
+                    // Trace index 1 is the query point
+                    Plotly.restyle('viz-plot', { visible: this.showQueryPoint }, [1]);
+                } else {
+                    // Plot doesn't exist yet, render it
+                    this.renderPlot(this.coordinates, this.queryCoords, this.results);
+                }
+            }
+        },
+
+        renderPlot(coordinates, queryCoords, results) {
+            // Get container dimensions before creating layout
+            const container = document.getElementById('viz-plot-container');
+            const width = container.clientWidth;
+            const height = container.clientHeight;
+
+            const scores = results.map(r => r.score);
+
+            // Trace 1: Document results (always visible)
+            const documentTrace = {
+                x: coordinates.map(c => c[0]),
+                y: coordinates.map(c => c[1]),
+                z: coordinates.map(c => c[2]),
+                mode: 'markers',
+                type: 'scatter3d',
+                name: 'Documents',
+                visible: true,
+                customdata: results.map((r, i) => ({
+                    title: r.title,
+                    raw_score: r.original_score,
+                    relative_score: r.score,
+                    x: coordinates[i][0],
+                    y: coordinates[i][1],
+                    z: coordinates[i][2]
+                })),
+                hovertemplate:
+                    '<b>%{customdata.title}</b><br>' +
+                    'Raw Score: %{customdata.raw_score:.3f} (%{customdata.relative_score:.0%} relative)<br>' +
+                    '(x=%{customdata.x}, y=%{customdata.y}, z=%{customdata.z})' +
+                    '<extra></extra>',
+                marker: {
+                    size: results.map(r => 4 + (Math.pow(r.score, 2) * 10)),
+                    opacity: results.map(r => 0.3 + (r.score * 0.7)),
+                    color: scores,
+                    colorscale: 'Viridis',
+                    showscale: true,
+                    colorbar: {
+                        title: 'Relative Score',
+                        x: 1.02,
+                        xanchor: 'left',
+                        thickness: 20,
+                        len: 0.8
+                    },
+                    cmin: 0,
+                    cmax: 1
+                }
+            };
+
+            // Trace 2: Query point (visibility controlled by toggle)
+            const queryTrace = {
+                x: [queryCoords[0]],
+                y: [queryCoords[1]],
+                z: [queryCoords[2]],
+                mode: 'markers',
+                type: 'scatter3d',
+                name: 'Query',
+                visible: this.showQueryPoint,  // Initial visibility from state
+                hovertemplate:
+                    '<b>Search Query</b><br>' +
+                    `(x=${queryCoords[0]}, y=${queryCoords[1]}, z=${queryCoords[2]})` +
+                    '<extra></extra>',
+                marker: {
+                    size: 10,
+                    color: '#ef5350',  // Subdued red (Material Design Red 400)
+                    line: {
+                        color: '#c62828',  // Darker red border (Material Design Red 800)
+                        width: 1
+                    }
+                }
+            };
+
+            const layout = {
+                title: `Vector Space (PCA 3D) - ${results.length} results`,
+                width: width,   // Explicit width from container
+                height: height, // Explicit height from container
+                scene: {
+                    xaxis: { title: 'PC1' },
+                    yaxis: { title: 'PC2' },
+                    zaxis: { title: 'PC3' },
+                    camera: {
+                        eye: { x: 1.5, y: 1.5, z: 1.5 }
+                    },
+                    // Full width for 3D scene
+                    domain: {
+                        x: [0, 1],
+                        y: [0, 1]
+                    }
+                },
+                hovermode: 'closest',
+                autosize: true,  // Enable auto-sizing for window resizes
+                showlegend: false,  // Hide legend
+                margin: { l: 0, r: 100, t: 40, b: 0 }  // Right margin for colorbar
+            };
+
+            // Always render both traces - visibility is controlled by the visible property
+            const traces = [documentTrace, queryTrace];
+
+            // Enable responsive resizing
+            const config = {
+                responsive: true,
+                displayModeBar: true
+            };
+
+            // Use newPlot() with explicit dimensions - renders at correct size immediately
+            // Camera position will be preserved by subsequent Plotly.restyle() calls in updatePlot()
+            Plotly.newPlot('viz-plot', traces, layout, config);
+        },
+
+        getNextcloudUrl(result) {
+            // Use global NEXTCLOUD_BASE_URL if set, otherwise construct from window location
+            const baseUrl = window.NEXTCLOUD_BASE_URL || '';
+            switch (result.doc_type) {
+                case 'note':
+                    return `${baseUrl}/apps/notes/note/${result.id}`;
+                case 'file':
+                    return `${baseUrl}/apps/files/?fileId=${result.id}`;
+                case 'calendar':
+                    return `${baseUrl}/apps/calendar`;
+                case 'contact':
+                    return `${baseUrl}/apps/contacts`;
+                case 'deck':
+                    return `${baseUrl}/apps/deck`;
+                default:
+                    return `${baseUrl}`;
+            }
+        },
+
+        hasChunkPosition(result) {
+            return result.chunk_start_offset != null && result.chunk_end_offset != null;
+        },
+
+        isChunkExpanded(resultKey) {
+            return this.expandedChunks[resultKey] !== undefined;
+        },
+
+        async toggleChunk(result) {
+            const resultKey = `${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`;
+
+            if (this.isChunkExpanded(resultKey)) {
+                delete this.expandedChunks[resultKey];
+                return;
+            }
+
+            this.chunkLoading[resultKey] = true;
+
+            try {
+                const params = new URLSearchParams({
+                    doc_type: result.doc_type,
+                    doc_id: result.id,
+                    start: result.chunk_start_offset,
+                    end: result.chunk_end_offset,
+                    context: 500
+                });
+
+                const response = await fetch(`/app/chunk-context?${params}`);
+                const data = await response.json();
+
+                if (data.success) {
+                    this.expandedChunks[resultKey] = data;
+                } else {
+                    alert('Failed to load chunk: ' + data.error);
+                }
+            } catch (error) {
+                alert('Error loading chunk: ' + error.message);
+            } finally {
+                delete this.chunkLoading[resultKey];
+            }
+        }
+    };
+}

nextcloud_mcp_server/auth/templates/base.html

@@ -0,0 +1,524 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta http-equiv="X-UA-Compatible" content="IE=edge">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1">
+    <meta name="apple-mobile-web-app-capable" content="yes">
+    <meta name="theme-color" content="#0082c9">
+    <title>{% block title %}Nextcloud MCP Server{% endblock %}</title>
+
+    <!-- Favicon -->
+    <link rel="icon" type="image/svg+xml" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' width='32' height='32' viewBox='0 0 512 512'><rect width='512' height='512' rx='80' ry='80' fill='%230082C9'/><path d='M255.9 21.04c-11.8 0-22.2 4.08-28.6 10.01-5.6 4.98-8.6 11.41-8.6 18.11 0 5.55 2.2 11.01 5.9 15.48-16.4 4.97-30.1 13.64-39 24.53 22.1-7.67 45.7-11.86 70.3-11.86 24.6 0 48.3 4.19 70.3 11.86-8.9-10.89-22.6-19.56-39-24.53 3.9-4.47 5.9-9.93 5.9-15.48 0-6.7-3-13.13-8.5-18.11-6.4-5.93-16.9-10.01-28.7-10.01zm0 20.34c5.3 0 10.1 1.27 13.6 3.52 1.7 1.16 3.4 2.43 3.4 4.27 0 1.76-1.7 3.03-3.4 4.19-3.5 2.33-8.3 3.61-13.6 3.61-5.3 0-10.1-1.28-13.6-3.61-1.6-1.16-3.3-2.43-3.3-4.19 0-1.84 1.7-3.11 3.3-4.27 3.5-2.25 8.3-3.52 13.6-3.52zm.1 48.1c-110.8 0-200.72 90.02-200.72 200.82S145.2 491 256 491s200.7-89.9 200.7-200.7c0-110.8-89.9-200.82-200.7-200.82zm0 32.62c92.9 0 168.2 75.3 168.2 168.2 0 92.8-75.3 168.2-168.2 168.2-92.9 0-168.26-75.4-168.26-168.2 0-92.9 75.36-168.2 168.26-168.2zm-8.2 6.3c-9.6.5-19 1.9-28.3 4.1l2.3 7.8c8.4-2 17.1-3.3 26-3.8v-8.1zm16.2 0v8.1c9 .5 17.7 1.8 26 3.8l2.2-7.8c-9.1-2.2-18.6-3.6-28.2-4.1zm-60 8.5c-9 3.2-17.6 7-25.8 11.6l4.1 7.1c7.7-4.3 15.6-7.9 23.9-10.8l-2.2-7.9zm103.7 0-2 7.9c8.4 2.9 16.2 6.5 23.8 10.8l4.2-7.1c-8.2-4.6-16.9-8.4-26-11.6zm-143.3 20.3c-7.5 5.4-14.6 11.4-21.1 17.9l5.8 5.8c5.9-6.1 12.5-11.7 19.5-16.6l-4.2-7.1zm182.9 0-4 7.1c6.9 4.9 13.5 10.5 19.5 16.6l5.7-5.8c-6.5-6.5-13.7-12.5-21.2-17.9zm-91.4 11.5c-37 0-67.4 28.6-70.3 64.9l15.9 4.7c.7-29.6 24.7-53.4 54.4-53.4 30.1 0 54.4 24.4 54.4 54.3 0 15-6.2 28.7-16 38.5l.1.1c1.7 2.7 3 5.6 4.1 8.6.9 3 1.7 5.7 2.3 8.6v.4c33.8-16.7 57.2-51.5 57.2-91.7 0-3.8-.2-7.3-.6-10.9-3.2-3.3-6.3-6.4-9.8-9.5 1.5 6.5 2.3 13.4 2.3 20.4 0 28.7-13 54.7-33.5 71.8 6.3-10.6 10.1-23 10.1-36.3 0-38.9-31.7-70.5-70.6-70.5zm-91.8 14.6c-3.3 3.1-6.5 6.2-9.7 9.5-.3 3.6-.5 7.1-.5 10.9 0 7.3.7 14.2 2.1 20.9l9.1 2.7c-2.1-7.5-3.1-15.4-3.1-23.6 0-7 .7-13.9 2.1-20.4zm-31.6 4c-5.8 7.1-10.9 14.6-15.4 22.6l7.1 4c4.1-7.4 8.8-14.3 14-20.8l-5.7-5.8zm246.8 0-5.7 5.8c5.3 6.5 10 13.4 13.9 20.8l7.1-4c-4.4-8-9.5-15.5-15.3-22.6zm-269.2 37.1c-2.5 5.7-4.6 11.4-6.4 17.6l.1-.3c3.4-5 7.9-9.3 12.9-12.5l.3-.6-6.9-4.2zm291.8 0-7.2 4.2c3.2 7.3 5.7 15.1 7.6 23.1l7.9-2.1c-2.1-8.8-4.9-17.3-8.3-25.2zm-261.2 11.5c-13.4.1-25.7 9-29.7 22.5l114.8 34.2c-4.9 16.7 4.6 34.2 21.2 39.2L361.7 366c16.6 5 34.1-4.4 39.1-21l-114.6-34.4c4.9-16.5-4.7-34.1-21.3-39.1 0 0-72.4-21.5-114.8-34.3-3.1-.9-6.3-1.4-9.4-1.3zm-42.09 29.7c-.9 6.9-1.4 14-1.4 21.3 0 1.3.1 2.9.1 4.2h8.09v-4.2c0-6.5.4-12.9 1.2-19.2l-7.99-2.1zm314.59 0-7.9 2.1c.7 6.3 1.3 12.7 1.3 19.2 0 1.3 0 2.9-.2 4.2h8.2v-4.2c0-7.3-.5-14.4-1.4-21.3zm-157.3 24.7c6.3 0 11.5 5 11.5 11.3 0 6.4-5.2 11.6-11.5 11.6s-11.5-5.2-11.5-11.6c0-6.3 5.2-11.3 11.5-11.3zM98.51 307.4c1 8.2 2.89 16.4 5.09 24.3l7.9-2.1c-2.1-7.2-3.8-14.6-4.8-22.2h-8.19zm306.69 0c-1.1 7.6-2.7 15-4.8 22.2l7.8 2.1c2.2-7.9 4.1-16.1 5.2-24.3h-8.2zm-191.3 10.9c-19 13.3-31.4 35.3-31.4 60.1 0 10.4 2.3 20.4 6.2 29.7 8.8 4.9 17.9 8.8 27.6 11.7-10.8-10.7-17.5-25.2-17.5-41.4 0-19 9.3-36 23.7-46.3-3.8-4.1-6.7-8.7-8.6-13.8zM116.8 345l-7.9 2c3.1 7.6 6.8 14.7 11 21.6l6.9-4.2c-3.8-6.2-7-12.8-10-19.4zm194.8 20.5c.9 4.1 1.4 8.5 1.4 12.9 0 16.2-6.7 30.7-17.4 41.4 9.6-2.9 18.8-6.8 27.5-11.7 4-9.3 6.2-19.3 6.2-29.7 0-2.7-.2-5.2-.4-7.7l-17.3-5.2zM136 377.9l-7.1 4.1c4.7 6.2 9.7 12.1 15.3 17.3l5.7-5.5c-5.1-5-9.7-10.3-13.9-15.9zm243.9 2.3-.2.1c-2.1.3-4 .6-6.2.7h-.1c-3.6 4.5-7.3 8.8-11.5 12.8l5.8 5.5c5.5-5.2 10.5-11.1 15.2-17.3l-3-1.8zm-217.8 24-5.9 5.9c6 4.8 12.2 9.7 18.8 13.6l3.8-7.8c-5.7-2.9-11.4-6.8-16.7-11.7zm187.7 0c-5.4 4.9-11.1 8.8-16.8 11.7l3.9 7.8c6.5-3.9 12.8-8.8 18.7-13.6l-5.8-5.9zm-156.4 19.5-4.1 6.8c6.6 4 13.7 5.8 20.7 8.8l2.2-7.9c-6.5-1.9-12.7-4.8-18.8-7.7zm125.2 0c-6.2 2.9-12.5 5.8-19.1 7.7l2.3 7.9c7.2-3 14-4.8 20.7-8.8l-3.9-6.8zm-90.7 11.7-2 7.8c7.1 1 14.5 1.9 21.9 1.9v-7.7c-6.8 0-13.5-1.1-19.9-2zm55.9 0c-6.3.9-13 2-19.8 2v7.7c7.5 0 14.8-.9 22.1-1.9l-2.3-7.8z' fill='%23fff'/></svg>">
+
+    <!-- Open Sans font -->
+    <style>
+        @font-face {
+            font-family: 'Open Sans';
+            font-style: normal;
+            font-weight: normal;
+            src: local('Open Sans'), local('OpenSans');
+        }
+        @font-face {
+            font-family: 'Open Sans';
+            font-style: normal;
+            font-weight: bold;
+            src: local('Open Sans Semibold'), local('OpenSans-Semibold');
+        }
+    </style>
+
+    {% block extra_head %}{% endblock %}
+
+    <style>
+        /* Nextcloud App Design System */
+
+        /* CSS Variables */
+        :root {
+            /* Primary Colors */
+            --color-primary: #00679e;
+            --color-primary-element: #00679e;
+            --color-primary-light: #e5eff5;
+            --color-primary-element-light: #e5eff5;
+
+            /* Background Colors */
+            --color-main-background: #ffffff;
+            --color-background-dark: #ededed;
+            --color-background-hover: #f5f5f5;
+
+            /* Text Colors */
+            --color-main-text: #222222;
+            --color-text-maxcontrast: #6b6b6b;
+            --color-text-light: #767676;
+
+            /* Border Colors */
+            --color-border: #ededed;
+            --color-border-dark: #dbdbdb;
+
+            /* Borders & Radius */
+            --border-radius: 3px;
+            --border-radius-large: 10px;
+            --border-radius-pill: 100px;
+
+            /* Spacing */
+            --default-grid-baseline: 4px;
+            --default-clickable-area: 44px;
+        }
+
+        /* SVG Icon Styles */
+        .nav-icon {
+            width: 20px;
+            height: 20px;
+            display: inline-block;
+            fill: var(--color-main-text);
+            opacity: 0.7;
+        }
+
+        .app-navigation-entry.active .nav-icon {
+            fill: var(--color-primary-element);
+            opacity: 1;
+        }
+
+        /* General */
+        * {
+            box-sizing: border-box;
+        }
+
+        body {
+            font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
+            color: var(--color-main-text);
+            background: var(--color-main-background);
+            margin: 0;
+            padding: 0;
+        }
+
+        h1, h2, h3 {
+            font-weight: 300;
+            line-height: 1.2;
+        }
+
+        h1 {
+            font-size: 32px;
+            margin: 0 0 20px 0;
+            color: var(--color-main-text);
+        }
+
+        h2 {
+            font-size: 20px;
+            margin: 20px 0 12px 0;
+            color: var(--color-main-text);
+            border-bottom: 1px solid var(--color-border);
+            padding-bottom: 8px;
+        }
+
+        h3 {
+            font-size: 16px;
+            margin: 16px 0 8px 0;
+            color: var(--color-main-text);
+            font-weight: 500;
+        }
+
+        img {
+            max-width: 100%;
+        }
+
+        /* App Header (simplified, no full menu) */
+        .app-header {
+            height: 50px;
+            background: var(--color-primary-element);
+            box-shadow: 0 2px 4px rgba(0,0,0,0.1);
+            position: sticky;
+            top: 0;
+            z-index: 100;
+            display: flex;
+            align-items: center;
+            padding: 0 20px;
+        }
+
+        .app-header__brand {
+            color: white;
+            font-size: 18px;
+            font-weight: 600;
+            text-decoration: none;
+            display: flex;
+            align-items: center;
+            gap: 12px;
+        }
+
+        .app-header__brand:hover {
+            opacity: 0.9;
+        }
+
+        .app-header__logo {
+            height: 32px;
+            width: 32px;
+            fill: white;
+        }
+
+        /* App Layout */
+        .app-content-wrapper {
+            display: flex;
+            height: calc(100vh - 50px);
+            overflow: hidden;
+        }
+
+        /* Side Navigation */
+        #app-navigation {
+            width: 250px;
+            background: var(--color-main-background);
+            border-right: 1px solid var(--color-border);
+            display: flex;
+            flex-direction: column;
+            flex-shrink: 0;
+            transition: margin-left 0.3s ease;
+        }
+
+        #app-navigation.app-navigation--closed {
+            margin-left: -250px;
+        }
+
+        .app-navigation__content {
+            flex: 1;
+            overflow-y: auto;
+            padding: 8px;
+            display: flex;
+            flex-direction: column;
+        }
+
+        .app-navigation-list {
+            list-style: none;
+            padding: 0;
+            margin: 0;
+            flex: 1;
+        }
+
+        .app-navigation-entry {
+            position: relative;
+            margin-bottom: 2px;
+        }
+
+        .app-navigation-entry__wrapper {
+            display: flex;
+            align-items: center;
+            position: relative;
+        }
+
+        .app-navigation-entry-link {
+            display: flex;
+            align-items: center;
+            padding: 0 8px;
+            min-height: var(--default-clickable-area);
+            border-radius: var(--border-radius);
+            transition: background-color 100ms ease-in-out;
+            text-decoration: none;
+            color: var(--color-main-text);
+            flex: 1;
+            font-size: 14px;
+        }
+
+        .app-navigation-entry-link:hover {
+            background-color: var(--color-background-hover);
+        }
+
+        .app-navigation-entry.active .app-navigation-entry-link {
+            background-color: var(--color-primary-element-light);
+            font-weight: 500;
+        }
+
+        .app-navigation-entry-icon {
+            width: var(--default-clickable-area);
+            height: var(--default-clickable-area);
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            margin-right: 0;
+        }
+
+        .app-navigation-entry__name {
+            flex: 1;
+            white-space: nowrap;
+            overflow: hidden;
+            text-overflow: ellipsis;
+        }
+
+        .app-navigation-entry__counter {
+            margin-left: auto;
+            padding: 2px 6px;
+            border-radius: var(--border-radius-pill);
+            background-color: var(--color-background-dark);
+            font-size: 11px;
+            color: var(--color-text-maxcontrast);
+            min-width: 20px;
+            text-align: center;
+        }
+
+        .app-navigation__settings {
+            list-style: none;
+            padding: 8px 0 0 0;
+            margin: 8px 0 0 0;
+            border-top: 1px solid var(--color-border);
+            flex-shrink: 0;
+        }
+
+        .app-navigation-toggle {
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            position: fixed;
+            top: 60px;
+            left: 10px;
+            z-index: 110;
+            background: var(--color-main-background);
+            border: 1px solid var(--color-border);
+            border-radius: var(--border-radius);
+            padding: 8px 12px;
+            cursor: pointer;
+            box-shadow: 0 0 5px rgba(0,0,0,0.1);
+            transition: left 0.3s ease;
+        }
+
+        .app-navigation-toggle:hover {
+            background: var(--color-background-hover);
+        }
+
+        #app-navigation:not(.app-navigation--closed) ~ * .app-navigation-toggle {
+            left: 260px;
+        }
+
+        /* Main Content Area */
+        #app-content {
+            flex: 1;
+            overflow-y: auto;
+            background: var(--color-main-background);
+        }
+
+        .page-content {
+            max-width: 1000px;
+            margin: 0 auto;
+            padding: 24px;
+        }
+
+        .content-section {
+            background: var(--color-main-background);
+            border-radius: 0;
+            padding: 0;
+            box-shadow: none;
+        }
+
+        .content-section h1 {
+            font-size: 24px;
+            font-weight: 600;
+            margin-bottom: 24px;
+        }
+
+        .content-section h2 {
+            font-size: 18px;
+            font-weight: 500;
+            margin: 24px 0 12px 0;
+            border-bottom: none;
+            padding-bottom: 0;
+        }
+
+        .content-section h3 {
+            font-size: 16px;
+            font-weight: 500;
+        }
+
+        /* Responsive */
+        @media (max-width: 768px) {
+            #app-navigation {
+                position: fixed;
+                height: calc(100vh - 50px);
+                z-index: 105;
+                box-shadow: 2px 0 8px rgba(0,0,0,0.1);
+            }
+
+            .page-content {
+                padding: 16px;
+            }
+        }
+
+        /* Footer */
+        footer.page-footer {
+            background-color: #0F0833;
+            color: #ffffff;
+            padding: 40px 0;
+            margin-top: 60px;
+        }
+
+        footer.page-footer .bootstrap-container {
+            max-width: 1200px;
+            margin: 0 auto;
+            padding: 0 20px;
+        }
+
+        footer.page-footer h1 {
+            font-size: 15px;
+            font-weight: bold;
+            line-height: 1.8;
+            color: #ffffff;
+            margin-top: 20px;
+        }
+
+        footer.page-footer ul {
+            list-style-type: none;
+            padding-left: 0;
+        }
+
+        footer.page-footer li {
+            font-size: 13px;
+            line-height: 1.8;
+            color: #ffffff;
+            margin-top: 0;
+        }
+
+        footer.page-footer li a {
+            color: #ffffff;
+            text-decoration: none;
+            display: block;
+            padding: 4px 0;
+        }
+
+        footer.page-footer li a:hover {
+            text-decoration: underline;
+        }
+
+        footer.page-footer p {
+            font-size: 15px;
+            line-height: 1.8;
+            color: #ffffff;
+        }
+
+        footer.page-footer p.copyright {
+            color: rgba(255, 255, 255, 0.5);
+            font-size: 13px;
+            text-align: center;
+            margin-top: 30px;
+        }
+
+        /* Buttons */
+        .btn {
+            border-radius: 50px;
+            padding: 10px 20px;
+            text-decoration: none;
+            display: inline-block;
+            cursor: pointer;
+            border: none;
+            font-size: 14px;
+            transition: all 0.3s;
+        }
+
+        .btn-primary {
+            background: #0082C9;
+            border: 1px solid #0062C9;
+            color: #fff;
+        }
+
+        .btn-primary:hover {
+            background: #006ba3;
+        }
+
+        /* Tables */
+        table {
+            width: 100%;
+            border-collapse: collapse;
+            margin: 20px 0;
+        }
+
+        td {
+            padding: 12px 8px;
+            border-bottom: 1px solid var(--color-border);
+            font-size: 14px;
+        }
+
+        td:first-child {
+            width: 180px;
+            color: var(--color-text-maxcontrast);
+            font-weight: 500;
+        }
+
+        code {
+            background-color: var(--color-background-dark);
+            padding: 2px 6px;
+            border-radius: var(--border-radius);
+            font-family: 'SFMono-Regular', 'Consolas', 'Liberation Mono', 'Menlo', monospace;
+            font-size: 90%;
+            color: var(--color-main-text);
+        }
+
+        /* Badges */
+        .badge {
+            display: inline-block;
+            padding: 3px 8px;
+            border-radius: 12px;
+            font-size: 12px;
+            font-weight: bold;
+            text-transform: uppercase;
+        }
+
+        .badge-oauth {
+            background-color: #4caf50;
+            color: white;
+        }
+
+        .badge-basic {
+            background-color: #2196f3;
+            color: white;
+        }
+
+        /* Messages */
+        .warning {
+            background-color: #fff3cd;
+            border-left: 4px solid #ffc107;
+            padding: 15px;
+            margin: 15px 0;
+            color: #856404;
+        }
+
+        .info-message {
+            background-color: #e3f2fd;
+            border-left: 4px solid #2196f3;
+            padding: 15px;
+            margin: 15px 0;
+            color: #1565c0;
+        }
+
+        .error {
+            background-color: #ffebee;
+            border-left: 4px solid #d32f2f;
+            padding: 15px;
+            margin: 15px 0;
+            color: #c62828;
+        }
+
+        .success {
+            background-color: #e8f5e9;
+            border: 2px solid #4caf50;
+            padding: 30px;
+            border-radius: 8px;
+            text-align: center;
+        }
+
+        .success h1 {
+            color: #4caf50;
+        }
+
+        {% block extra_styles %}{% endblock %}
+    </style>
+</head>
+<body>
+    <!-- App Header -->
+    <header class="app-header">
+        <a href="/app" class="app-header__brand">
+            <svg class="app-header__logo" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512">
+                <path d="M255.9 21.04c-11.8 0-22.2 4.08-28.6 10.01-5.6 4.98-8.6 11.41-8.6 18.11 0 5.55 2.2 11.01 5.9 15.48-16.4 4.97-30.1 13.64-39 24.53 22.1-7.67 45.7-11.86 70.3-11.86 24.6 0 48.3 4.19 70.3 11.86-8.9-10.89-22.6-19.56-39-24.53 3.9-4.47 5.9-9.93 5.9-15.48 0-6.7-3-13.13-8.5-18.11-6.4-5.93-16.9-10.01-28.7-10.01zm0 20.34c5.3 0 10.1 1.27 13.6 3.52 1.7 1.16 3.4 2.43 3.4 4.27 0 1.76-1.7 3.03-3.4 4.19-3.5 2.33-8.3 3.61-13.6 3.61-5.3 0-10.1-1.28-13.6-3.61-1.6-1.16-3.3-2.43-3.3-4.19 0-1.84 1.7-3.11 3.3-4.27 3.5-2.25 8.3-3.52 13.6-3.52zm.1 48.1c-110.8 0-200.72 90.02-200.72 200.82S145.2 491 256 491s200.7-89.9 200.7-200.7c0-110.8-89.9-200.82-200.7-200.82zm0 32.62c92.9 0 168.2 75.3 168.2 168.2 0 92.8-75.3 168.2-168.2 168.2-92.9 0-168.26-75.4-168.26-168.2 0-92.9 75.36-168.2 168.26-168.2zm-8.2 6.3c-9.6.5-19 1.9-28.3 4.1l2.3 7.8c8.4-2 17.1-3.3 26-3.8v-8.1zm16.2 0v8.1c9 .5 17.7 1.8 26 3.8l2.2-7.8c-9.1-2.2-18.6-3.6-28.2-4.1zm-60 8.5c-9 3.2-17.6 7-25.8 11.6l4.1 7.1c7.7-4.3 15.6-7.9 23.9-10.8l-2.2-7.9zm103.7 0-2 7.9c8.4 2.9 16.2 6.5 23.8 10.8l4.2-7.1c-8.2-4.6-16.9-8.4-26-11.6zm-143.3 20.3c-7.5 5.4-14.6 11.4-21.1 17.9l5.8 5.8c5.9-6.1 12.5-11.7 19.5-16.6l-4.2-7.1zm182.9 0-4 7.1c6.9 4.9 13.5 10.5 19.5 16.6l5.7-5.8c-6.5-6.5-13.7-12.5-21.2-17.9zm-91.4 11.5c-37 0-67.4 28.6-70.3 64.9l15.9 4.7c.7-29.6 24.7-53.4 54.4-53.4 30.1 0 54.4 24.4 54.4 54.3 0 15-6.2 28.7-16 38.5l.1.1c1.7 2.7 3 5.6 4.1 8.6.9 3 1.7 5.7 2.3 8.6v.4c33.8-16.7 57.2-51.5 57.2-91.7 0-3.8-.2-7.3-.6-10.9-3.2-3.3-6.3-6.4-9.8-9.5 1.5 6.5 2.3 13.4 2.3 20.4 0 28.7-13 54.7-33.5 71.8 6.3-10.6 10.1-23 10.1-36.3 0-38.9-31.7-70.5-70.6-70.5zm-91.8 14.6c-3.3 3.1-6.5 6.2-9.7 9.5-.3 3.6-.5 7.1-.5 10.9 0 7.3.7 14.2 2.1 20.9l9.1 2.7c-2.1-7.5-3.1-15.4-3.1-23.6 0-7 .7-13.9 2.1-20.4zm-31.6 4c-5.8 7.1-10.9 14.6-15.4 22.6l7.1 4c4.1-7.4 8.8-14.3 14-20.8l-5.7-5.8zm246.8 0-5.7 5.8c5.3 6.5 10 13.4 13.9 20.8l7.1-4c-4.4-8-9.5-15.5-15.3-22.6zm-269.2 37.1c-2.5 5.7-4.6 11.4-6.4 17.6l.1-.3c3.4-5 7.9-9.3 12.9-12.5l.3-.6-6.9-4.2zm291.8 0-7.2 4.2c3.2 7.3 5.7 15.1 7.6 23.1l7.9-2.1c-2.1-8.8-4.9-17.3-8.3-25.2zm-261.2 11.5c-13.4.1-25.7 9-29.7 22.5l114.8 34.2c-4.9 16.7 4.6 34.2 21.2 39.2L361.7 366c16.6 5 34.1-4.4 39.1-21l-114.6-34.4c4.9-16.5-4.7-34.1-21.3-39.1 0 0-72.4-21.5-114.8-34.3-3.1-.9-6.3-1.4-9.4-1.3zm-42.09 29.7c-.9 6.9-1.4 14-1.4 21.3 0 1.3.1 2.9.1 4.2h8.09v-4.2c0-6.5.4-12.9 1.2-19.2l-7.99-2.1zm314.59 0-7.9 2.1c.7 6.3 1.3 12.7 1.3 19.2 0 1.3 0 2.9-.2 4.2h8.2v-4.2c0-7.3-.5-14.4-1.4-21.3zm-157.3 24.7c6.3 0 11.5 5 11.5 11.3 0 6.4-5.2 11.6-11.5 11.6s-11.5-5.2-11.5-11.6c0-6.3 5.2-11.3 11.5-11.3zM98.51 307.4c1 8.2 2.89 16.4 5.09 24.3l7.9-2.1c-2.1-7.2-3.8-14.6-4.8-22.2h-8.19zm306.69 0c-1.1 7.6-2.7 15-4.8 22.2l7.8 2.1c2.2-7.9 4.1-16.1 5.2-24.3h-8.2zm-191.3 10.9c-19 13.3-31.4 35.3-31.4 60.1 0 10.4 2.3 20.4 6.2 29.7 8.8 4.9 17.9 8.8 27.6 11.7-10.8-10.7-17.5-25.2-17.5-41.4 0-19 9.3-36 23.7-46.3-3.8-4.1-6.7-8.7-8.6-13.8zM116.8 345l-7.9 2c3.1 7.6 6.8 14.7 11 21.6l6.9-4.2c-3.8-6.2-7-12.8-10-19.4zm194.8 20.5c.9 4.1 1.4 8.5 1.4 12.9 0 16.2-6.7 30.7-17.4 41.4 9.6-2.9 18.8-6.8 27.5-11.7 4-9.3 6.2-19.3 6.2-29.7 0-2.7-.2-5.2-.4-7.7l-17.3-5.2zM136 377.9l-7.1 4.1c4.7 6.2 9.7 12.1 15.3 17.3l5.7-5.5c-5.1-5-9.7-10.3-13.9-15.9zm243.9 2.3-.2.1c-2.1.3-4 .6-6.2.7h-.1c-3.6 4.5-7.3 8.8-11.5 12.8l5.8 5.5c5.5-5.2 10.5-11.1 15.2-17.3l-3-1.8zm-217.8 24-5.9 5.9c6 4.8 12.2 9.7 18.8 13.6l3.8-7.8c-5.7-2.9-11.4-6.8-16.7-11.7zm187.7 0c-5.4 4.9-11.1 8.8-16.8 11.7l3.9 7.8c6.5-3.9 12.8-8.8 18.7-13.6l-5.8-5.9zm-156.4 19.5-4.1 6.8c6.6 4 13.7 5.8 20.7 8.8l2.2-7.9c-6.5-1.9-12.7-4.8-18.8-7.7zm125.2 0c-6.2 2.9-12.5 5.8-19.1 7.7l2.3 7.9c7.2-3 14-4.8 20.7-8.8l-3.9-6.8zm-90.7 11.7-2 7.8c7.1 1 14.5 1.9 21.9 1.9v-7.7c-6.8 0-13.5-1.1-19.9-2zm55.9 0c-6.3.9-13 2-19.8 2v7.7c7.5 0 14.8-.9 22.1-1.9l-2.3-7.8z" fill="#fff"/>
+            </svg>
+            <span>Nextcloud MCP Server</span>
+        </a>
+    </header>
+
+    <!-- App Content Wrapper (Sidebar + Main Content) -->
+    {% block content %}{% endblock %}
+
+    {% block scripts %}{% endblock %}
+</body>
+</html>

nextcloud_mcp_server/auth/templates/error.html

@@ -0,0 +1,19 @@
+{% extends "base.html" %}
+
+{% block title %}{{ error_title|default('Error') }} - Nextcloud MCP Server{% endblock %}
+
+{% block content %}
+<h1>{{ error_title|default('Error') }}</h1>
+
+<div class="error">
+    <strong>Error:</strong> {{ error_message }}
+</div>
+
+{% if login_url %}
+<p><a href="{{ login_url }}" class="btn btn-primary">Login again</a></p>
+{% endif %}
+
+{% if back_url %}
+<p><a href="{{ back_url }}" class="btn">Go Back</a></p>
+{% endif %}
+{% endblock %}

nextcloud_mcp_server/auth/templates/success.html

@@ -0,0 +1,21 @@
+{% extends "base.html" %}
+
+{% block title %}{{ success_title|default('Success') }} - Nextcloud MCP Server{% endblock %}
+
+{% block extra_head %}
+{% if redirect_url and redirect_delay %}
+<meta http-equiv="refresh" content="{{ redirect_delay }};url={{ redirect_url }}">
+{% endif %}
+{% endblock %}
+
+{% block content %}
+<div class="success">
+    <h1>{{ success_title|default('✓ Success') }}</h1>
+    {% for message in success_messages %}
+    <p>{{ message }}</p>
+    {% endfor %}
+    {% if redirect_url %}
+    <p>Redirecting...</p>
+    {% endif %}
+</div>
+{% endblock %}

nextcloud_mcp_server/auth/templates/user_info.html

@@ -0,0 +1,650 @@
+{% extends "base.html" %}
+
+{% block title %}Nextcloud MCP Server{% endblock %}
+
+{% block extra_head %}
+    <!-- htmx for dynamic loading -->
+    <script src="https://unpkg.com/htmx.org@1.9.10"></script>
+
+    <!-- Alpine.js for state management -->
+    <script defer src="https://cdn.jsdelivr.net/npm/alpinejs@3.x.x/dist/cdn.min.js"></script>
+
+    <!-- Plotly.js for vector visualization -->
+    <script src="https://cdn.plot.ly/plotly-3.3.0.min.js"></script>
+
+    <!-- Vector Viz static assets -->
+    <link rel="stylesheet" href="/app/static/vector-viz.css">
+{% endblock %}
+
+{% block extra_styles %}
+    /* Smooth htmx transitions */
+    .htmx-swapping {
+        opacity: 0;
+        transition: opacity 200ms ease-out;
+    }
+
+    .htmx-settling {
+        opacity: 1;
+        transition: opacity 200ms ease-in;
+    }
+
+    /* Logout button styling */
+    .logout-section {
+        margin-top: 20px;
+        padding-top: 20px;
+        border-top: 1px solid var(--color-border);
+    }
+
+    /* Welcome tab specific styles */
+    .hero-section {
+        background: linear-gradient(135deg, var(--color-primary-element) 0%, #0082c9 100%);
+        color: white;
+        padding: 60px 24px;
+        margin: -24px -24px 40px -24px;
+        border-radius: 0 0 var(--border-radius-large) var(--border-radius-large);
+        text-align: center;
+    }
+
+    .hero-section h1 {
+        color: white;
+        font-size: 36px;
+        margin: 0 0 16px 0;
+        font-weight: 600;
+    }
+
+    .hero-section p {
+        font-size: 18px;
+        opacity: 0.95;
+        max-width: 700px;
+        margin: 0 auto;
+        line-height: 1.6;
+    }
+
+    .feature-grid {
+        display: grid;
+        grid-template-columns: repeat(auto-fit, minmax(280px, 1fr));
+        gap: 24px;
+        margin: 32px 0;
+    }
+
+    .feature-card {
+        background: var(--color-main-background);
+        border: 2px solid var(--color-border);
+        border-radius: var(--border-radius-large);
+        padding: 24px;
+        transition: all 0.2s;
+        cursor: pointer;
+        text-decoration: none;
+        color: inherit;
+        display: block;
+    }
+
+    .feature-card:hover {
+        border-color: var(--color-primary-element);
+        box-shadow: 0 4px 12px rgba(0, 103, 158, 0.15);
+        transform: translateY(-2px);
+    }
+
+    .feature-card h3 {
+        color: var(--color-primary-element);
+        font-size: 20px;
+        margin: 12px 0 8px 0;
+        font-weight: 600;
+        display: flex;
+        align-items: center;
+        gap: 12px;
+    }
+
+    .feature-card p {
+        color: var(--color-text-maxcontrast);
+        font-size: 14px;
+        line-height: 1.6;
+        margin: 8px 0 0 0;
+    }
+
+    .feature-icon {
+        width: 48px;
+        height: 48px;
+        background: var(--color-primary-element-light);
+        border-radius: var(--border-radius);
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        margin-bottom: 8px;
+    }
+
+    .feature-icon svg {
+        width: 28px;
+        height: 28px;
+        fill: var(--color-primary-element);
+    }
+
+    .info-section {
+        background: var(--color-background-hover);
+        border-radius: var(--border-radius-large);
+        padding: 32px;
+        margin: 32px 0;
+    }
+
+    .info-section h2 {
+        color: var(--color-main-text);
+        font-size: 24px;
+        margin: 0 0 16px 0;
+        border: none;
+        padding: 0;
+    }
+
+    .info-section p {
+        color: var(--color-text-maxcontrast);
+        line-height: 1.7;
+        margin: 12px 0;
+    }
+
+    .info-section ul {
+        margin: 12px 0;
+        padding-left: 24px;
+    }
+
+    .info-section li {
+        color: var(--color-text-maxcontrast);
+        line-height: 1.7;
+        margin: 8px 0;
+    }
+
+    .info-section code {
+        background: var(--color-main-background);
+        padding: 2px 8px;
+        border-radius: var(--border-radius);
+        font-size: 13px;
+    }
+
+    .auth-status {
+        background: var(--color-primary-element-light);
+        border-left: 4px solid var(--color-primary-element);
+        padding: 16px 20px;
+        margin: 24px 0;
+        border-radius: var(--border-radius);
+        display: flex;
+        align-items: center;
+        gap: 12px;
+    }
+
+    .auth-status svg {
+        width: 24px;
+        height: 24px;
+        fill: var(--color-primary-element);
+        flex-shrink: 0;
+    }
+
+    .auth-status-text {
+        flex: 1;
+    }
+
+    .auth-status-text strong {
+        display: block;
+        color: var(--color-main-text);
+        font-size: 14px;
+        margin-bottom: 4px;
+    }
+
+    .auth-status-text span {
+        color: var(--color-text-maxcontrast);
+        font-size: 13px;
+    }
+{% endblock %}
+
+{% block content %}
+<div class="app-content-wrapper" x-data="{ activeSection: 'welcome', navOpen: true }">
+    <!-- Side Navigation -->
+    <nav id="app-navigation" :class="{ 'app-navigation--closed': !navOpen }">
+        <div class="app-navigation__content">
+            <!-- Navigation List -->
+            <ul class="app-navigation-list">
+                <li class="app-navigation-entry" :class="{ 'active': activeSection === 'welcome' }">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="#"
+                           @click.prevent="activeSection = 'welcome'"
+                           class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M10,20V14H14V20H19V12H22L12,3L2,12H5V20H10Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">Welcome</span>
+                        </a>
+                    </div>
+                </li>
+
+                <li class="app-navigation-entry" :class="{ 'active': activeSection === 'user-info' }">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="#"
+                           @click.prevent="activeSection = 'user-info'"
+                           class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">User Info</span>
+                        </a>
+                    </div>
+                </li>
+
+                {% if show_vector_sync_tab %}
+                <li class="app-navigation-entry" :class="{ 'active': activeSection === 'vector-sync' }">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="#"
+                           @click.prevent="activeSection = 'vector-sync'"
+                           class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M12,18A6,6 0 0,1 6,12C6,11 6.25,10.03 6.7,9.2L5.24,7.74C4.46,8.97 4,10.43 4,12A8,8 0 0,0 12,20V23L16,19L12,15M12,4V1L8,5L12,9V6A6,6 0 0,1 18,12C18,13 17.75,13.97 17.3,14.8L18.76,16.26C19.54,15.03 20,13.57 20,12A8,8 0 0,0 12,4Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">Vector Sync</span>
+                        </a>
+                    </div>
+                </li>
+
+                <li class="app-navigation-entry" :class="{ 'active': activeSection === 'vector-viz' }">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="#"
+                           @click.prevent="activeSection = 'vector-viz'"
+                           class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M22,21H2V3H4V19H6V10H10V19H12V6H16V19H18V14H22V21Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">Vector Viz</span>
+                        </a>
+                    </div>
+                </li>
+                {% endif %}
+
+                {% if show_webhooks_tab %}
+                <li class="app-navigation-entry" :class="{ 'active': activeSection === 'webhooks' }">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="#"
+                           @click.prevent="activeSection = 'webhooks'"
+                           class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M10.59,13.41C11,13.8 11,14.44 10.59,14.83C10.2,15.22 9.56,15.22 9.17,14.83C7.22,12.88 7.22,9.71 9.17,7.76V7.76L12.71,4.22C14.66,2.27 17.83,2.27 19.78,4.22C21.73,6.17 21.73,9.34 19.78,11.29L18.29,12.78C18.3,11.96 18.17,11.14 17.89,10.36L18.36,9.88C19.54,8.71 19.54,6.81 18.36,5.64C17.19,4.46 15.29,4.46 14.12,5.64L10.59,9.17C9.41,10.34 9.41,12.24 10.59,13.41M13.41,9.17C13.8,8.78 14.44,8.78 14.83,9.17C16.78,11.12 16.78,14.29 14.83,16.24V16.24L11.29,19.78C9.34,21.73 6.17,21.73 4.22,19.78C2.27,17.83 2.27,14.66 4.22,12.71L5.71,11.22C5.7,12.04 5.83,12.86 6.11,13.65L5.64,14.12C4.46,15.29 4.46,17.19 5.64,18.36C6.81,19.54 8.71,19.54 9.88,18.36L13.41,14.83C14.59,13.66 14.59,11.76 13.41,10.59C13,10.2 13,9.56 13.41,9.17Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">Webhooks</span>
+                        </a>
+                    </div>
+                </li>
+                {% endif %}
+            </ul>
+
+            <!-- Settings/Logout at bottom -->
+            {% if logout_url %}
+            <ul class="app-navigation__settings">
+                <li class="app-navigation-entry">
+                    <div class="app-navigation-entry__wrapper">
+                        <a href="{{ logout_url }}" class="app-navigation-entry-link">
+                            <span class="app-navigation-entry-icon">
+                                <svg class="nav-icon" viewBox="0 0 24 24">
+                                    <path d="M16,17V14H9V10H16V7L21,12L16,17M14,2A2,2 0 0,1 16,4V6H14V4H5V20H14V18H16V20A2,2 0 0,1 14,22H5A2,2 0 0,1 3,20V4A2,2 0 0,1 5,2H14Z" />
+                                </svg>
+                            </span>
+                            <span class="app-navigation-entry__name">Logout</span>
+                        </a>
+                    </div>
+                </li>
+            </ul>
+            {% endif %}
+        </div>
+
+        <!-- Toggle Button (mobile) -->
+        <button @click="navOpen = !navOpen"
+                class="app-navigation-toggle"
+                :aria-expanded="navOpen.toString()">
+            ☰
+        </button>
+    </nav>
+
+    <!-- Main Content Area -->
+    <main id="app-content">
+        <div class="page-content">
+            <!-- Welcome Section -->
+            <div x-show="activeSection === 'welcome'">
+                <!-- Hero Section -->
+                <div class="hero-section">
+                    <h1>Welcome to Nextcloud MCP Server</h1>
+                    <p>
+                        Interactive user interface for semantic search and document retrieval.
+                        Test queries, visualize results, and explore your Nextcloud content using RAG workflows.
+                    </p>
+                </div>
+
+                <!-- Authentication Status -->
+                <div class="auth-status">
+                    <svg viewBox="0 0 24 24">
+                        <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                    </svg>
+                    <div class="auth-status-text">
+                        <strong>Authenticated as: {{ username }}</strong>
+                        <span>Authentication mode: <code>{{ auth_mode }}</code></span>
+                    </div>
+                </div>
+
+                {% if vector_sync_enabled %}
+                <!-- Vector Sync Enabled Content -->
+                <div class="info-section">
+                    <h2>About Semantic Search</h2>
+                    <p>
+                        This interface provides access to <strong>semantic search</strong> capabilities powered by vector embeddings.
+                        Unlike traditional keyword search, semantic search understands the <em>meaning</em> of your queries and finds
+                        conceptually similar content across your Nextcloud apps.
+                    </p>
+                    <p>
+                        <strong>How it works:</strong>
+                    </p>
+                    <ul>
+                        <li>Documents from Notes, Calendar, Files, Contacts, and Deck are indexed into a vector database</li>
+                        <li>Each document chunk is converted to a 768-dimensional vector embedding that captures semantic meaning</li>
+                        <li>Queries are also converted to embeddings and matched against document vectors using similarity search</li>
+                        <li>Results can be retrieved using pure semantic search or hybrid BM25 search combining keywords and semantics</li>
+                    </ul>
+                </div>
+
+                <div class="info-section">
+                    <h2>RAG Workflow Integration</h2>
+                    <p>
+                        This UI allows you to <strong>test the same queries that Large Language Models (LLMs) would use</strong> in a
+                        Retrieval-Augmented Generation (RAG) workflow. When an AI assistant needs to answer questions about your data:
+                    </p>
+                    <ul>
+                        <li><strong>Step 1:</strong> The assistant converts your question into a search query</li>
+                        <li><strong>Step 2:</strong> The MCP server retrieves relevant document chunks using semantic search</li>
+                        <li><strong>Step 3:</strong> Retrieved context is passed to the LLM to generate an informed answer</li>
+                    </ul>
+
+                    <!-- RAG Workflow Diagram -->
+                    <div style="background: var(--color-main-background); border: 2px solid var(--color-primary-element); border-radius: var(--border-radius-large); padding: 24px; margin: 24px 0; overflow-x: auto;">
+                        <div style="text-align: center; font-weight: 600; margin-bottom: 20px; color: var(--color-primary-element); font-size: 16px;">
+                            MCP Sampling RAG Workflow
+                        </div>
+
+                        <!-- Four-component bidirectional flow -->
+                        <div style="max-width: 1000px; margin: 0 auto;">
+                            <div style="display: grid; grid-template-columns: 0.7fr auto 1fr auto 1fr auto 0.9fr; gap: 10px; align-items: center;">
+                                <!-- User -->
+                                <div style="background: var(--color-background-hover); border: 2px solid var(--color-border); border-radius: var(--border-radius-large); padding: 14px; text-align: center;">
+                                    <div style="font-size: 26px; margin-bottom: 5px;">👤</div>
+                                    <div style="font-weight: 600; color: var(--color-main-text); font-size: 12px;">User</div>
+                                    <div style="font-size: 9px; color: var(--color-text-maxcontrast); font-style: italic; margin-top: 5px; line-height: 1.2;">
+                                        "What are health<br>benefits of coffee?"
+                                    </div>
+                                </div>
+
+                                <!-- Arrow User <-> Client -->
+                                <div style="text-align: center;">
+                                    <div style="font-size: 20px; color: var(--color-text-maxcontrast);">↔</div>
+                                </div>
+
+                                <!-- MCP Client + LLM (combined) -->
+                                <div style="background: var(--color-primary-element-light); border: 2px solid var(--color-primary-element); border-radius: var(--border-radius-large); padding: 12px; text-align: center;">
+                                    <div style="font-weight: 600; color: var(--color-primary-element); font-size: 13px; margin-bottom: 8px;">MCP Client + LLM</div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 8px; margin-bottom: 6px;">
+                                        <div style="font-size: 9px; color: var(--color-text-maxcontrast);">(Claude Code)</div>
+                                    </div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 8px; border: 2px solid var(--color-primary-element);">
+                                        <div style="font-size: 16px; margin-bottom: 2px;">🧠</div>
+                                        <div style="font-weight: 600; color: var(--color-main-text); font-size: 10px;">Client's LLM</div>
+                                        <div style="font-size: 8px; color: var(--color-text-maxcontrast);">(Claude)</div>
+                                    </div>
+
+                                    <div style="margin-top: 8px; font-size: 8px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                        <strong>Enables RAG:</strong><br>
+                                        Receives context,<br>
+                                        generates answer
+                                    </div>
+                                </div>
+
+                                <!-- Arrow Client <-> Server -->
+                                <div style="text-align: center;">
+                                    <div style="font-size: 20px; color: var(--color-primary-element);">↔</div>
+                                    <div style="font-size: 7px; color: var(--color-text-maxcontrast); margin-top: 2px; font-weight: 600; line-height: 1.1;">
+                                        Query +<br>
+                                        Sampling
+                                    </div>
+                                </div>
+
+                                <!-- MCP Server -->
+                                <div style="background: var(--color-primary-element-light); border: 2px solid var(--color-primary-element); border-radius: var(--border-radius-large); padding: 12px; text-align: center;">
+                                    <div style="font-weight: 600; color: var(--color-primary-element); font-size: 13px; margin-bottom: 8px;">MCP Server</div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 7px; margin-bottom: 5px;">
+                                        <div style="font-weight: 600; color: var(--color-main-text); font-size: 9px; margin-bottom: 2px;">1. Semantic Search</div>
+                                        <div style="font-size: 7px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                            Vector embeddings<br>
+                                            BM25 Hybrid + RRF
+                                        </div>
+                                    </div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 7px; margin-bottom: 5px;">
+                                        <div style="font-weight: 600; color: var(--color-main-text); font-size: 9px; margin-bottom: 2px;">2. Retrieve Context</div>
+                                        <div style="font-size: 7px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                            Top relevant docs<br>
+                                            with scores
+                                        </div>
+                                    </div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 7px; margin-bottom: 5px;">
+                                        <div style="font-weight: 600; color: var(--color-main-text); font-size: 9px; margin-bottom: 2px;">3. Format Response</div>
+                                        <div style="font-size: 7px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                            Document chunks<br>
+                                            with citations
+                                        </div>
+                                    </div>
+
+                                    <div style="background: var(--color-main-background); border-radius: var(--border-radius); padding: 7px;">
+                                        <div style="font-weight: 600; color: var(--color-main-text); font-size: 9px; margin-bottom: 2px;">4. Send to LLM</div>
+                                        <div style="font-size: 7px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                            Via MCP sampling<br>
+                                            for answer generation
+                                        </div>
+                                    </div>
+                                </div>
+
+                                <!-- Arrow Server <-> Nextcloud -->
+                                <div style="text-align: center;">
+                                    <div style="font-size: 20px; color: var(--color-primary-element);">↔</div>
+                                    <div style="font-size: 7px; color: var(--color-text-maxcontrast); margin-top: 2px; font-weight: 600; line-height: 1.1;">
+                                        Retrieve
+                                    </div>
+                                </div>
+
+                                <!-- Nextcloud -->
+                                <div style="background: var(--color-background-hover); border: 2px solid var(--color-border); border-radius: var(--border-radius-large); padding: 12px; text-align: center; position: relative;">
+                                    <img src="/app/static/nextcloud-logo.png" alt="Nextcloud" style="width: 40px; height: 40px; margin-bottom: 6px;" />
+                                    <div style="font-weight: 600; color: var(--color-main-text); font-size: 12px; margin-bottom: 4px;">Nextcloud</div>
+                                    <div style="font-size: 8px; color: var(--color-text-maxcontrast); line-height: 1.2;">
+                                        Notes, Calendar,<br>
+                                        Files, Contacts,<br>
+                                        Deck
+                                    </div>
+                                </div>
+                            </div>
+
+                            <!-- Explanation below diagram -->
+                            <div style="margin-top: 24px; padding: 16px; background: var(--color-background-hover); border-radius: var(--border-radius); border-left: 4px solid var(--color-primary-element);">
+                                <div style="font-size: 12px; color: var(--color-main-text); line-height: 1.6;">
+                                    <strong>How RAG works via MCP Sampling:</strong>
+                                </div>
+                                <ol style="margin: 8px 0 0 0; padding-left: 20px; font-size: 11px; color: var(--color-text-maxcontrast); line-height: 1.6;">
+                                    <li>User asks question through MCP Client</li>
+                                    <li>Client sends query to MCP Server</li>
+                                    <li>Server retrieves relevant document context from Nextcloud</li>
+                                    <li><strong>Server sends context back to Client's LLM</strong> (MCP Sampling)</li>
+                                    <li>Client's LLM generates answer with citations using retrieved context</li>
+                                    <li>Answer returned to user</li>
+                                </ol>
+                                <div style="margin-top: 8px; font-size: 10px; color: var(--color-text-maxcontrast); font-style: italic;">
+                                    The server has no LLM - it only retrieves context. The client's existing LLM is reused for answer generation.
+                                </div>
+                            </div>
+                        </div>
+                    </div>
+
+                    <p style="margin-top: 16px;">
+                        <strong>Key Point:</strong> The MCP server retrieves context but doesn't generate answers itself.
+                        Through <strong>MCP sampling</strong>, it requests the client's LLM to generate responses, giving users
+                        full control over which model is used and ensuring all processing happens client-side.
+                    </p>
+
+                    <p>
+                        By using this interface, you can preview search results, understand relevance scores, and verify
+                        that the system retrieves the right information before it reaches the LLM.
+                    </p>
+                </div>
+
+                <!-- Feature Cards -->
+                <h2>Available Features</h2>
+                <div class="feature-grid">
+                    <a href="#" @click.prevent="activeSection = 'user-info'" class="feature-card">
+                        <div class="feature-icon">
+                            <svg viewBox="0 0 24 24">
+                                <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                            </svg>
+                        </div>
+                        <h3>User Information</h3>
+                        <p>
+                            View your authentication details, session information, and IdP profile.
+                            Manage background access permissions.
+                        </p>
+                    </a>
+
+                    <a href="#" @click.prevent="activeSection = 'vector-sync'" class="feature-card">
+                        <div class="feature-icon">
+                            <svg viewBox="0 0 24 24">
+                                <path d="M12,18A6,6 0 0,1 6,12C6,11 6.25,10.03 6.7,9.2L5.24,7.74C4.46,8.97 4,10.43 4,12A8,8 0 0,0 12,20V23L16,19L12,15M12,4V1L8,5L12,9V6A6,6 0 0,1 18,12C18,13 17.75,13.97 17.3,14.8L18.76,16.26C19.54,15.03 20,13.57 20,12A8,8 0 0,0 12,4Z" />
+                            </svg>
+                        </div>
+                        <h3>Vector Sync Status</h3>
+                        <p>
+                            Monitor real-time indexing progress with metrics for indexed documents, pending queue,
+                            and synchronization status.
+                        </p>
+                    </a>
+
+                    <a href="#" @click.prevent="activeSection = 'vector-viz'" class="feature-card">
+                        <div class="feature-icon">
+                            <svg viewBox="0 0 24 24">
+                                <path d="M22,21H2V3H4V19H6V10H10V19H12V6H16V19H18V14H22V21Z" />
+                            </svg>
+                        </div>
+                        <h3>Vector Visualization</h3>
+                        <p>
+                            Interactive search interface with 2D PCA visualization. Compare algorithms,
+                            view relevance scores, and explore matched document chunks.
+                        </p>
+                    </a>
+                </div>
+
+                {% else %}
+                <!-- Vector Sync Disabled Content -->
+                <div class="warning">
+                    <h3 style="margin-top: 0;">Vector Sync is Disabled</h3>
+                    <p>
+                        Semantic search and vector visualization features are currently disabled.
+                        To enable these features, set <code>VECTOR_SYNC_ENABLED=true</code> in your environment configuration.
+                    </p>
+                    <p style="margin-bottom: 0;">
+                        <strong>Learn more:</strong>
+                        <a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/configuration.md" target="_blank" style="color: inherit; text-decoration: underline;">
+                            Configuration Guide
+                        </a>
+                    </p>
+                </div>
+
+                <!-- Limited Feature Card -->
+                <h2>Available Features</h2>
+                <div class="feature-grid">
+                    <a href="#" @click.prevent="activeSection = 'user-info'" class="feature-card">
+                        <div class="feature-icon">
+                            <svg viewBox="0 0 24 24">
+                                <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                            </svg>
+                        </div>
+                        <h3>User Information</h3>
+                        <p>
+                            View your authentication details, session information, and IdP profile.
+                            Manage background access permissions.
+                        </p>
+                    </a>
+                </div>
+                {% endif %}
+
+                <!-- Documentation Section -->
+                <div class="info-section" style="margin-top: 40px;">
+                    <h2>Documentation</h2>
+                    <p>
+                        For detailed information about configuration, authentication modes, and advanced features,
+                        please refer to the project documentation:
+                    </p>
+                    <ul>
+                        <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/installation.md" target="_blank">Installation Guide</a></li>
+                        <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/configuration.md" target="_blank">Configuration Options</a></li>
+                        <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/authentication.md" target="_blank">Authentication Modes</a></li>
+                        {% if vector_sync_enabled %}
+                        <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/user-guide/vector-sync-ui.md" target="_blank">Vector Sync UI Guide</a></li>
+                        {% endif %}
+                    </ul>
+                </div>
+            </div>
+
+            <!-- User Info Section -->
+            <div x-show="activeSection === 'user-info'">
+                <div class="content-section">
+                    <h1>User Information</h1>
+                    {{ user_info_tab_html|safe }}
+                </div>
+            </div>
+
+            {% if show_vector_sync_tab %}
+            <!-- Vector Sync Section -->
+            <div x-show="activeSection === 'vector-sync'">
+                <div class="content-section">
+                    <h1>Vector Sync Status</h1>
+                    {{ vector_sync_tab_html|safe }}
+                </div>
+            </div>
+
+            <!-- Vector Viz Section -->
+            <div x-show="activeSection === 'vector-viz'">
+                <div class="content-section">
+                    <h1>Vector Visualization</h1>
+                    <div hx-get="/app/vector-viz" hx-trigger="load" hx-swap="outerHTML">
+                        <p style="color: #999;">Loading vector visualization...</p>
+                    </div>
+                </div>
+            </div>
+            {% endif %}
+
+            {% if show_webhooks_tab %}
+            <!-- Webhooks Section -->
+            <div x-show="activeSection === 'webhooks'">
+                <div class="content-section">
+                    <h1>Webhook Management</h1>
+                    {{ webhooks_tab_html|safe }}
+                </div>
+            </div>
+            {% endif %}
+        </div>
+    </main>
+</div>
+
+<script>
+    // Set global Nextcloud base URL for use in external JS
+    window.NEXTCLOUD_BASE_URL = '{{ nextcloud_host_for_links }}';
+</script>
+<script src="/app/static/vector-viz.js"></script>
+{% endblock %}

nextcloud_mcp_server/auth/templates/vector_viz.html

@@ -0,0 +1,180 @@
+<div x-data="vizApp()">
+    <div class="viz-layout">
+        <!-- Top: Search Controls -->
+        <div class="viz-card viz-controls-card">
+            <form @submit.prevent="executeSearch">
+                <div class="viz-controls-grid">
+                    <div class="viz-control-group">
+                        <label>Search Query</label>
+                        <input type="text" x-model="query" placeholder="Enter search query..." required />
+                    </div>
+
+                    <div class="viz-control-group">
+                        <label>Algorithm</label>
+                        <select x-model="algorithm">
+                            <option value="semantic">Semantic (Dense)</option>
+                            <option value="bm25_hybrid" selected>BM25 Hybrid</option>
+                        </select>
+                    </div>
+
+                    <div class="viz-control-group">
+                        <label>Fusion</label>
+                        <select x-model="fusion" :disabled="algorithm !== 'bm25_hybrid'" :style="algorithm !== 'bm25_hybrid' ? 'opacity: 0.5; cursor: not-allowed;' : ''">
+                            <option value="rrf" selected>RRF</option>
+                            <option value="dbsf">DBSF</option>
+                        </select>
+                    </div>
+
+                    <div class="viz-control-group">
+                        <label>&nbsp;</label>
+                        <button type="submit" class="viz-btn">Search</button>
+                    </div>
+
+                    <div class="viz-control-group">
+                        <label>&nbsp;</label>
+                        <button type="button" class="viz-btn-secondary" @click="showAdvanced = !showAdvanced">
+                            <span x-text="showAdvanced ? 'Hide' : 'Advanced'"></span>
+                        </button>
+                    </div>
+                </div>
+
+                <!-- Advanced Options (Collapsible) -->
+                <div x-show="showAdvanced" style="margin-top: 16px;">
+                    <div class="viz-controls-grid" style="grid-template-columns: repeat(auto-fit, minmax(150px, 1fr));">
+                        <div class="viz-control-group">
+                            <label>Document Types</label>
+                            <div style="display: grid; grid-template-columns: 1fr 1fr; gap: 8px; font-size: 13px;">
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="" style="margin-right: 4px;">
+                                    <span>All</span>
+                                </label>
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="note" style="margin-right: 4px;">
+                                    <span>Notes</span>
+                                </label>
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="file" style="margin-right: 4px;">
+                                    <span>Files</span>
+                                </label>
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="calendar" style="margin-right: 4px;">
+                                    <span>Calendar</span>
+                                </label>
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="contact" style="margin-right: 4px;">
+                                    <span>Contacts</span>
+                                </label>
+                                <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal;">
+                                    <input type="checkbox" x-model="docTypes" value="deck" style="margin-right: 4px;">
+                                    <span>Deck</span>
+                                </label>
+                            </div>
+                        </div>
+
+                        <div class="viz-control-group">
+                            <label>Score Threshold</label>
+                            <input type="number" x-model.number="scoreThreshold" min="0" max="1" step="any" />
+                        </div>
+
+                        <div class="viz-control-group">
+                            <label>Result Limit</label>
+                            <input type="number" x-model.number="limit" min="1" max="1000" />
+                        </div>
+
+                        <div class="viz-control-group">
+                            <label>Display Options</label>
+                            <label style="display: flex; align-items: center; cursor: pointer; font-weight: normal; margin-top: 4px;">
+                                <input type="checkbox" x-model="showQueryPoint" @change="updatePlot()" style="margin-right: 6px;">
+                                <span>Show Query Point</span>
+                            </label>
+                        </div>
+                    </div>
+                </div>
+            </form>
+        </div>
+
+        <!-- Plot -->
+        <div class="viz-card viz-card-plot">
+            <div id="viz-plot-container">
+                <div x-show="loading" class="viz-loading-overlay" x-transition.opacity.duration.200ms>
+                    Executing search and computing PCA projection...
+                </div>
+                <div id="viz-plot" x-show="!loading" x-transition.opacity.duration.200ms></div>
+            </div>
+        </div>
+
+        <!-- Results -->
+        <div class="viz-card" style="flex: 0 0 auto;">
+            <h3 style="margin-top: 0;">Search Results (<span x-text="loading ? '...' : results.length"></span>)</h3>
+
+        <div x-show="loading" class="viz-loading" x-transition.opacity.duration.200ms>
+            Loading results...
+        </div>
+
+        <div x-show="!loading && results.length === 0" class="viz-no-results" x-transition.opacity.duration.200ms>
+            No results found. Try a different query or adjust your search parameters.
+        </div>
+
+        <template x-if="!loading && results.length > 0">
+            <div x-transition.opacity.duration.200ms>
+                <template x-for="result in results" :key="`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`">
+                    <div style="padding: 12px; border-bottom: 1px solid #eee;">
+                        <a :href="getNextcloudUrl(result)" target="_blank" style="font-weight: 500; color: #0066cc; text-decoration: none;">
+                            <span x-text="result.title"></span>
+                        </a>
+                        <div style="font-size: 14px; color: #666; margin-top: 4px;"
+                             x-text="result.excerpt.length > 200 ? result.excerpt.substring(0, 200) + '...' : result.excerpt"></div>
+                        <div style="font-size: 12px; color: #999; margin-top: 4px;">
+                            Raw Score: <span x-text="result.original_score.toFixed(3)"></span>
+                            (<span x-text="(result.score * 100).toFixed(0)"></span>% relative) |
+                            Type: <span x-text="result.doc_type"></span>
+                        </div>
+
+                        <!-- Show Chunk button (only if chunk position is available) -->
+                        <template x-if="hasChunkPosition(result)">
+                            <button
+                                class="chunk-toggle-btn"
+                                @click="toggleChunk(result)"
+                                x-text="isChunkExpanded(`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`) ? 'Hide Chunk' : 'Show Chunk'"
+                            ></button>
+                        </template>
+
+                        <!-- Chunk context (expanded inline) -->
+                        <template x-if="isChunkExpanded(`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`)">
+                            <div class="chunk-context" x-transition.opacity.duration.200ms>
+                                <template x-if="chunkLoading[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]">
+                                    <div style="color: #666; font-style: italic;">Loading chunk...</div>
+                                </template>
+                                <template x-if="!chunkLoading[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]">
+                                    <div>
+                                        <!-- Highlighted page image for PDFs -->
+                                        <template x-if="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.highlighted_page_image">
+                                            <div class="chunk-image-container">
+                                                <div class="chunk-image-header">
+                                                    <span>Page <span x-text="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.page_number"></span></span>
+                                                </div>
+                                                <img
+                                                    :src="'data:image/png;base64,' + expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.highlighted_page_image"
+                                                    :alt="'Page ' + expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.page_number"
+                                                    class="chunk-highlighted-image"
+                                                />
+                                            </div>
+                                        </template>
+                                        <!-- Text context -->
+                                        <template x-if="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.has_more_before">
+                                            <span class="chunk-ellipsis">...</span>
+                                        </template>
+                                        <span class="chunk-text" x-text="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.before_context"></span><span class="chunk-matched" x-text="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.chunk_text"></span><span class="chunk-text" x-text="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.after_context"></span><template x-if="expandedChunks[`${result.doc_type}_${result.id}_${result.chunk_start_offset || 0}`]?.has_more_after">
+                                            <span class="chunk-ellipsis">...</span>
+                                        </template>
+                                    </div>
+                                </template>
+                            </div>
+                        </template>
+                    </div>
+                </template>
+            </div>
+        </template>
+        </div><!-- Search Results -->
+    </div><!-- .viz-layout -->
+</div><!-- x-data="vizApp()" -->

nextcloud_mcp_server/auth/templates/welcome.html

@@ -0,0 +1,392 @@
+{% extends "base.html" %}
+
+{% block title %}Welcome - Nextcloud MCP Server{% endblock %}
+
+{% block extra_head %}
+    <!-- Alpine.js for interactive elements -->
+    <script defer src="https://cdn.jsdelivr.net/npm/alpinejs@3.x.x/dist/cdn.min.js"></script>
+{% endblock %}
+
+{% block extra_styles %}
+    /* Welcome page specific styles */
+    .hero-section {
+        background: linear-gradient(135deg, var(--color-primary-element) 0%, #0082c9 100%);
+        color: white;
+        padding: 60px 24px;
+        margin: -24px -24px 40px -24px;
+        border-radius: 0 0 var(--border-radius-large) var(--border-radius-large);
+        text-align: center;
+    }
+
+    .hero-section h1 {
+        color: white;
+        font-size: 36px;
+        margin: 0 0 16px 0;
+        font-weight: 600;
+    }
+
+    .hero-section p {
+        font-size: 18px;
+        opacity: 0.95;
+        max-width: 700px;
+        margin: 0 auto;
+        line-height: 1.6;
+    }
+
+    .feature-grid {
+        display: grid;
+        grid-template-columns: repeat(auto-fit, minmax(280px, 1fr));
+        gap: 24px;
+        margin: 32px 0;
+    }
+
+    .feature-card {
+        background: var(--color-main-background);
+        border: 2px solid var(--color-border);
+        border-radius: var(--border-radius-large);
+        padding: 24px;
+        transition: all 0.2s;
+        cursor: pointer;
+        text-decoration: none;
+        color: inherit;
+        display: block;
+    }
+
+    .feature-card:hover {
+        border-color: var(--color-primary-element);
+        box-shadow: 0 4px 12px rgba(0, 103, 158, 0.15);
+        transform: translateY(-2px);
+    }
+
+    .feature-card h3 {
+        color: var(--color-primary-element);
+        font-size: 20px;
+        margin: 12px 0 8px 0;
+        font-weight: 600;
+        display: flex;
+        align-items: center;
+        gap: 12px;
+    }
+
+    .feature-card p {
+        color: var(--color-text-maxcontrast);
+        font-size: 14px;
+        line-height: 1.6;
+        margin: 8px 0 0 0;
+    }
+
+    .feature-icon {
+        width: 48px;
+        height: 48px;
+        background: var(--color-primary-element-light);
+        border-radius: var(--border-radius);
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        margin-bottom: 8px;
+    }
+
+    .feature-icon svg {
+        width: 28px;
+        height: 28px;
+        fill: var(--color-primary-element);
+    }
+
+    .info-section {
+        background: var(--color-background-hover);
+        border-radius: var(--border-radius-large);
+        padding: 32px;
+        margin: 32px 0;
+    }
+
+    .info-section h2 {
+        color: var(--color-main-text);
+        font-size: 24px;
+        margin: 0 0 16px 0;
+        border: none;
+        padding: 0;
+    }
+
+    .info-section p {
+        color: var(--color-text-maxcontrast);
+        line-height: 1.7;
+        margin: 12px 0;
+    }
+
+    .info-section ul {
+        margin: 12px 0;
+        padding-left: 24px;
+    }
+
+    .info-section li {
+        color: var(--color-text-maxcontrast);
+        line-height: 1.7;
+        margin: 8px 0;
+    }
+
+    .info-section code {
+        background: var(--color-main-background);
+        padding: 2px 8px;
+        border-radius: var(--border-radius);
+        font-size: 13px;
+    }
+
+    .auth-status {
+        background: var(--color-primary-element-light);
+        border-left: 4px solid var(--color-primary-element);
+        padding: 16px 20px;
+        margin: 24px 0;
+        border-radius: var(--border-radius);
+        display: flex;
+        align-items: center;
+        gap: 12px;
+    }
+
+    .auth-status svg {
+        width: 24px;
+        height: 24px;
+        fill: var(--color-primary-element);
+        flex-shrink: 0;
+    }
+
+    .auth-status-text {
+        flex: 1;
+    }
+
+    .auth-status-text strong {
+        display: block;
+        color: var(--color-main-text);
+        font-size: 14px;
+        margin-bottom: 4px;
+    }
+
+    .auth-status-text span {
+        color: var(--color-text-maxcontrast);
+        font-size: 13px;
+    }
+{% endblock %}
+
+{% block content %}
+<div class="app-content-wrapper">
+    <!-- Main Content Area -->
+    <main id="app-content">
+        <div class="page-content">
+            <!-- Hero Section -->
+            <div class="hero-section">
+                <h1>Welcome to Nextcloud MCP Server</h1>
+                <p>
+                    Interactive user interface for semantic search and document retrieval.
+                    Test queries, visualize results, and explore your Nextcloud content using RAG workflows.
+                </p>
+            </div>
+
+            <!-- Authentication Status -->
+            <div class="auth-status">
+                <svg viewBox="0 0 24 24">
+                    <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                </svg>
+                <div class="auth-status-text">
+                    <strong>Authenticated as: {{ username }}</strong>
+                    <span>Authentication mode: <code>{{ auth_mode }}</code></span>
+                </div>
+            </div>
+
+            {% if vector_sync_enabled %}
+            <!-- Vector Sync Enabled Content -->
+            <div class="info-section">
+                <h2>About Semantic Search</h2>
+                <p>
+                    This interface provides access to <strong>semantic search</strong> capabilities powered by vector embeddings.
+                    Unlike traditional keyword search, semantic search understands the <em>meaning</em> of your queries and finds
+                    conceptually similar content across your Nextcloud apps.
+                </p>
+                <p>
+                    <strong>How it works:</strong>
+                </p>
+                <ul>
+                    <li>Documents from Notes, Calendar, Files, Contacts, and Deck are indexed into a vector database</li>
+                    <li>Each document chunk is converted to a 768-dimensional vector embedding that captures semantic meaning</li>
+                    <li>Queries are also converted to embeddings and matched against document vectors using similarity search</li>
+                    <li>Results can be retrieved using pure semantic search or hybrid BM25 search combining keywords and semantics</li>
+                </ul>
+            </div>
+
+            <div class="info-section">
+                <h2>RAG Workflow Integration</h2>
+                <p>
+                    This UI allows you to <strong>test the same queries that Large Language Models (LLMs) would use</strong> in a
+                    Retrieval-Augmented Generation (RAG) workflow. When an AI assistant needs to answer questions about your data:
+                </p>
+                <ul>
+                    <li><strong>Step 1:</strong> The assistant converts your question into a search query</li>
+                    <li><strong>Step 2:</strong> The MCP server retrieves relevant document chunks using semantic search</li>
+                    <li><strong>Step 3:</strong> Retrieved context is passed to the LLM to generate an informed answer</li>
+                </ul>
+
+                <!-- RAG Workflow Diagram -->
+                <div style="background: var(--color-main-background); border: 2px solid var(--color-primary-element); border-radius: var(--border-radius-large); padding: 24px; margin: 24px 0; font-family: 'SFMono-Regular', 'Consolas', 'Liberation Mono', 'Menlo', monospace; font-size: 13px; line-height: 1.8; overflow-x: auto;">
+                    <div style="text-align: center; font-weight: 600; margin-bottom: 16px; color: var(--color-primary-element); font-size: 14px;">
+                        MCP Sampling RAG Workflow
+                    </div>
+                    <pre style="margin: 0; color: var(--color-main-text);">
+┌─────────────────┐
+│   <strong>MCP Client</strong>   │  User asks: "What are health benefits of coffee?"
+│  (Claude Code)  │
+└────────┬────────┘
+         │ (1) User question
+         ↓
+┌────────────────────────────────────────────────────────────────────────┐
+│                      <strong>Nextcloud MCP Server</strong>                          │
+│  ┌──────────────────────────────────────────────────────────────────┐  │
+│  │ <strong>nc_semantic_search_answer</strong> Tool (MCP Sampling-enabled)      │  │
+│  │                                                                  │  │
+│  │  (2) Semantic Search                                             │  │
+│  │  ┌────────────────────────────────────────────────────────┐     │  │
+│  │  │ Query: "health benefits of coffee"                     │     │  │
+│  │  │ → Convert to 768D vector embedding                     │     │  │
+│  │  │ → Search Qdrant (BM25 Hybrid + RRF fusion)             │     │  │
+│  │  │ → Retrieve top 5 relevant document chunks              │     │  │
+│  │  └────────────────────────────────────────────────────────┘     │  │
+│  │                                                                  │  │
+│  │  (3) Construct Prompt with Context                               │  │
+│  │  ┌────────────────────────────────────────────────────────┐     │  │
+│  │  │ "What are health benefits of coffee?                   │     │  │
+│  │  │                                                         │     │  │
+│  │  │  Documents:                                             │     │  │
+│  │  │  - [MED-2155] Effects of habitual coffee consumption...│     │  │
+│  │  │  - [MED-1646] Beverage consumption guidance...         │     │  │
+│  │  │  - [MED-1627] Coffee and depression risk...            │     │  │
+│  │  │  ...                                                    │     │  │
+│  │  │                                                         │     │  │
+│  │  │  Provide answer with citations."                        │     │  │
+│  │  └────────────────────────────────────────────────────────┘     │  │
+│  │                                                                  │  │
+│  │  (4) MCP Sampling Request                                        │  │
+│  │  ─────────────────────────────────────────────────────────────> │  │
+│  └──────────────────────────────────────────────────────────────────┘  │
+└────────────────────────────────────────────────────────────────────────┘
+         │
+         │ Sampling request with prompt + context
+         ↓
+┌─────────────────┐
+│   <strong>MCP Client</strong>   │  (5) Client's LLM generates answer using retrieved context
+│    (Claude)     │      → "Coffee consumption (2-3 cups/day) is associated with
+└────────┬────────┘         reduced risk of type 2 diabetes, cardiovascular disease,
+         │                  and improved liver health (Document 1, 2)..."
+         │
+         │ (6) Answer with citations
+         ↓
+┌─────────────────┐
+│      User       │  Receives comprehensive answer with source citations
+└─────────────────┘</pre>
+                </div>
+
+                <p style="margin-top: 16px;">
+                    <strong>Key Point:</strong> The MCP server retrieves context but doesn't generate answers itself.
+                    Through <strong>MCP sampling</strong>, it requests the client's LLM to generate responses, giving users
+                    full control over which model is used and ensuring all processing happens client-side.
+                </p>
+
+                <p>
+                    By using this interface, you can preview search results, understand relevance scores, and verify
+                    that the system retrieves the right information before it reaches the LLM.
+                </p>
+            </div>
+
+            <!-- Feature Cards -->
+            <h2>Available Features</h2>
+            <div class="feature-grid">
+                <a href="/app/user-info" class="feature-card">
+                    <div class="feature-icon">
+                        <svg viewBox="0 0 24 24">
+                            <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                        </svg>
+                    </div>
+                    <h3>User Information</h3>
+                    <p>
+                        View your authentication details, session information, and IdP profile.
+                        Manage background access permissions.
+                    </p>
+                </a>
+
+                <a href="/app/user-info#vector-sync" class="feature-card">
+                    <div class="feature-icon">
+                        <svg viewBox="0 0 24 24">
+                            <path d="M12,18A6,6 0 0,1 6,12C6,11 6.25,10.03 6.7,9.2L5.24,7.74C4.46,8.97 4,10.43 4,12A8,8 0 0,0 12,20V23L16,19L12,15M12,4V1L8,5L12,9V6A6,6 0 0,1 18,12C18,13 17.75,13.97 17.3,14.8L18.76,16.26C19.54,15.03 20,13.57 20,12A8,8 0 0,0 12,4Z" />
+                        </svg>
+                    </div>
+                    <h3>Vector Sync Status</h3>
+                    <p>
+                        Monitor real-time indexing progress with metrics for indexed documents, pending queue,
+                        and synchronization status.
+                    </p>
+                </a>
+
+                <a href="/app/user-info#vector-viz" class="feature-card">
+                    <div class="feature-icon">
+                        <svg viewBox="0 0 24 24">
+                            <path d="M22,21H2V3H4V19H6V10H10V19H12V6H16V19H18V14H22V21Z" />
+                        </svg>
+                    </div>
+                    <h3>Vector Visualization</h3>
+                    <p>
+                        Interactive search interface with 2D PCA visualization. Compare algorithms,
+                        view relevance scores, and explore matched document chunks.
+                    </p>
+                </a>
+            </div>
+
+            {% else %}
+            <!-- Vector Sync Disabled Content -->
+            <div class="warning">
+                <h3 style="margin-top: 0;">Vector Sync is Disabled</h3>
+                <p>
+                    Semantic search and vector visualization features are currently disabled.
+                    To enable these features, set <code>VECTOR_SYNC_ENABLED=true</code> in your environment configuration.
+                </p>
+                <p style="margin-bottom: 0;">
+                    <strong>Learn more:</strong>
+                    <a href="https://github.com/YOUR_REPO/docs/configuration.md" target="_blank" style="color: inherit; text-decoration: underline;">
+                        Configuration Guide
+                    </a>
+                </p>
+            </div>
+
+            <!-- Limited Feature Card -->
+            <h2>Available Features</h2>
+            <div class="feature-grid">
+                <a href="/app/user-info" class="feature-card">
+                    <div class="feature-icon">
+                        <svg viewBox="0 0 24 24">
+                            <path d="M12,4A4,4 0 0,1 16,8A4,4 0 0,1 12,12A4,4 0 0,1 8,8A4,4 0 0,1 12,4M12,14C16.42,14 20,15.79 20,18V20H4V18C4,15.79 7.58,14 12,14Z" />
+                        </svg>
+                    </div>
+                    <h3>User Information</h3>
+                    <p>
+                        View your authentication details, session information, and IdP profile.
+                        Manage background access permissions.
+                    </p>
+                </a>
+            </div>
+            {% endif %}
+
+            <!-- Documentation Section -->
+            <div class="info-section" style="margin-top: 40px;">
+                <h2>Documentation</h2>
+                <p>
+                    For detailed information about configuration, authentication modes, and advanced features,
+                    please refer to the project documentation:
+                </p>
+                <ul>
+                    <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/installation.md" target="_blank">Installation Guide</a></li>
+                    <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/configuration.md" target="_blank">Configuration Options</a></li>
+                    <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/authentication.md" target="_blank">Authentication Modes</a></li>
+                    {% if vector_sync_enabled %}
+                    <li><a href="https://github.com/cbcoutinho/nextcloud-mcp-server/blob/master/docs/user-guide/vector-sync-ui.md" target="_blank">Vector Sync UI Guide</a></li>
+                    {% endif %}
+                </ul>
+            </div>
+        </div>
+    </main>
+</div>
+{% endblock %}

nextcloud_mcp_server/auth/unified_verifier.py

@@ -303,10 +303,13 @@ class UnifiedTokenVerifier(TokenVerifier):
 
         try:
             # Introspection requires client authentication
+            client_id = self.settings.oidc_client_id
+            client_secret = self.settings.oidc_client_secret
+            assert client_id is not None and client_secret is not None
             response = await self.http_client.post(
                 self.introspection_uri,
                 data={"token": token},
-                auth=(self.settings.oidc_client_id, self.settings.oidc_client_secret),
+                auth=(client_id, client_secret),
             )
 
             if response.status_code == 200:

nextcloud_mcp_server/auth/userinfo_routes.py

@@ -9,24 +9,38 @@ For OAuth mode: Requires browser-based OAuth login to establish session.
 
 import logging
 import os
+from pathlib import Path
 from typing import Any
 
 import httpx
+from jinja2 import Environment, FileSystemLoader
 from starlette.authentication import requires
 from starlette.requests import Request
 from starlette.responses import HTMLResponse, JSONResponse
 
+from nextcloud_mcp_server.client import NextcloudClient
+
 logger = logging.getLogger(__name__)
 
+# Setup Jinja2 environment for templates
+_template_dir = Path(__file__).parent / "templates"
+_jinja_env = Environment(loader=FileSystemLoader(_template_dir))
 
-async def _get_authenticated_client_for_userinfo(request: Request) -> httpx.AsyncClient:
-    """Get an authenticated HTTP client for user info page operations.
+
+async def _get_authenticated_client_for_userinfo(request: Request) -> NextcloudClient:
+    """Get an authenticated Nextcloud client for user info page operations.
+
+    This is a shared helper for authenticated routes that need to access
+    Nextcloud APIs. It handles both BasicAuth and OAuth authentication modes.
 
     Args:
         request: Starlette request object
 
     Returns:
-        Authenticated httpx.AsyncClient
+        Authenticated NextcloudClient
+
+    Raises:
+        RuntimeError: If credentials/session not configured
     """
     oauth_ctx = getattr(request.app.state, "oauth_context", None)
 
@@ -39,11 +53,15 @@ async def _get_authenticated_client_for_userinfo(request: Request) -> httpx.Asyn
         if not all([nextcloud_host, username, password]):
             raise RuntimeError("BasicAuth credentials not configured")
 
-        assert nextcloud_host is not None  # Type narrowing for type checker
-        return httpx.AsyncClient(
+        from httpx import BasicAuth
+
+        assert nextcloud_host is not None
+        assert username is not None
+        assert password is not None
+        return NextcloudClient(
             base_url=nextcloud_host,
-            auth=(username, password),
-            timeout=30.0,
+            username=username,
+            auth=BasicAuth(username, password),
         )
 
     # OAuth mode - get token from session
@@ -58,15 +76,14 @@ async def _get_authenticated_client_for_userinfo(request: Request) -> httpx.Asyn
         raise RuntimeError("No access token found in session")
 
     access_token = token_data["access_token"]
+    username = token_data.get("username")
     nextcloud_host = oauth_ctx.get("config", {}).get("nextcloud_host", "")
 
-    if not nextcloud_host:
-        raise RuntimeError("Nextcloud host not configured")
+    if not nextcloud_host or not username:
+        raise RuntimeError("Nextcloud host or username not configured")
 
-    return httpx.AsyncClient(
-        base_url=nextcloud_host,
-        headers={"Authorization": f"Bearer {access_token}"},
-        timeout=30.0,
+    return NextcloudClient.from_token(
+        base_url=nextcloud_host, token=access_token, username=username
     )
 
 
@@ -417,10 +434,10 @@ async def user_info_html(request: Request) -> HTMLResponse:
     try:
         from nextcloud_mcp_server.auth.permissions import is_nextcloud_admin
 
-        # Get authenticated HTTP client
-        http_client = await _get_authenticated_client_for_userinfo(request)
-        is_admin = await is_nextcloud_admin(request, http_client)
-        await http_client.aclose()
+        # Get authenticated Nextcloud client
+        nc_client = await _get_authenticated_client_for_userinfo(request)
+        is_admin = await is_nextcloud_admin(request, nc_client._client)
+        await nc_client.close()
     except Exception as e:
         logger.warning(f"Failed to check admin status: {e}")
         # Default to not admin if check fails
@@ -431,51 +448,14 @@ async def user_info_html(request: Request) -> HTMLResponse:
         oauth_ctx = getattr(request.app.state, "oauth_context", None)
         login_url = str(request.url_for("oauth_login")) if oauth_ctx else "/oauth/login"
 
-        error_html = f"""
-        <!DOCTYPE html>
-        <html lang="en">
-        <head>
-            <meta charset="UTF-8">
-            <meta name="viewport" content="width=device-width, initial-scale=1.0">
-            <title>Error - Nextcloud MCP Server</title>
-            <style>
-                body {{
-                    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
-                    max-width: 800px;
-                    margin: 50px auto;
-                    padding: 20px;
-                    background-color: #f5f5f5;
-                }}
-                .container {{
-                    background: white;
-                    border-radius: 8px;
-                    padding: 30px;
-                    box-shadow: 0 2px 4px rgba(0,0,0,0.1);
-                }}
-                h1 {{
-                    color: #d32f2f;
-                    margin-top: 0;
-                }}
-                .error {{
-                    background-color: #ffebee;
-                    border-left: 4px solid #d32f2f;
-                    padding: 15px;
-                    margin: 20px 0;
-                }}
-            </style>
-        </head>
-        <body>
-            <div class="container">
-                <h1>Error Retrieving User Info</h1>
-                <div class="error">
-                    <strong>Error:</strong> {user_context["error"]}
-                </div>
-                <p><a href="{login_url}">Login again</a></p>
-            </div>
-        </body>
-        </html>
-        """
-        return HTMLResponse(content=error_html)
+        template = _jinja_env.get_template("error.html")
+        return HTMLResponse(
+            content=template.render(
+                error_title="Error Retrieving User Info",
+                error_message=user_context["error"],
+                login_url=login_url,
+            )
+        )
 
     # Build HTML response
     auth_mode = user_context.get("auth_mode", "unknown")
@@ -654,410 +634,26 @@ async def user_info_html(request: Request) -> HTMLResponse:
             </div>
         """
 
-    html_content = f"""
-    <!DOCTYPE html>
-    <html lang="en">
-    <head>
-        <meta charset="UTF-8">
-        <meta name="viewport" content="width=device-width, initial-scale=1.0">
-        <title>Nextcloud MCP Server</title>
+    # Check if vector sync is enabled (needed for Welcome tab)
+    vector_sync_enabled = os.getenv("VECTOR_SYNC_ENABLED", "false").lower() == "true"
 
-        <!-- htmx for dynamic loading -->
-        <script src="https://unpkg.com/htmx.org@1.9.10"></script>
-
-        <!-- Alpine.js for tab state management -->
-        <script defer src="https://cdn.jsdelivr.net/npm/alpinejs@3.x.x/dist/cdn.min.js"></script>
-
-        <!-- Plotly.js for vector visualization -->
-        <script src="https://cdn.plot.ly/plotly-2.27.0.min.js"></script>
-
-        <!-- Vector visualization app (Alpine.js component) -->
-        <script>
-            function vizApp() {{
-                return {{
-                    query: '',
-                    algorithm: 'hybrid',
-                    showAdvanced: false,
-                    docTypes: [''],  // Default to "All Types"
-                    limit: 50,
-                    scoreThreshold: 0.7,
-                    semanticWeight: 0.5,
-                    keywordWeight: 0.3,
-                    fuzzyWeight: 0.2,
-                    loading: false,
-                    results: [],
-
-                    async executeSearch() {{
-                        this.loading = true;
-                        this.results = [];
-
-                        try {{
-                            const params = new URLSearchParams({{
-                                query: this.query,
-                                algorithm: this.algorithm,
-                                limit: this.limit,
-                                score_threshold: this.scoreThreshold,
-                                semantic_weight: this.semanticWeight,
-                                keyword_weight: this.keywordWeight,
-                                fuzzy_weight: this.fuzzyWeight,
-                            }});
-
-                            // Add doc_types parameter (filter out empty string for "All Types")
-                            const selectedTypes = this.docTypes.filter(t => t !== '');
-                            if (selectedTypes.length > 0) {{
-                                params.append('doc_types', selectedTypes.join(','));
-                            }}
-
-                            const response = await fetch(`/app/vector-viz/search?${{params}}`);
-                            const data = await response.json();
-
-                            if (data.success) {{
-                                this.results = data.results;
-                                this.renderPlot(data.coordinates_2d, data.results);
-                            }} else {{
-                                alert('Search failed: ' + data.error);
-                            }}
-                        }} catch (error) {{
-                            alert('Error: ' + error.message);
-                        }} finally {{
-                            this.loading = false;
-                        }}
-                    }},
-
-                    renderPlot(coordinates, results) {{
-                        // Calculate score range for auto-scaling
-                        const scores = results.map(r => r.score);
-                        const minScore = Math.min(...scores);
-                        const maxScore = Math.max(...scores);
-
-                        const trace = {{
-                            x: coordinates.map(c => c[0]),
-                            y: coordinates.map(c => c[1]),
-                            mode: 'markers',
-                            type: 'scatter',
-                            text: results.map(r => `${{r.title}}<br>Score: ${{r.score.toFixed(3)}}`),
-                            marker: {{
-                                // Multi-channel encoding: size + opacity + color for visual hierarchy
-                                // Power scaling (score^2) amplifies visual differences dramatically
-                                // score=0.0 → 6px, score=0.5 → 9.5px, score=1.0 → 20px
-                                size: results.map(r => 6 + (Math.pow(r.score, 2) * 14)),
-                                // Linear opacity scaling (0.2-1.0 range keeps all points visible)
-                                opacity: results.map(r => 0.2 + (r.score * 0.8)),
-                                // Color gradient shows score
-                                color: scores,
-                                colorscale: 'Viridis',
-                                showscale: true,
-                                colorbar: {{ title: 'Relative Score' }},
-                                // Scores are normalized 0-1 within result set
-                                cmin: 0,
-                                cmax: 1
-                            }}
-                        }};
-
-                        const layout = {{
-                            title: `Vector Space (PCA 2D) - ${{results.length}} results`,
-                            xaxis: {{ title: 'PC1' }},
-                            yaxis: {{ title: 'PC2' }},
-                            hovermode: 'closest',
-                            height: 600
-                        }};
-
-                        Plotly.newPlot('viz-plot', [trace], layout);
-                    }},
-
-                    getNextcloudUrl(result) {{
-                        // Generate Nextcloud URL based on document type
-                        // Use the actual Nextcloud host (port 8080), not the MCP server
-                        const baseUrl = '{nextcloud_host_for_links}';
-
-                        switch (result.doc_type) {{
-                            case 'note':
-                                return `${{baseUrl}}/apps/notes/note/${{result.id}}`;
-                            case 'file':
-                                return `${{baseUrl}}/apps/files/?fileId=${{result.id}}`;
-                            case 'calendar':
-                                return `${{baseUrl}}/apps/calendar`;
-                            case 'contact':
-                                return `${{baseUrl}}/apps/contacts`;
-                            case 'deck':
-                                return `${{baseUrl}}/apps/deck`;
-                            default:
-                                return `${{baseUrl}}`;
-                        }}
-                    }}
-                }}
-            }}
-        </script>
-
-        <style>
-            body {{
-                font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
-                max-width: 900px;
-                margin: 50px auto;
-                padding: 20px;
-                background-color: #f5f5f5;
-            }}
-            .container {{
-                background: white;
-                border-radius: 8px;
-                padding: 30px;
-                box-shadow: 0 2px 4px rgba(0,0,0,0.1);
-                min-height: calc(100vh - 200px);
-            }}
-            h1 {{
-                color: #0082c9;
-                margin-top: 0;
-                border-bottom: 2px solid #0082c9;
-                padding-bottom: 10px;
-            }}
-            h2 {{
-                color: #333;
-                margin-top: 20px;
-                border-bottom: 1px solid #e0e0e0;
-                padding-bottom: 5px;
-            }}
-
-            /* Tab navigation */
-            .tabs {{
-                display: flex;
-                gap: 0;
-                margin: 20px 0 0 0;
-                border-bottom: 2px solid #e0e0e0;
-            }}
-            .tab {{
-                padding: 12px 24px;
-                cursor: pointer;
-                background: transparent;
-                border: none;
-                font-size: 14px;
-                font-weight: 500;
-                color: #666;
-                border-bottom: 2px solid transparent;
-                margin-bottom: -2px;
-                transition: all 0.2s;
-            }}
-            .tab:hover {{
-                color: #0082c9;
-                background-color: #f5f5f5;
-            }}
-            .tab.active {{
-                color: #0082c9;
-                border-bottom-color: #0082c9;
-            }}
-
-            /* Tab content - use grid to overlay panes */
-            .tab-content {{
-                padding: 20px 0;
-                display: grid;
-            }}
-
-            /* Tab panes - all occupy the same grid cell to overlay */
-            .tab-pane {{
-                grid-area: 1 / 1;
-            }}
-
-            /* Tables */
-            table {{
-                width: 100%;
-                border-collapse: collapse;
-                margin: 15px 0;
-            }}
-            td {{
-                padding: 10px;
-                border-bottom: 1px solid #e0e0e0;
-            }}
-            td:first-child {{
-                width: 200px;
-                color: #666;
-            }}
-            code {{
-                background-color: #f5f5f5;
-                padding: 2px 6px;
-                border-radius: 3px;
-                font-family: 'Courier New', monospace;
-            }}
-
-            /* Badges */
-            .badge {{
-                display: inline-block;
-                padding: 3px 8px;
-                border-radius: 12px;
-                font-size: 12px;
-                font-weight: bold;
-                text-transform: uppercase;
-            }}
-            .badge-oauth {{
-                background-color: #4caf50;
-                color: white;
-            }}
-            .badge-basic {{
-                background-color: #2196f3;
-                color: white;
-            }}
-
-            /* Messages */
-            .warning {{
-                background-color: #fff3cd;
-                border-left: 4px solid #ffc107;
-                padding: 15px;
-                margin: 15px 0;
-                color: #856404;
-            }}
-            .info-message {{
-                background-color: #e3f2fd;
-                border-left: 4px solid #2196f3;
-                padding: 15px;
-                margin: 15px 0;
-                color: #1565c0;
-            }}
-
-            /* Buttons */
-            .button {{
-                display: inline-block;
-                padding: 10px 20px;
-                background-color: #d32f2f;
-                color: white;
-                text-decoration: none;
-                border-radius: 4px;
-                transition: background-color 0.3s;
-                border: none;
-                cursor: pointer;
-                font-size: 14px;
-            }}
-            .button:hover {{
-                background-color: #b71c1c;
-            }}
-            .button-primary {{
-                background-color: #0082c9;
-            }}
-            .button-primary:hover {{
-                background-color: #006ba3;
-            }}
-
-            /* Logout section */
-            .logout {{
-                margin-top: 30px;
-                padding-top: 20px;
-                border-top: 1px solid #e0e0e0;
-            }}
-
-            /* Smooth htmx content swaps */
-            .htmx-swapping {{
-                opacity: 0;
-                transition: opacity 200ms ease-out;
-            }}
-
-            /* Smooth htmx content settling */
-            .htmx-settling {{
-                opacity: 1;
-                transition: opacity 200ms ease-in;
-            }}
-        </style>
-    </head>
-    <body>
-        <div class="container" x-data="{{ activeTab: 'user-info' }}">
-            <h1>Nextcloud MCP Server</h1>
-
-            <!-- Tab Navigation -->
-            <div class="tabs">
-                <button
-                    class="tab"
-                    :class="activeTab === 'user-info' ? 'active' : ''"
-                    @click="activeTab = 'user-info'">
-                    User Info
-                </button>
-                {
-        ""
-        if not show_vector_sync_tab
-        else '''
-                <button
-                    class="tab"
-                    :class="activeTab === 'vector-sync' ? 'active' : ''"
-                    @click="activeTab = 'vector-sync'">
-                    Vector Sync
-                </button>
-                '''
-    }
-                {
-        ""
-        if not show_vector_sync_tab
-        else '''
-                <button
-                    class="tab"
-                    :class="activeTab === 'vector-viz' ? 'active' : ''"
-                    @click="activeTab = 'vector-viz'">
-                    Vector Viz
-                </button>
-                '''
-    }
-                {
-        ""
-        if not show_webhooks_tab
-        else '''
-                <button
-                    class="tab"
-                    :class="activeTab === 'webhooks' ? 'active' : ''"
-                    @click="activeTab = 'webhooks'">
-                    Webhooks
-                </button>
-                '''
-    }
-            </div>
-
-            <!-- Tab Content -->
-            <div class="tab-content">
-                <!-- User Info Tab -->
-                <div class="tab-pane" x-show="activeTab === 'user-info'" x-transition.opacity.duration.150ms>
-                    {user_info_tab_html}
-                </div>
-
-                {
-        ""
-        if not show_vector_sync_tab
-        else f'''
-                <!-- Vector Sync Tab -->
-                <div class="tab-pane" x-show="activeTab === 'vector-sync'" x-transition.opacity.duration.150ms>
-                    {vector_sync_tab_html}
-                </div>
-                '''
-    }
-
-                {
-        ""
-        if not show_vector_sync_tab
-        else '''
-                <!-- Vector Viz Tab -->
-                <div class="tab-pane" x-show="activeTab === 'vector-viz'" x-transition.opacity.duration.150ms>
-                    <div hx-get="/app/vector-viz" hx-trigger="load" hx-swap="outerHTML">
-                        <p style="color: #999;">Loading vector visualization...</p>
-                    </div>
-                </div>
-                '''
-    }
-
-                {
-        ""
-        if not show_webhooks_tab
-        else f'''
-                <!-- Webhooks Tab (admin-only, loaded dynamically) -->
-                <div class="tab-pane" x-show="activeTab === 'webhooks'" x-transition.opacity.duration.150ms>
-                    {webhooks_tab_html}
-                </div>
-                '''
-    }
-            </div>
-
-            {
-        f'<div class="logout"><a href="{logout_url}" class="button">Logout</a></div>'
-        if auth_mode == "oauth"
-        else ""
-    }
-        </div>
-    </body>
-    </html>
-    """
-
-    return HTMLResponse(content=html_content)
+    # Render template
+    template = _jinja_env.get_template("user_info.html")
+    return HTMLResponse(
+        content=template.render(
+            user_info_tab_html=user_info_tab_html,
+            vector_sync_tab_html=vector_sync_tab_html,
+            webhooks_tab_html=webhooks_tab_html,
+            show_vector_sync_tab=show_vector_sync_tab,
+            show_webhooks_tab=show_webhooks_tab,
+            logout_url=logout_url if auth_mode == "oauth" else None,
+            nextcloud_host_for_links=nextcloud_host_for_links,
+            # Additional context for Welcome tab
+            vector_sync_enabled=vector_sync_enabled,
+            username=username,
+            auth_mode=auth_mode,
+        )
+    )
 
 
 @requires("authenticated", redirect="oauth_login")
@@ -1077,17 +673,12 @@ async def revoke_session(request: Request) -> HTMLResponse:
     oauth_ctx = getattr(request.app.state, "oauth_context", None)
 
     if not oauth_ctx:
+        template = _jinja_env.get_template("error.html")
         return HTMLResponse(
-            """
-            <!DOCTYPE html>
-            <html>
-            <head><title>Error</title></head>
-            <body>
-                <h1>Error</h1>
-                <p>OAuth mode not enabled</p>
-            </body>
-            </html>
-            """,
+            content=template.render(
+                error_title="Error",
+                error_message="OAuth mode not enabled",
+            ),
             status_code=400,
         )
 
@@ -1095,17 +686,12 @@ async def revoke_session(request: Request) -> HTMLResponse:
     session_id = request.cookies.get("mcp_session")
 
     if not storage or not session_id:
+        template = _jinja_env.get_template("error.html")
         return HTMLResponse(
-            """
-            <!DOCTYPE html>
-            <html>
-            <head><title>Error</title></head>
-            <body>
-                <h1>Error</h1>
-                <p>Session not found</p>
-            </body>
-            </html>
-            """,
+            content=template.render(
+                error_title="Error",
+                error_message="Session not found",
+            ),
             status_code=400,
         )
 
@@ -1118,57 +704,26 @@ async def revoke_session(request: Request) -> HTMLResponse:
         # Redirect back to user page
         user_page_url = str(request.url_for("user_info_html"))
 
+        template = _jinja_env.get_template("success.html")
         return HTMLResponse(
-            f"""
-            <!DOCTYPE html>
-            <html lang="en">
-            <head>
-                <meta charset="UTF-8">
-                <meta http-equiv="refresh" content="2;url={user_page_url}">
-                <title>Background Access Revoked</title>
-                <style>
-                    body {{
-                        font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif;
-                        max-width: 600px;
-                        margin: 50px auto;
-                        padding: 20px;
-                        text-align: center;
-                    }}
-                    .success {{
-                        background-color: #e8f5e9;
-                        border: 2px solid #4caf50;
-                        padding: 30px;
-                        border-radius: 8px;
-                    }}
-                    h1 {{
-                        color: #4caf50;
-                    }}
-                </style>
-            </head>
-            <body>
-                <div class="success">
-                    <h1>✓ Background Access Revoked</h1>
-                    <p>Your refresh token has been deleted successfully.</p>
-                    <p>Browser session remains active.</p>
-                    <p>Redirecting back to user page...</p>
-                </div>
-            </body>
-            </html>
-            """
+            content=template.render(
+                success_title="✓ Background Access Revoked",
+                success_messages=[
+                    "Your refresh token has been deleted successfully.",
+                    "Browser session remains active.",
+                ],
+                redirect_url=user_page_url,
+                redirect_delay=2,
+            )
         )
 
     except Exception as e:
         logger.error(f"Failed to revoke background access: {e}")
+        template = _jinja_env.get_template("error.html")
         return HTMLResponse(
-            f"""
-            <!DOCTYPE html>
-            <html>
-            <head><title>Error</title></head>
-            <body>
-                <h1>Error</h1>
-                <p>Failed to revoke background access: {e}</p>
-            </body>
-            </html>
-            """,
+            content=template.render(
+                error_title="Error",
+                error_message=f"Failed to revoke background access: {e}",
+            ),
             status_code=500,
         )

nextcloud_mcp_server/auth/viz_routes.py

@@ -1,35 +1,42 @@
 """Vector visualization routes for testing search algorithms.
 
 Provides a web UI for users to test different search algorithms on their own
-indexed documents and visualize results in 2D space using PCA.
+indexed documents and visualize results in 3D space using PCA.
 
 All processing happens server-side following ADR-012:
 - Search execution via shared search/algorithms.py
-- PCA dimensionality reduction (768-dim → 2D)
-- Only 2D coordinates + metadata sent to client
-- Bandwidth-efficient (2 floats per doc vs 768)
+- Query embedding generation
+- PCA dimensionality reduction (768-dim → 3D)
+- Only 3D coordinates + metadata sent to client
+- Bandwidth-efficient (3 floats per doc vs 768)
 """
 
 import logging
 import time
+from pathlib import Path
 
 import numpy as np
+from jinja2 import Environment, FileSystemLoader
 from starlette.authentication import requires
 from starlette.requests import Request
 from starlette.responses import HTMLResponse, JSONResponse
 
 from nextcloud_mcp_server.config import get_settings
+from nextcloud_mcp_server.observability.tracing import trace_operation
 from nextcloud_mcp_server.search import (
-    FuzzySearchAlgorithm,
-    HybridSearchAlgorithm,
-    KeywordSearchAlgorithm,
+    BM25HybridSearchAlgorithm,
     SemanticSearchAlgorithm,
 )
 from nextcloud_mcp_server.vector.pca import PCA
+from nextcloud_mcp_server.vector.placeholder import get_placeholder_filter
 from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
 logger = logging.getLogger(__name__)
 
+# Setup Jinja2 environment for templates
+_template_dir = Path(__file__).parent / "templates"
+_jinja_env = Environment(loader=FileSystemLoader(_template_dir))
+
 
 @requires("authenticated", redirect="oauth_login")
 async def vector_visualization_html(request: Request) -> HTMLResponse:
@@ -65,284 +72,28 @@ async def vector_visualization_html(request: Request) -> HTMLResponse:
         else "unknown"
     )
 
-    html_content = f"""
-        <style>
-            .viz-card {{
-                background: white;
-                border-radius: 8px;
-                padding: 20px;
-                margin-bottom: 20px;
-                box-shadow: 0 2px 4px rgba(0,0,0,0.1);
-            }}
-            .viz-controls {{
-                margin-bottom: 20px;
-            }}
-            .viz-control-row {{
-                display: grid;
-                grid-template-columns: 2fr 1fr auto;
-                gap: 12px;
-                margin-bottom: 12px;
-                align-items: end;
-            }}
-            .viz-control-group {{
-                margin-bottom: 15px;
-            }}
-            .viz-control-group label {{
-                display: block;
-                margin-bottom: 5px;
-                font-weight: 500;
-                color: #333;
-            }}
-            .viz-control-group input[type="text"],
-            .viz-control-group input[type="number"],
-            .viz-control-group select {{
-                width: 100%;
-                padding: 8px 12px;
-                border: 1px solid #ddd;
-                border-radius: 4px;
-                font-size: 14px;
-            }}
-            .viz-control-group input[type="range"] {{
-                width: 100%;
-            }}
-            .viz-control-group select[multiple] {{
-                min-height: 100px;
-            }}
-            .viz-weight-display {{
-                display: inline-block;
-                min-width: 40px;
-                text-align: right;
-                color: #666;
-            }}
-            .viz-btn {{
-                background: #0066cc;
-                color: white;
-                border: none;
-                padding: 10px 20px;
-                border-radius: 4px;
-                cursor: pointer;
-                font-size: 14px;
-                font-weight: 500;
-            }}
-            .viz-btn:hover {{
-                background: #0052a3;
-            }}
-            .viz-btn-secondary {{
-                background: #6c757d;
-                color: white;
-                border: none;
-                padding: 6px 12px;
-                border-radius: 4px;
-                cursor: pointer;
-                font-size: 13px;
-                margin-bottom: 12px;
-            }}
-            .viz-btn-secondary:hover {{
-                background: #5a6268;
-            }}
-            #viz-plot-container {{
-                width: 100%;
-                height: 600px;
-                position: relative;
-            }}
-            #viz-plot {{
-                width: 100%;
-                height: 100%;
-            }}
-            .viz-loading {{
-                text-align: center;
-                padding: 40px;
-                color: #666;
-            }}
-            .viz-loading-overlay {{
-                position: absolute;
-                inset: 0;
-                display: flex;
-                align-items: center;
-                justify-content: center;
-                background: white;
-                color: #666;
-            }}
-            .viz-no-results {{
-                text-align: center;
-                padding: 40px;
-                color: #666;
-                font-style: italic;
-            }}
-            .viz-advanced-section {{
-                margin-top: 16px;
-                padding: 16px;
-                background: #f8f9fa;
-                border-radius: 4px;
-                border: 1px solid #dee2e6;
-            }}
-            .viz-advanced-grid {{
-                display: grid;
-                grid-template-columns: 1fr 1fr;
-                gap: 20px;
-            }}
-            .viz-info-box {{
-                background: #e3f2fd;
-                border-left: 4px solid #2196f3;
-                padding: 12px;
-                margin-bottom: 20px;
-                font-size: 14px;
-            }}
-        </style>
-
-        <div x-data="vizApp()">
-            <div class="viz-card">
-                <h2>Vector Visualization</h2>
-                <div class="viz-info-box">
-                    Testing search algorithms on your indexed documents. User: <strong>{username}</strong>
-                </div>
-
-                <form @submit.prevent="executeSearch">
-                    <div class="viz-controls">
-                        <!-- Main Controls -->
-                        <div class="viz-control-group">
-                            <label>Search Query</label>
-                            <input type="text" x-model="query" placeholder="Enter search query..." required />
-                        </div>
-
-                        <div class="viz-control-row">
-                            <div class="viz-control-group" style="margin-bottom: 0;">
-                                <label>Algorithm</label>
-                                <select x-model="algorithm">
-                                    <option value="semantic">Semantic (Vector Similarity)</option>
-                                    <option value="keyword">Keyword (Token Matching)</option>
-                                    <option value="fuzzy">Fuzzy (Character Overlap)</option>
-                                    <option value="hybrid" selected>Hybrid (RRF Fusion)</option>
-                                </select>
-                            </div>
-
-                            <div style="display: flex; align-items: flex-end;">
-                                <button type="submit" class="viz-btn" style="width: 100%;">Search & Visualize</button>
-                            </div>
-
-                            <div style="display: flex; align-items: flex-end;">
-                                <button type="button" class="viz-btn-secondary" @click="showAdvanced = !showAdvanced" style="white-space: nowrap;">
-                                    <span x-text="showAdvanced ? 'Hide Advanced' : 'Advanced'"></span>
-                                </button>
-                            </div>
-                        </div>
-
-                        <!-- Advanced Options (Collapsible) -->
-                        <div class="viz-advanced-section" x-show="showAdvanced" x-transition.opacity.duration.200ms>
-                            <h3 style="margin-top: 0; margin-bottom: 16px; font-size: 16px;">Advanced Options</h3>
-
-                            <div class="viz-advanced-grid">
-                                <div class="viz-control-group">
-                                    <label>Document Types</label>
-                                    <select x-model="docTypes" multiple>
-                                        <option value="">All Types (cross-app search)</option>
-                                        <option value="note">Notes</option>
-                                        <option value="file">Files</option>
-                                        <option value="calendar">Calendar Events</option>
-                                        <option value="contact">Contacts</option>
-                                        <option value="deck">Deck Cards</option>
-                                    </select>
-                                    <small style="color: #666; display: block; margin-top: 4px;">
-                                        Hold Ctrl/Cmd to select multiple
-                                    </small>
-                                </div>
-
-                                <div>
-                                    <div class="viz-control-group">
-                                        <label>Score Threshold (Semantic/Hybrid)</label>
-                                        <input type="number" x-model.number="scoreThreshold" min="0" max="1" step="0.1" />
-                                    </div>
-
-                                    <div class="viz-control-group">
-                                        <label>Result Limit</label>
-                                        <input type="number" x-model.number="limit" min="1" max="100" />
-                                    </div>
-                                </div>
-                            </div>
-
-                            <!-- Hybrid Weights (only when hybrid selected) -->
-                            <div x-show="algorithm === 'hybrid'" style="margin-top: 16px; padding: 12px; background: #e9ecef; border-radius: 4px;">
-                                <label style="margin-bottom: 12px; display: block;">Hybrid Algorithm Weights</label>
-
-                                <div style="margin-bottom: 8px;">
-                                    <label style="display: inline-block; width: 100px; font-weight: normal;">Semantic:</label>
-                                    <input type="range" x-model.number="semanticWeight" min="0" max="1" step="0.1" style="width: 200px; display: inline-block;">
-                                    <span class="viz-weight-display" x-text="semanticWeight.toFixed(1)"></span>
-                                </div>
-                                <div style="margin-bottom: 8px;">
-                                    <label style="display: inline-block; width: 100px; font-weight: normal;">Keyword:</label>
-                                    <input type="range" x-model.number="keywordWeight" min="0" max="1" step="0.1" style="width: 200px; display: inline-block;">
-                                    <span class="viz-weight-display" x-text="keywordWeight.toFixed(1)"></span>
-                                </div>
-                                <div>
-                                    <label style="display: inline-block; width: 100px; font-weight: normal;">Fuzzy:</label>
-                                    <input type="range" x-model.number="fuzzyWeight" min="0" max="1" step="0.1" style="width: 200px; display: inline-block;">
-                                    <span class="viz-weight-display" x-text="fuzzyWeight.toFixed(1)"></span>
-                                </div>
-                            </div>
-                        </div>
-                    </div>
-                </form>
-            </div>
-
-            <div class="viz-card">
-                <div id="viz-plot-container">
-                    <div x-show="loading" class="viz-loading-overlay" x-transition.opacity.duration.200ms>
-                        Executing search and computing PCA projection...
-                    </div>
-                    <div id="viz-plot" x-show="!loading" x-transition.opacity.duration.200ms></div>
-                </div>
-            </div>
-
-            <div class="viz-card">
-                <h3>Search Results (<span x-text="loading ? '...' : results.length"></span>)</h3>
-
-                <div x-show="loading" class="viz-loading" x-transition.opacity.duration.200ms>
-                    Loading results...
-                </div>
-
-                <div x-show="!loading && results.length === 0" class="viz-no-results" x-transition.opacity.duration.200ms>
-                    No results found. Try a different query or adjust your search parameters.
-                </div>
-
-                <template x-if="!loading && results.length > 0">
-                    <div x-transition.opacity.duration.200ms>
-                        <template x-for="result in results" :key="result.id">
-                            <div style="padding: 12px; border-bottom: 1px solid #eee;">
-                                <a :href="getNextcloudUrl(result)" target="_blank" style="font-weight: 500; color: #0066cc; text-decoration: none;">
-                                    <span x-text="result.title"></span>
-                                </a>
-                                <div style="font-size: 14px; color: #666; margin-top: 4px;" x-text="result.excerpt"></div>
-                                <div style="font-size: 12px; color: #999; margin-top: 4px;">
-                                    Score: <span x-text="result.score.toFixed(3)"></span> |
-                                    Type: <span x-text="result.doc_type"></span>
-                                </div>
-                            </div>
-                        </template>
-                    </div>
-                </template>
-            </div>
-        </div>
-    """
-
+    # Load and render template
+    template = _jinja_env.get_template("vector_viz.html")
+    html_content = template.render(username=username)
     return HTMLResponse(content=html_content)
 
 
 @requires("authenticated", redirect="oauth_login")
 async def vector_visualization_search(request: Request) -> JSONResponse:
-    """Execute server-side search and return 2D coordinates + results.
+    """Execute server-side search and return 3D coordinates + results.
 
     All processing happens server-side:
     1. Execute search via shared algorithm module
-    2. Fetch matching vectors from Qdrant
-    3. Apply PCA reduction (768-dim → 2D)
-    4. Return coordinates + metadata only
+    2. Generate query embedding
+    3. Fetch matching vectors from Qdrant
+    4. Apply PCA reduction (768-dim → 3D) to query + documents
+    5. Return coordinates + metadata only
 
     Args:
         request: Starlette request with query parameters
 
     Returns:
-        JSON response with coordinates_2d and results
+        JSON response with coordinates_3d and results (including query point)
     """
     settings = get_settings()
 
@@ -365,12 +116,10 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
 
     # Parse query parameters
     query = request.query_params.get("query", "")
-    algorithm = request.query_params.get("algorithm", "hybrid")
+    algorithm = request.query_params.get("algorithm", "bm25_hybrid")
     limit = int(request.query_params.get("limit", "50"))
-    score_threshold = float(request.query_params.get("score_threshold", "0.7"))
-    semantic_weight = float(request.query_params.get("semantic_weight", "0.5"))
-    keyword_weight = float(request.query_params.get("keyword_weight", "0.3"))
-    fuzzy_weight = float(request.query_params.get("fuzzy_weight", "0.2"))
+    score_threshold = float(request.query_params.get("score_threshold", "0.0"))
+    fusion = request.query_params.get("fusion", "rrf")  # Default to RRF
 
     # Parse doc_types (comma-separated list, None = all types)
     doc_types_param = request.query_params.get("doc_types", "")
@@ -378,7 +127,7 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
 
     logger.info(
         f"Viz search: user={username}, query='{query}', "
-        f"algorithm={algorithm}, limit={limit}, doc_types={doc_types}"
+        f"algorithm={algorithm}, fusion={fusion}, limit={limit}, doc_types={doc_types}"
     )
 
     try:
@@ -391,19 +140,16 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
             _get_authenticated_client_for_userinfo,
         )
 
-        async with await _get_authenticated_client_for_userinfo(request) as http_client:  # noqa: F841
+        with trace_operation("vector_viz.get_auth_client"):
+            auth_client_ctx = await _get_authenticated_client_for_userinfo(request)
+
+        async with auth_client_ctx as nc_client:  # noqa: F841
             # Create search algorithm (no client needed - verification removed)
             if algorithm == "semantic":
                 search_algo = SemanticSearchAlgorithm(score_threshold=score_threshold)
-            elif algorithm == "keyword":
-                search_algo = KeywordSearchAlgorithm()
-            elif algorithm == "fuzzy":
-                search_algo = FuzzySearchAlgorithm()
-            elif algorithm == "hybrid":
-                search_algo = HybridSearchAlgorithm(
-                    semantic_weight=semantic_weight,
-                    keyword_weight=keyword_weight,
-                    fuzzy_weight=fuzzy_weight,
+            elif algorithm == "bm25_hybrid":
+                search_algo = BM25HybridSearchAlgorithm(
+                    score_threshold=score_threshold, fusion=fusion
                 )
             else:
                 return JSONResponse(
@@ -417,24 +163,40 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
             all_results = []
             if doc_types is None or len(doc_types) == 0:
                 # Cross-app search - search all indexed types
-                unverified_results = await search_algo.search(
-                    query=query,
-                    user_id=username,
-                    limit=limit * 2,  # Buffer for verification filtering
-                    doc_type=None,  # Search all types
-                    score_threshold=score_threshold,
-                )
-                all_results.extend(unverified_results)
-            else:
-                # Search each document type and combine
-                for doc_type in doc_types:
+                with trace_operation(
+                    "vector_viz.search_execute",
+                    attributes={
+                        "search.algorithm": algorithm,
+                        "search.limit": limit * 2,
+                        "search.doc_type": "all",
+                    },
+                ):
                     unverified_results = await search_algo.search(
                         query=query,
                         user_id=username,
                         limit=limit * 2,  # Buffer for verification filtering
-                        doc_type=doc_type,
+                        doc_type=None,  # Search all types
                         score_threshold=score_threshold,
                     )
+                all_results.extend(unverified_results)
+            else:
+                # Search each document type and combine
+                for doc_type in doc_types:
+                    with trace_operation(
+                        "vector_viz.search_execute",
+                        attributes={
+                            "search.algorithm": algorithm,
+                            "search.limit": limit * 2,
+                            "search.doc_type": doc_type,
+                        },
+                    ):
+                        unverified_results = await search_algo.search(
+                            query=query,
+                            user_id=username,
+                            limit=limit * 2,  # Buffer for verification filtering
+                            doc_type=doc_type,
+                            score_threshold=score_threshold,
+                        )
                     all_results.extend(unverified_results)
                 # Sort by score before verification
                 all_results.sort(key=lambda r: r.score, reverse=True)
@@ -445,78 +207,87 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
             search_results = all_results[:limit]
             search_duration = time.perf_counter() - search_start
 
-        # Normalize scores relative to this result set for better visualization
+        # Store original scores and normalize for visualization
         # (best result = 1.0, worst result = 0.0 within THIS result set)
         # This makes visual encoding meaningful regardless of RRF normalization
-        if search_results:
-            scores = [r.score for r in search_results]
-            min_score, max_score = min(scores), max(scores)
-            score_range = max_score - min_score if max_score > min_score else 1.0
+        with trace_operation(
+            "vector_viz.score_normalize",
+            attributes={"normalize.num_results": len(search_results)},
+        ):
+            if search_results:
+                scores = [r.score for r in search_results]
+                min_score, max_score = min(scores), max(scores)
+                score_range = max_score - min_score if max_score > min_score else 1.0
 
-            logger.info(
-                f"Normalizing scores for viz: original range [{min_score:.3f}, {max_score:.3f}] "
-                f"→ [0.0, 1.0]"
-            )
+                logger.info(
+                    f"Normalizing scores for viz: original range [{min_score:.3f}, {max_score:.3f}] "
+                    f"→ [0.0, 1.0]"
+                )
 
-            # Rescale each result's score to 0-1 within this result set
-            for r in search_results:
-                r.score = (r.score - min_score) / score_range
+                # Store original score and rescale to 0-1 for visualization
+                for r in search_results:
+                    # Store original score before normalization
+                    r.original_score = r.score
+                    # Rescale for visual encoding
+                    r.score = (r.score - min_score) / score_range
 
         if not search_results:
             return JSONResponse(
                 {
                     "success": True,
                     "results": [],
-                    "coordinates_2d": [],
+                    "coordinates_3d": [],
+                    "query_coords": [],
                     "message": "No results found",
                 }
             )
 
-        # Fetch vectors for matching results from Qdrant
+        # Fetch vectors for specific matching chunks from Qdrant using batch retrieve
         vector_fetch_start = time.perf_counter()
-        qdrant_client = await get_qdrant_client()
-        doc_ids = [r.id for r in search_results]
 
-        # Retrieve vectors for the matching documents
-        from qdrant_client.models import FieldCondition, Filter, MatchAny
+        with trace_operation("vector_viz.get_qdrant_client"):
+            qdrant_client = await get_qdrant_client()
 
-        points_response = await qdrant_client.scroll(
-            collection_name=settings.get_collection_name(),
-            scroll_filter=Filter(
-                must=[
-                    FieldCondition(
-                        key="doc_id",
-                        match=MatchAny(any=[str(doc_id) for doc_id in doc_ids]),
-                    ),
-                    FieldCondition(
-                        key="user_id",
-                        match={"value": username},
-                    ),
-                ]
-            ),
-            limit=len(doc_ids) * 2,  # Account for multiple chunks per doc
-            with_vectors=True,
-            with_payload=["doc_id"],  # Need doc_id to map vectors to results
-        )
+        chunk_vectors_map = {}  # Map (doc_id, chunk_start, chunk_end) -> vector
 
-        points = points_response[0]
+        # Collect point IDs from search results for batch retrieval
+        # point_id is the Qdrant internal ID returned by search algorithms
+        point_ids = [r.point_id for r in search_results if r.point_id]
 
-        if not points:
-            return JSONResponse(
-                {
-                    "success": True,
-                    "results": [],
-                    "coordinates_2d": [],
-                    "message": "No vectors found for results",
-                }
-            )
+        if point_ids:
+            # Single batch retrieve call instead of N sequential scroll calls
+            # This is ~50x faster for 50 results (1 HTTP request vs 50)
+            with trace_operation(
+                "vector_viz.vector_retrieve",
+                attributes={"retrieve.num_points": len(point_ids)},
+            ):
+                points_response = await qdrant_client.retrieve(
+                    collection_name=settings.get_collection_name(),
+                    ids=point_ids,
+                    with_vectors=["dense"],
+                    with_payload=["doc_id", "chunk_start_offset", "chunk_end_offset"],
+                )
+
+            # Build chunk_vectors_map from batch response
+            for point in points_response:
+                if point.vector is not None:
+                    # Extract dense vector (handle both named and unnamed vectors)
+                    if isinstance(point.vector, dict):
+                        vector = point.vector.get("dense")
+                    else:
+                        vector = point.vector
+
+                    if vector is not None and point.payload:
+                        doc_id = point.payload.get("doc_id")
+                        chunk_start = point.payload.get("chunk_start_offset")
+                        chunk_end = point.payload.get("chunk_end_offset")
+                        chunk_key = (doc_id, chunk_start, chunk_end)
+                        chunk_vectors_map[chunk_key] = vector
 
-        # Extract vectors
-        vectors = np.array([p.vector for p in points if p.vector is not None])
         vector_fetch_duration = time.perf_counter() - vector_fetch_start
 
-        if len(vectors) < 2:
-            # Not enough points for PCA
+        if len(chunk_vectors_map) < 2:
+            # Not enough chunks for PCA
             return JSONResponse(
                 {
                     "success": True,
@@ -530,35 +301,149 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
                         }
                         for r in search_results
                     ],
-                    "coordinates_2d": [[0, 0]] * len(search_results),
-                    "message": "Not enough vectors for PCA",
+                    "coordinates_3d": [[0, 0, 0]] * len(search_results),
+                    "query_coords": [0, 0, 0],
+                    "message": "Not enough chunks for PCA",
                 }
             )
 
-        # Apply PCA dimensionality reduction (768-dim → 2D)
+        # Detect embedding dimension from first available vector
+        embedding_dim = None
+        for vector in chunk_vectors_map.values():
+            if vector is not None:
+                embedding_dim = len(vector)
+                break
+
+        if embedding_dim is None:
+            return JSONResponse(
+                {
+                    "success": False,
+                    "error": "Could not determine embedding dimension",
+                },
+                status_code=500,
+            )
+
+        logger.info(f"Detected embedding dimension: {embedding_dim}")
+
+        # Build chunk vectors array in search_results order (1:1 mapping)
+        chunk_vectors = []
+        for result in search_results:
+            chunk_key = (result.id, result.chunk_start_offset, result.chunk_end_offset)
+            if chunk_key in chunk_vectors_map:
+                chunk_vectors.append(chunk_vectors_map[chunk_key])
+            else:
+                # Chunk not found in vectors (shouldn't happen)
+                logger.warning(
+                    f"Chunk {chunk_key} not found in fetched vectors, using zero vector"
+                )
+                # Use zero vector as fallback
+                chunk_vectors.append(np.zeros(embedding_dim))
+
+        chunk_vectors = np.array(chunk_vectors)
+
+        # Reuse query embedding from search algorithm (avoids redundant embedding call)
+        query_embed_start = time.perf_counter()
+        if search_algo.query_embedding is not None:
+            query_embedding = search_algo.query_embedding
+            logger.info(
+                f"Reusing query embedding from search algorithm "
+                f"(dimension={len(query_embedding)})"
+            )
+        else:
+            # Fallback: generate embedding if not available from search
+            from nextcloud_mcp_server.embedding.service import get_embedding_service
+
+            embedding_service = get_embedding_service()
+            query_embedding = await embedding_service.embed(query)
+            logger.info(f"Generated query embedding (dimension={len(query_embedding)})")
+        query_embed_duration = time.perf_counter() - query_embed_start
+
+        # Combine query vector with chunk vectors for PCA
+        # Query will be the last point in the array
+        all_vectors = np.vstack([chunk_vectors, np.array([query_embedding])])
+
+        # Normalize vectors to unit length (L2 normalization)
+        # This is critical because Qdrant uses COSINE distance, which only measures
+        # vector direction (angle), not magnitude. PCA uses Euclidean distance which
+        # considers both direction and magnitude. By normalizing to unit length,
+        # Euclidean distances in PCA space will match cosine distances.
+        norms = np.linalg.norm(all_vectors, axis=1, keepdims=True)
+
+        # Check for zero-norm vectors (can happen with empty/corrupted embeddings)
+        zero_norm_mask = norms[:, 0] < 1e-10
+        if zero_norm_mask.any():
+            zero_indices = np.where(zero_norm_mask)[0]
+            logger.warning(
+                f"Found {zero_norm_mask.sum()} zero-norm vectors at indices {zero_indices.tolist()}. "
+                "Replacing with small epsilon to avoid division by zero."
+            )
+            # Replace zero norms with small epsilon to avoid NaN
+            norms[zero_norm_mask] = 1e-10
+
+        all_vectors_normalized = all_vectors / norms
+        logger.info(
+            f"Normalized vectors: query_norm={norms[-1][0]:.3f}, "
+            f"doc_norm_range=[{norms[:-1].min():.3f}, {norms[:-1].max():.3f}]"
+        )
+
+        # Apply PCA dimensionality reduction (768-dim → 3D) on normalized vectors
+        # Run in thread pool to avoid blocking the event loop (CPU-bound)
         pca_start = time.perf_counter()
-        pca = PCA(n_components=2)
-        coords_2d = pca.fit_transform(vectors)
+
+        def _compute_pca(vectors: np.ndarray) -> tuple[np.ndarray, PCA]:
+            pca = PCA(n_components=3)
+            coords = pca.fit_transform(vectors)
+            return coords, pca
+
+        import anyio
+
+        with trace_operation(
+            "vector_viz.pca_compute",
+            attributes={
+                "pca.num_vectors": len(all_vectors_normalized),
+                "pca.embedding_dim": embedding_dim,
+            },
+        ):
+            coords_3d, pca = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+                lambda: _compute_pca(all_vectors_normalized)
+            )
         pca_duration = time.perf_counter() - pca_start
 
         # After fit, these attributes are guaranteed to be set
         assert pca.explained_variance_ratio_ is not None
 
+        # Check for NaN values in PCA output (numerical instability)
+        nan_mask = np.isnan(coords_3d)
+        if nan_mask.any():
+            nan_rows = np.where(nan_mask.any(axis=1))[0]
+            logger.error(
+                f"Found NaN values in PCA output at {len(nan_rows)} points: {nan_rows.tolist()[:10]}. "
+                "Replacing NaN with 0.0 to prevent JSON serialization error."
+            )
+            # Replace NaN with 0 to allow JSON serialization
+            coords_3d = np.nan_to_num(coords_3d, nan=0.0)
+
+        # Split query coords from chunk coords
+        # Round to 2 decimal places for cleaner display
+        query_coords_3d = [
+            round(float(x), 2) for x in coords_3d[-1]
+        ]  # Last point is query
+        chunk_coords_3d = coords_3d[:-1]  # All but last are chunks
+
         logger.info(
             f"PCA explained variance: PC1={pca.explained_variance_ratio_[0]:.3f}, "
-            f"PC2={pca.explained_variance_ratio_[1]:.3f}"
+            f"PC2={pca.explained_variance_ratio_[1]:.3f}, "
+            f"PC3={pca.explained_variance_ratio_[2]:.3f}"
+        )
+        logger.info(
+            f"Embedding stats: chunks={len(chunk_vectors)}, "
+            f"query_dim={len(query_embedding)}, chunk_vector_dim={chunk_vectors.shape[1] if chunk_vectors.size > 0 else 0}"
         )
 
-        # Map results to coordinates (use first chunk per document)
-        result_coords = []
-        seen_doc_ids = set()
-
-        for point, coord in zip(points, coords_2d):
-            if point.payload:
-                doc_id = int(point.payload.get("doc_id", 0))
-                if doc_id not in seen_doc_ids and doc_id in doc_ids:
-                    seen_doc_ids.add(doc_id)
-                    result_coords.append(coord.tolist())
+        # Coordinates already match search_results order (1:1 mapping)
+        result_coords = [
+            [round(float(x), 2) for x in coord] for coord in chunk_coords_3d
+        ]
 
         # Build response
         response_results = [
@@ -567,7 +452,12 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
                 "doc_type": r.doc_type,
                 "title": r.title,
                 "excerpt": r.excerpt,
-                "score": r.score,
+                "score": r.score,  # Normalized score for visual encoding (0-1)
+                "original_score": getattr(
+                    r, "original_score", r.score
+                ),  # Raw score from algorithm
+                "chunk_start_offset": r.chunk_start_offset,
+                "chunk_end_offset": r.chunk_end_offset,
             }
             for r in search_results
         ]
@@ -580,26 +470,30 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
             f"Viz search timing: total={total_duration * 1000:.1f}ms, "
             f"search={search_duration * 1000:.1f}ms ({search_duration / total_duration * 100:.1f}%), "
             f"vector_fetch={vector_fetch_duration * 1000:.1f}ms ({vector_fetch_duration / total_duration * 100:.1f}%), "
+            f"query_embed={query_embed_duration * 1000:.1f}ms ({query_embed_duration / total_duration * 100:.1f}%), "
             f"pca={pca_duration * 1000:.1f}ms ({pca_duration / total_duration * 100:.1f}%), "
-            f"results={len(search_results)}, vectors={len(vectors)}"
+            f"results={len(search_results)}, chunk_vectors={len(chunk_vectors)}"
         )
 
         return JSONResponse(
             {
                 "success": True,
                 "results": response_results,
-                "coordinates_2d": result_coords[: len(search_results)],
+                "coordinates_3d": result_coords[: len(search_results)],
+                "query_coords": query_coords_3d,
                 "pca_variance": {
                     "pc1": float(pca.explained_variance_ratio_[0]),
                     "pc2": float(pca.explained_variance_ratio_[1]),
+                    "pc3": float(pca.explained_variance_ratio_[2]),
                 },
                 "timing": {
                     "total_ms": round(total_duration * 1000, 2),
                     "search_ms": round(search_duration * 1000, 2),
                     "vector_fetch_ms": round(vector_fetch_duration * 1000, 2),
+                    "query_embed_ms": round(query_embed_duration * 1000, 2),
                     "pca_ms": round(pca_duration * 1000, 2),
                     "num_results": len(search_results),
-                    "num_vectors": len(vectors),
+                    "num_chunk_vectors": len(chunk_vectors),
                 },
             }
         )
@@ -610,3 +504,166 @@ async def vector_visualization_search(request: Request) -> JSONResponse:
             {"success": False, "error": str(e)},
             status_code=500,
         )
+
+
+@requires("authenticated", redirect="oauth_login")
+async def chunk_context_endpoint(request: Request) -> JSONResponse:
+    """Fetch chunk text with surrounding context for visualization.
+
+    This endpoint retrieves the matched chunk along with surrounding text
+    to provide context for the search result. Used by the viz pane to
+    display chunks inline.
+
+    Query parameters:
+        doc_type: Document type (e.g., "note")
+        doc_id: Document ID
+        start: Chunk start offset (character position)
+        end: Chunk end offset (character position)
+        context: Characters of context before/after (default: 500)
+
+    Returns:
+        JSON with chunk_text, before_context, after_context, and flags
+    """
+    try:
+        # Get query parameters
+        doc_type = request.query_params.get("doc_type")
+        doc_id = request.query_params.get("doc_id")
+        start_str = request.query_params.get("start")
+        end_str = request.query_params.get("end")
+        context_chars = int(request.query_params.get("context", "500"))
+
+        # Validate required parameters
+        if not all([doc_type, doc_id, start_str, end_str]):
+            return JSONResponse(
+                {
+                    "success": False,
+                    "error": "Missing required parameters: doc_type, doc_id, start, end",
+                },
+                status_code=400,
+            )
+
+        # Type assertions - we validated these above
+        assert doc_type is not None
+        assert doc_id is not None
+        assert start_str is not None
+        assert end_str is not None
+
+        start = int(start_str)
+        end = int(end_str)
+        # Convert doc_id to int (all document types use int IDs)
+        doc_id_int = int(doc_id)
+
+        # Get authenticated Nextcloud client
+        from nextcloud_mcp_server.auth.userinfo_routes import (
+            _get_authenticated_client_for_userinfo,
+        )
+        from nextcloud_mcp_server.search.context import get_chunk_with_context
+
+        # Use context expansion module to fetch chunk with surrounding context
+        async with await _get_authenticated_client_for_userinfo(request) as nc_client:
+            chunk_context = await get_chunk_with_context(
+                nc_client=nc_client,
+                user_id=request.user.display_name,  # User ID from auth
+                doc_id=doc_id_int,
+                doc_type=doc_type,
+                chunk_start=start,
+                chunk_end=end,
+                context_chars=context_chars,
+            )
+
+        # Check if context expansion succeeded
+        if chunk_context is None:
+            return JSONResponse(
+                {
+                    "success": False,
+                    "error": f"Failed to fetch chunk context for {doc_type} {doc_id}",
+                },
+                status_code=404,
+            )
+
+        logger.info(
+            f"Fetched chunk context for {doc_type}_{doc_id}: "
+            f"chunk_len={len(chunk_context.chunk_text)}, "
+            f"before_len={len(chunk_context.before_context)}, "
+            f"after_len={len(chunk_context.after_context)}"
+        )
+
+        # For PDF files, also fetch the highlighted page image from Qdrant
+        highlighted_page_image = None
+        page_number = None
+        if doc_type == "file":
+            try:
+                from qdrant_client.models import FieldCondition, Filter, MatchValue
+
+                settings = get_settings()
+                qdrant_client = await get_qdrant_client()
+                username = request.user.display_name
+
+                # Query for this specific chunk's highlighted image
+                points_response = await qdrant_client.scroll(
+                    collection_name=settings.get_collection_name(),
+                    scroll_filter=Filter(
+                        must=[
+                            get_placeholder_filter(),
+                            FieldCondition(
+                                key="doc_id", match=MatchValue(value=doc_id_int)
+                            ),
+                            FieldCondition(
+                                key="user_id", match=MatchValue(value=username)
+                            ),
+                            FieldCondition(
+                                key="chunk_start_offset", match=MatchValue(value=start)
+                            ),
+                            FieldCondition(
+                                key="chunk_end_offset", match=MatchValue(value=end)
+                            ),
+                        ]
+                    ),
+                    limit=1,
+                    with_vectors=False,
+                    with_payload=["highlighted_page_image", "page_number"],
+                )
+
+                points = points_response[0]
+                if points and points[0].payload:
+                    highlighted_page_image = points[0].payload.get(
+                        "highlighted_page_image"
+                    )
+                    page_number = points[0].payload.get("page_number")
+                    if highlighted_page_image:
+                        logger.info(
+                            f"Found highlighted image for chunk: "
+                            f"page={page_number}, image_size={len(highlighted_page_image)}"
+                        )
+            except Exception as e:
+                logger.warning(f"Failed to fetch highlighted image: {e}")
+
+        # Return response compatible with frontend expectations
+        response_data: dict = {
+            "success": True,
+            "chunk_text": chunk_context.chunk_text,
+            "before_context": chunk_context.before_context,
+            "after_context": chunk_context.after_context,
+            "has_more_before": chunk_context.has_before_truncation,
+            "has_more_after": chunk_context.has_after_truncation,
+        }
+
+        # Add image data if available
+        if highlighted_page_image:
+            response_data["highlighted_page_image"] = highlighted_page_image
+            response_data["page_number"] = page_number
+
+        return JSONResponse(response_data)
+
+    except ValueError as e:
+        logger.error(f"Invalid parameter format: {e}")
+        return JSONResponse(
+            {"success": False, "error": f"Invalid parameter format: {e}"},
+            status_code=400,
+        )
+    except Exception as e:
+        logger.error(f"Chunk context error: {e}", exc_info=True)
+        return JSONResponse(
+            {"success": False, "error": str(e)},
+            status_code=500,
+        )

nextcloud_mcp_server/auth/webhook_routes.py

@@ -139,6 +139,7 @@ async def _get_authenticated_client(request: Request) -> httpx.AsyncClient:
             raise RuntimeError("BasicAuth credentials not configured")
 
         assert nextcloud_host is not None  # Type narrowing for type checker
+        assert username is not None and password is not None  # Type narrowing
         return httpx.AsyncClient(
             base_url=nextcloud_host,
             auth=(username, password),

nextcloud_mcp_server/cli.py

@@ -29,9 +29,9 @@ from .app import get_app
 @click.option(
     "--transport",
     "-t",
-    default="sse",
+    default="streamable-http",
     show_default=True,
-    type=click.Choice(["sse", "streamable-http", "http"]),
+    type=click.Choice(["streamable-http", "http"]),
     help="MCP transport protocol",
 )
 @click.option(

nextcloud_mcp_server/client/__init__.py

@@ -18,6 +18,7 @@ from .contacts import ContactsClient
 from .cookbook import CookbookClient
 from .deck import DeckClient
 from .groups import GroupsClient
+from .news import NewsClient
 from .notes import NotesClient
 from .sharing import SharingClient
 from .tables import TablesClient
@@ -81,6 +82,7 @@ class NextcloudClient:
         self.contacts = ContactsClient(self._client, username)
         self.cookbook = CookbookClient(self._client, username)
         self.deck = DeckClient(self._client, username)
+        self.news = NewsClient(self._client, username)
         self.users = UsersClient(self._client, username)
         self.groups = GroupsClient(self._client, username)
         self.sharing = SharingClient(self._client, username)
@@ -130,10 +132,75 @@ class NextcloudClient:
         all_notes = self.notes.get_all_notes()
         return await self._notes_search.search_notes(all_notes, query)
 
+    async def find_files_by_tag(
+        self, tag_name: str, mime_type_filter: str | None = None
+    ) -> list[dict]:
+        """Find files by system tag name, optionally filtered by MIME type.
+
+        This method coordinates tag lookup and file retrieval via WebDAV:
+        1. Look up the tag ID by name
+        2. Get all files with that tag (via REPORT with full metadata)
+        3. Optionally filter by MIME type
+
+        Args:
+            tag_name: Name of the system tag to search for (e.g., "vector-index")
+            mime_type_filter: Optional MIME type filter (e.g., "application/pdf")
+
+        Returns:
+            List of file dictionaries with WebDAV properties (path, size, content_type, etc.)
+
+        Raises:
+            RuntimeError: If tag lookup or file query fails
+
+        Examples:
+            # Find all files with "vector-index" tag
+            files = await nc_client.find_files_by_tag("vector-index")
+
+            # Find only PDFs with the tag
+            pdfs = await nc_client.find_files_by_tag("vector-index", "application/pdf")
+        """
+        # Look up tag by name using WebDAV
+        tag = await self.webdav.get_tag_by_name(tag_name)
+        if not tag:
+            logger.debug(f"Tag '{tag_name}' not found, returning empty list")
+            return []
+
+        # Get files with this tag (returns full file info from REPORT)
+        files = await self.webdav.get_files_by_tag(tag["id"])
+        if not files:
+            logger.debug(f"No files found with tag '{tag_name}'")
+            return []
+
+        logger.debug(f"Found {len(files)} files with tag '{tag_name}'")
+
+        # Apply MIME type filter if specified
+        if mime_type_filter:
+            filtered_files = [
+                f
+                for f in files
+                if f.get("content_type", "").startswith(mime_type_filter)
+            ]
+            logger.info(
+                f"Returning {len(filtered_files)} files with tag '{tag_name}' (filtered by {mime_type_filter})"
+            )
+            return filtered_files
+
+        logger.info(f"Returning {len(files)} files with tag '{tag_name}'")
+        return files
+
     def _get_webdav_base_path(self) -> str:
         """Helper to get the base WebDAV path for the authenticated user."""
         return f"/remote.php/dav/files/{self.username}"
 
+    async def __aenter__(self):
+        """Async context manager entry."""
+        return self
+
+    async def __aexit__(self, exc_type, exc_val, exc_tb):
+        """Async context manager exit - closes all clients."""
+        await self.close()
+        return False  # Don't suppress exceptions
+
     async def close(self):
         """Close the HTTP client and CalDAV client."""
         await self._client.aclose()

nextcloud_mcp_server/client/news.py

@@ -0,0 +1,385 @@
+"""Client for Nextcloud News app operations."""
+
+import logging
+from enum import IntEnum
+from typing import Any
+
+from .base import BaseNextcloudClient
+
+logger = logging.getLogger(__name__)
+
+
+class NewsItemType(IntEnum):
+    """Type constants for News API item queries."""
+
+    FEED = 0  # Single feed
+    FOLDER = 1  # Folder and its feeds
+    STARRED = 2  # All starred items
+    ALL = 3  # All items
+
+
+class NewsClient(BaseNextcloudClient):
+    """Client for Nextcloud News app operations."""
+
+    app_name = "news"
+    API_BASE = "/apps/news/api/v1-3"
+
+    # --- Folders ---
+
+    async def get_folders(self) -> list[dict[str, Any]]:
+        """Get all folders."""
+        response = await self._make_request("GET", f"{self.API_BASE}/folders")
+        return response.json().get("folders", [])
+
+    async def create_folder(self, name: str) -> dict[str, Any]:
+        """Create a new folder.
+
+        Args:
+            name: Folder name
+
+        Returns:
+            Created folder data
+
+        Raises:
+            HTTPStatusError: 409 if folder name already exists,
+                            422 if name is empty
+        """
+        response = await self._make_request(
+            "POST", f"{self.API_BASE}/folders", json={"name": name}
+        )
+        folders = response.json().get("folders", [])
+        return folders[0] if folders else {}
+
+    async def rename_folder(self, folder_id: int, name: str) -> None:
+        """Rename a folder.
+
+        Args:
+            folder_id: Folder ID
+            name: New folder name
+
+        Raises:
+            HTTPStatusError: 404 if folder not found, 409 if name exists
+        """
+        await self._make_request(
+            "PUT", f"{self.API_BASE}/folders/{folder_id}", json={"name": name}
+        )
+
+    async def delete_folder(self, folder_id: int) -> None:
+        """Delete a folder and all its feeds/items.
+
+        Args:
+            folder_id: Folder ID
+
+        Raises:
+            HTTPStatusError: 404 if folder not found
+        """
+        await self._make_request("DELETE", f"{self.API_BASE}/folders/{folder_id}")
+
+    async def mark_folder_read(self, folder_id: int, newest_item_id: int) -> None:
+        """Mark all items in a folder as read.
+
+        Args:
+            folder_id: Folder ID
+            newest_item_id: ID of newest item to mark read (prevents marking
+                           items user hasn't seen yet)
+
+        Raises:
+            HTTPStatusError: 404 if folder not found
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/folders/{folder_id}/read",
+            json={"newestItemId": newest_item_id},
+        )
+
+    # --- Feeds ---
+
+    async def get_feeds(self) -> dict[str, Any]:
+        """Get all feeds with metadata.
+
+        Returns:
+            Dict with keys:
+                - feeds: List of feed objects
+                - starredCount: Number of starred items
+                - newestItemId: ID of newest item (omitted if no items)
+        """
+        response = await self._make_request("GET", f"{self.API_BASE}/feeds")
+        return response.json()
+
+    async def create_feed(
+        self, url: str, folder_id: int | None = None
+    ) -> dict[str, Any]:
+        """Subscribe to a new feed.
+
+        Args:
+            url: Feed URL
+            folder_id: Optional folder ID (None for root)
+
+        Returns:
+            Created feed data
+
+        Raises:
+            HTTPStatusError: 409 if feed already exists, 422 if URL is invalid
+        """
+        body: dict[str, Any] = {"url": url}
+        if folder_id is not None:
+            body["folderId"] = folder_id
+        response = await self._make_request("POST", f"{self.API_BASE}/feeds", json=body)
+        data = response.json()
+        feeds = data.get("feeds", [])
+        return feeds[0] if feeds else {}
+
+    async def delete_feed(self, feed_id: int) -> None:
+        """Unsubscribe from a feed (deletes all items).
+
+        Args:
+            feed_id: Feed ID
+
+        Raises:
+            HTTPStatusError: 404 if feed not found
+        """
+        await self._make_request("DELETE", f"{self.API_BASE}/feeds/{feed_id}")
+
+    async def move_feed(self, feed_id: int, folder_id: int | None) -> None:
+        """Move a feed to a different folder.
+
+        Args:
+            feed_id: Feed ID
+            folder_id: Target folder ID (None for root)
+
+        Raises:
+            HTTPStatusError: 404 if feed not found
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/feeds/{feed_id}/move",
+            json={"folderId": folder_id},
+        )
+
+    async def rename_feed(self, feed_id: int, title: str) -> None:
+        """Rename a feed.
+
+        Args:
+            feed_id: Feed ID
+            title: New feed title
+
+        Raises:
+            HTTPStatusError: 404 if feed not found
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/feeds/{feed_id}/rename",
+            json={"feedTitle": title},
+        )
+
+    async def mark_feed_read(self, feed_id: int, newest_item_id: int) -> None:
+        """Mark all items in a feed as read.
+
+        Args:
+            feed_id: Feed ID
+            newest_item_id: ID of newest item to mark read
+
+        Raises:
+            HTTPStatusError: 404 if feed not found
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/feeds/{feed_id}/read",
+            json={"newestItemId": newest_item_id},
+        )
+
+    # --- Items ---
+
+    async def get_items(
+        self,
+        batch_size: int = 50,
+        offset: int = 0,
+        type_: int = NewsItemType.ALL,
+        id_: int = 0,
+        get_read: bool = True,
+        oldest_first: bool = False,
+    ) -> list[dict[str, Any]]:
+        """Get items (articles) with filtering.
+
+        Args:
+            batch_size: Number of items to return (-1 for all)
+            offset: Item ID to start after (for pagination)
+            type_: Item type filter (NewsItemType)
+            id_: Feed/folder ID (ignored for STARRED/ALL types)
+            get_read: Include read items
+            oldest_first: Sort oldest first instead of newest
+
+        Returns:
+            List of item objects
+        """
+        params: dict[str, Any] = {
+            "batchSize": batch_size,
+            "offset": offset,
+            "type": type_,
+            "id": id_,
+            "getRead": str(get_read).lower(),
+            "oldestFirst": str(oldest_first).lower(),
+        }
+        response = await self._make_request(
+            "GET", f"{self.API_BASE}/items", params=params
+        )
+        return response.json().get("items", [])
+
+    async def get_item(self, item_id: int) -> dict[str, Any]:
+        """Get a specific item by ID.
+
+        Args:
+            item_id: Item ID
+
+        Returns:
+            Item data
+
+        Raises:
+            HTTPStatusError: 404 if item not found
+        """
+        response = await self._make_request("GET", f"{self.API_BASE}/items/{item_id}")
+        return response.json()
+
+    async def get_updated_items(
+        self,
+        last_modified: int,
+        type_: int = NewsItemType.ALL,
+        id_: int = 0,
+    ) -> list[dict[str, Any]]:
+        """Get items modified since a timestamp (for delta sync).
+
+        Args:
+            last_modified: Unix timestamp (seconds or microseconds)
+            type_: Item type filter
+            id_: Feed/folder ID
+
+        Returns:
+            List of modified items (includes deleted items)
+        """
+        params: dict[str, Any] = {
+            "lastModified": last_modified,
+            "type": type_,
+            "id": id_,
+        }
+        response = await self._make_request(
+            "GET", f"{self.API_BASE}/items/updated", params=params
+        )
+        return response.json().get("items", [])
+
+    async def mark_item_read(self, item_id: int) -> None:
+        """Mark a single item as read.
+
+        Args:
+            item_id: Item ID
+
+        Raises:
+            HTTPStatusError: 404 if item not found
+        """
+        await self._make_request("POST", f"{self.API_BASE}/items/{item_id}/read")
+
+    async def mark_item_unread(self, item_id: int) -> None:
+        """Mark a single item as unread.
+
+        Args:
+            item_id: Item ID
+
+        Raises:
+            HTTPStatusError: 404 if item not found
+        """
+        await self._make_request("POST", f"{self.API_BASE}/items/{item_id}/unread")
+
+    async def star_item(self, item_id: int) -> None:
+        """Star (favorite) a single item.
+
+        Args:
+            item_id: Item ID
+
+        Raises:
+            HTTPStatusError: 404 if item not found
+        """
+        await self._make_request("POST", f"{self.API_BASE}/items/{item_id}/star")
+
+    async def unstar_item(self, item_id: int) -> None:
+        """Unstar a single item.
+
+        Args:
+            item_id: Item ID
+
+        Raises:
+            HTTPStatusError: 404 if item not found
+        """
+        await self._make_request("POST", f"{self.API_BASE}/items/{item_id}/unstar")
+
+    async def mark_items_read(self, item_ids: list[int]) -> None:
+        """Mark multiple items as read.
+
+        Args:
+            item_ids: List of item IDs
+        """
+        await self._make_request(
+            "POST", f"{self.API_BASE}/items/read/multiple", json={"itemIds": item_ids}
+        )
+
+    async def mark_items_unread(self, item_ids: list[int]) -> None:
+        """Mark multiple items as unread.
+
+        Args:
+            item_ids: List of item IDs
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/items/unread/multiple",
+            json={"itemIds": item_ids},
+        )
+
+    async def star_items(self, item_ids: list[int]) -> None:
+        """Star multiple items.
+
+        Args:
+            item_ids: List of item IDs
+        """
+        await self._make_request(
+            "POST", f"{self.API_BASE}/items/star/multiple", json={"itemIds": item_ids}
+        )
+
+    async def unstar_items(self, item_ids: list[int]) -> None:
+        """Unstar multiple items.
+
+        Args:
+            item_ids: List of item IDs
+        """
+        await self._make_request(
+            "POST",
+            f"{self.API_BASE}/items/unstar/multiple",
+            json={"itemIds": item_ids},
+        )
+
+    async def mark_all_read(self, newest_item_id: int) -> None:
+        """Mark all items as read.
+
+        Args:
+            newest_item_id: ID of newest item to mark read
+        """
+        await self._make_request(
+            "POST", f"{self.API_BASE}/items/read", json={"newestItemId": newest_item_id}
+        )
+
+    # --- Status ---
+
+    async def get_status(self) -> dict[str, Any]:
+        """Get News app status and configuration.
+
+        Returns:
+            Dict with version and warnings
+        """
+        response = await self._make_request("GET", f"{self.API_BASE}/status")
+        return response.json()
+
+    async def get_version(self) -> str:
+        """Get News app version.
+
+        Returns:
+            Version string (e.g., "25.0.0")
+        """
+        response = await self._make_request("GET", f"{self.API_BASE}/version")
+        return response.json().get("version", "")

nextcloud_mcp_server/client/webdav.py

@@ -821,6 +821,20 @@ class WebDAVClient(BaseNextcloudClient):
                     item["file_id"] = int(value) if value else None
                 elif tag == "favorite":
                     item["is_favorite"] = value == "1"
+                elif tag == "tags":
+                    # Tags can be comma-separated or have multiple child elements
+                    if value:
+                        # Handle comma-separated tags
+                        item["tags"] = [
+                            t.strip() for t in value.split(",") if t.strip()
+                        ]
+                    else:
+                        # Check for child tag elements (alternative format)
+                        tag_elements = child.findall(".//{http://owncloud.org/ns}tag")
+                        if tag_elements:
+                            item["tags"] = [t.text for t in tag_elements if t.text]
+                        else:
+                            item["tags"] = []
                 elif tag == "permissions":
                     item["permissions"] = value
                 elif tag == "size":
@@ -948,3 +962,574 @@ class WebDAVClient(BaseNextcloudClient):
             properties=properties,
             limit=limit,
         )
+
+    async def find_by_tag(
+        self, tag_name: str, scope: str = "", limit: Optional[int] = None
+    ) -> List[Dict[str, Any]]:
+        """Find files by tag name.
+
+        DEPRECATED: Use NextcloudClient.find_files_by_tag() instead, which uses
+        the proper OCS Tags API rather than WebDAV SEARCH.
+
+        Args:
+            tag_name: Tag to filter by (e.g., "vector-index")
+            scope: Directory path to search in (empty string for user root)
+            limit: Maximum number of results to return
+
+        Returns:
+            List of files/directories with the specified tag
+
+        Examples:
+            # Find all files tagged with "vector-index"
+            results = await find_by_tag("vector-index")
+
+            # Find tagged files in a specific folder
+            results = await find_by_tag("vector-index", scope="Documents")
+        """
+        # Use LIKE for tag matching since tags can be comma-separated
+        where_conditions = f"""
+            <d:like>
+                <d:prop>
+                    <oc:tags/>
+                </d:prop>
+                <d:literal>%{tag_name}%</d:literal>
+            </d:like>
+        """
+
+        # Request tag property along with standard properties
+        properties = [
+            "displayname",
+            "getcontentlength",
+            "getcontenttype",
+            "getlastmodified",
+            "resourcetype",
+            "getetag",
+            "fileid",
+            "tags",
+        ]
+
+        return await self.search_files(
+            scope=scope,
+            where_conditions=where_conditions,
+            properties=properties,
+            limit=limit,
+        )
+
+    async def _get_file_info_by_id(self, file_id: int) -> Dict[str, Any]:
+        """Get file information by Nextcloud file ID using WebDAV.
+
+        Args:
+            file_id: Nextcloud internal file ID
+
+        Returns:
+            File information dictionary with path, size, content_type, etc.
+
+        Raises:
+            HTTPStatusError: If file not found or request fails
+        """
+        # Nextcloud allows accessing files by ID via special meta endpoint
+        meta_path = f"/remote.php/dav/meta/{file_id}/"
+
+        propfind_body = """<?xml version="1.0"?>
+        <d:propfind xmlns:d="DAV:" xmlns:oc="http://owncloud.org/ns">
+            <d:prop>
+                <d:displayname/>
+                <d:getcontentlength/>
+                <d:getcontenttype/>
+                <d:getlastmodified/>
+                <d:resourcetype/>
+                <d:getetag/>
+                <oc:fileid/>
+            </d:prop>
+        </d:propfind>"""
+
+        headers = {"Depth": "0", "Content-Type": "text/xml", "OCS-APIRequest": "true"}
+
+        response = await self._make_request(
+            "PROPFIND", meta_path, content=propfind_body, headers=headers
+        )
+        response.raise_for_status()
+
+        # Parse the XML response
+        root = ET.fromstring(response.content)
+        responses = root.findall(".//{DAV:}response")
+
+        if not responses:
+            raise RuntimeError(f"File ID {file_id} not found")
+
+        response_elem = responses[0]
+        href = response_elem.find(".//{DAV:}href")
+        if href is None:
+            raise RuntimeError(f"No href in response for file ID {file_id}")
+
+        propstat = response_elem.find(".//{DAV:}propstat")
+        if propstat is None:
+            raise RuntimeError(f"No propstat for file ID {file_id}")
+
+        prop = propstat.find(".//{DAV:}prop")
+        if prop is None:
+            raise RuntimeError(f"No prop for file ID {file_id}")
+
+        # Extract file path from displayname or construct from file ID
+        displayname_elem = prop.find(".//{DAV:}displayname")
+        name = (
+            displayname_elem.text if displayname_elem is not None else f"file_{file_id}"
+        )
+
+        # Get file properties
+        size_elem = prop.find(".//{DAV:}getcontentlength")
+        size = int(size_elem.text) if size_elem is not None and size_elem.text else 0
+
+        content_type_elem = prop.find(".//{DAV:}getcontenttype")
+        content_type = content_type_elem.text if content_type_elem is not None else None
+
+        modified_elem = prop.find(".//{DAV:}getlastmodified")
+        modified = modified_elem.text if modified_elem is not None else None
+
+        etag_elem = prop.find(".//{DAV:}getetag")
+        etag = (
+            etag_elem.text.strip('"')
+            if etag_elem is not None and etag_elem.text
+            else None
+        )
+
+        # Check if it's a directory
+        resourcetype = prop.find(".//{DAV:}resourcetype")
+        is_directory = (
+            resourcetype is not None
+            and resourcetype.find(".//{DAV:}collection") is not None
+        )
+
+        # Try to get actual file path - meta endpoint doesn't give us the real path
+        # so we'll construct a reasonable path from the name
+        # The calling code in NextcloudClient will have the context to determine the actual path
+        file_info = {
+            "name": name,
+            "path": f"/{name}",  # Placeholder - caller should use WebDAV to get real path if needed
+            "size": size,
+            "content_type": content_type,
+            "last_modified": modified,
+            "etag": etag,
+            "is_directory": is_directory,
+            "file_id": file_id,
+        }
+
+        logger.debug(f"Retrieved file info for ID {file_id}: {name}")
+        return file_info
+
+    async def get_tag_by_name(self, tag_name: str) -> dict[str, Any] | None:
+        """Get a system tag by its name via WebDAV.
+
+        Args:
+            tag_name: Name of the tag to find (case-sensitive)
+
+        Returns:
+            Tag dictionary if found, None otherwise
+        """
+        # Use WebDAV PROPFIND to list all systemtags
+        propfind_body = """<?xml version="1.0"?>
+<d:propfind xmlns:d="DAV:" xmlns:oc="http://owncloud.org/ns">
+  <d:prop>
+    <oc:id/>
+    <oc:display-name/>
+    <oc:user-visible/>
+    <oc:user-assignable/>
+  </d:prop>
+</d:propfind>"""
+
+        response = await self._client.request(
+            "PROPFIND",
+            "/remote.php/dav/systemtags/",
+            headers={"Depth": "1"},
+            content=propfind_body,
+        )
+        response.raise_for_status()
+
+        # Parse XML response
+        root = ET.fromstring(response.content)
+        ns = {
+            "d": "DAV:",
+            "oc": "http://owncloud.org/ns",
+        }
+
+        for response_elem in root.findall("d:response", ns):
+            href = response_elem.find("d:href", ns)
+            if href is None or href.text == "/remote.php/dav/systemtags/":
+                # Skip the collection itself
+                continue
+
+            propstat = response_elem.find("d:propstat", ns)
+            if propstat is None:
+                continue
+
+            prop = propstat.find("d:prop", ns)
+            if prop is None:
+                continue
+
+            # Extract tag properties
+            tag_id_elem = prop.find("oc:id", ns)
+            display_name_elem = prop.find("oc:display-name", ns)
+            user_visible_elem = prop.find("oc:user-visible", ns)
+            user_assignable_elem = prop.find("oc:user-assignable", ns)
+
+            if display_name_elem is not None and display_name_elem.text == tag_name:
+                tag_info = {
+                    "id": int(tag_id_elem.text)
+                    if tag_id_elem is not None and tag_id_elem.text is not None
+                    else None,
+                    "name": display_name_elem.text,
+                    "userVisible": user_visible_elem.text.lower() == "true"
+                    if user_visible_elem is not None
+                    else True,
+                    "userAssignable": user_assignable_elem.text.lower() == "true"
+                    if user_assignable_elem is not None
+                    else True,
+                }
+                logger.debug(f"Found tag '{tag_name}' with ID {tag_info['id']}")
+                return tag_info
+
+        logger.debug(f"Tag '{tag_name}' not found")
+        return None
+
+    async def get_files_by_tag(self, tag_id: int) -> list[dict[str, Any]]:
+        """Get all files tagged with a specific system tag via WebDAV REPORT.
+
+        Args:
+            tag_id: Numeric ID of the tag
+
+        Returns:
+            List of file info dictionaries with path, size, content_type, etc.
+        """
+        # Use WebDAV REPORT method with systemtag filter, requesting all properties
+        report_body = f"""<?xml version="1.0"?>
+<oc:filter-files xmlns:d="DAV:" xmlns:oc="http://owncloud.org/ns" xmlns:nc="http://nextcloud.org/ns">
+  <d:prop>
+    <oc:fileid/>
+    <d:displayname/>
+    <d:getcontentlength/>
+    <d:getcontenttype/>
+    <d:getlastmodified/>
+    <d:getetag/>
+  </d:prop>
+  <oc:filter-rules>
+    <oc:systemtag>{tag_id}</oc:systemtag>
+  </oc:filter-rules>
+</oc:filter-files>"""
+
+        response = await self._client.request(
+            "REPORT",
+            f"{self._get_webdav_base_path()}/",
+            content=report_body,
+        )
+        response.raise_for_status()
+
+        # Parse XML response
+        root = ET.fromstring(response.content)
+        ns = {
+            "d": "DAV:",
+            "oc": "http://owncloud.org/ns",
+        }
+
+        files = []
+        for response_elem in root.findall("d:response", ns):
+            # Extract href (file path)
+            href_elem = response_elem.find("d:href", ns)
+            if href_elem is None or not href_elem.text:
+                continue
+
+            propstat = response_elem.find("d:propstat", ns)
+            if propstat is None:
+                continue
+
+            prop = propstat.find("d:prop", ns)
+            if prop is None:
+                continue
+
+            # Extract all properties
+            fileid_elem = prop.find("oc:fileid", ns)
+            displayname_elem = prop.find("d:displayname", ns)
+            contentlength_elem = prop.find("d:getcontentlength", ns)
+            contenttype_elem = prop.find("d:getcontenttype", ns)
+            lastmodified_elem = prop.find("d:getlastmodified", ns)
+            etag_elem = prop.find("d:getetag", ns)
+
+            if fileid_elem is None or not fileid_elem.text:
+                continue
+
+            # Decode href path and extract the file path
+            from urllib.parse import unquote
+
+            href_path = unquote(href_elem.text)
+            # Remove WebDAV prefix to get user-relative path
+            webdav_prefix = f"/remote.php/dav/files/{self.username}/"
+            file_path = href_path.replace(webdav_prefix, "/")
+
+            # Parse last modified timestamp
+            last_modified_timestamp = None
+            if lastmodified_elem is not None and lastmodified_elem.text:
+                from email.utils import parsedate_to_datetime
+
+                try:
+                    dt = parsedate_to_datetime(lastmodified_elem.text)
+                    last_modified_timestamp = int(dt.timestamp())
+                except Exception:
+                    pass
+
+            file_info = {
+                "id": int(fileid_elem.text),
+                "path": file_path,
+                "name": displayname_elem.text
+                if displayname_elem is not None
+                else file_path.split("/")[-1],
+                "size": int(contentlength_elem.text)
+                if contentlength_elem is not None and contentlength_elem.text
+                else 0,
+                "content_type": contenttype_elem.text
+                if contenttype_elem is not None
+                else "",
+                "last_modified": lastmodified_elem.text
+                if lastmodified_elem is not None
+                else None,
+                "last_modified_timestamp": last_modified_timestamp,
+                "etag": etag_elem.text if etag_elem is not None else None,
+            }
+            files.append(file_info)
+
+        logger.debug(f"Found {len(files)} files with tag ID {tag_id}")
+        return files
+
+    async def get_file_info(self, path: str) -> dict[str, Any] | None:
+        """Get file info including file ID via WebDAV PROPFIND.
+
+        Args:
+            path: Path to the file (relative to user's files directory)
+
+        Returns:
+            File info dictionary with id, name, size, content_type, etc.
+            Returns None if file not found.
+        """
+        webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
+
+        propfind_body = """<?xml version="1.0"?>
+<d:propfind xmlns:d="DAV:" xmlns:oc="http://owncloud.org/ns">
+  <d:prop>
+    <oc:fileid/>
+    <d:displayname/>
+    <d:getcontentlength/>
+    <d:getcontenttype/>
+    <d:getlastmodified/>
+    <d:getetag/>
+    <d:resourcetype/>
+  </d:prop>
+</d:propfind>"""
+
+        try:
+            response = await self._client.request(
+                "PROPFIND",
+                webdav_path,
+                headers={"Depth": "0"},
+                content=propfind_body,
+            )
+            response.raise_for_status()
+        except HTTPStatusError as e:
+            if e.response.status_code == 404:
+                logger.debug(f"File not found: {path}")
+                return None
+            raise
+
+        # Parse XML response
+        root = ET.fromstring(response.content)
+        ns = {
+            "d": "DAV:",
+            "oc": "http://owncloud.org/ns",
+        }
+
+        response_elem = root.find("d:response", ns)
+        if response_elem is None:
+            return None
+
+        propstat = response_elem.find("d:propstat", ns)
+        if propstat is None:
+            return None
+
+        prop = propstat.find("d:prop", ns)
+        if prop is None:
+            return None
+
+        # Extract properties
+        fileid_elem = prop.find("oc:fileid", ns)
+        displayname_elem = prop.find("d:displayname", ns)
+        contentlength_elem = prop.find("d:getcontentlength", ns)
+        contenttype_elem = prop.find("d:getcontenttype", ns)
+        lastmodified_elem = prop.find("d:getlastmodified", ns)
+        etag_elem = prop.find("d:getetag", ns)
+        resourcetype_elem = prop.find("d:resourcetype", ns)
+
+        is_directory = (
+            resourcetype_elem is not None
+            and resourcetype_elem.find("d:collection", ns) is not None
+        )
+
+        file_info = {
+            "id": int(fileid_elem.text)
+            if fileid_elem is not None and fileid_elem.text is not None
+            else None,
+            "path": path,
+            "name": displayname_elem.text
+            if displayname_elem is not None
+            else path.split("/")[-1],
+            "size": int(contentlength_elem.text)
+            if contentlength_elem is not None and contentlength_elem.text
+            else 0,
+            "content_type": contenttype_elem.text
+            if contenttype_elem is not None
+            else "",
+            "last_modified": lastmodified_elem.text
+            if lastmodified_elem is not None
+            else None,
+            "etag": etag_elem.text.strip('"')
+            if etag_elem is not None and etag_elem.text
+            else None,
+            "is_directory": is_directory,
+        }
+
+        logger.debug(f"Got file info for '{path}': id={file_info['id']}")
+        return file_info
+
+    async def create_tag(
+        self,
+        name: str,
+        user_visible: bool = True,
+        user_assignable: bool = True,
+    ) -> dict[str, Any]:
+        """Create a system tag via WebDAV.
+
+        Args:
+            name: Name of the tag to create
+            user_visible: Whether the tag is visible to users
+            user_assignable: Whether users can assign this tag
+
+        Returns:
+            Tag dictionary with id, name, userVisible, userAssignable
+
+        Raises:
+            HTTPStatusError: If tag creation fails (409 if already exists)
+        """
+        # Use WebDAV POST with JSON body to create tag
+        response = await self._client.post(
+            "/remote.php/dav/systemtags/",
+            headers={"Content-Type": "application/json"},
+            json={
+                "name": name,
+                "userVisible": user_visible,
+                "userAssignable": user_assignable,
+            },
+        )
+        response.raise_for_status()
+
+        # Extract tag ID from Content-Location header (e.g., /remote.php/dav/systemtags/42)
+        content_location = response.headers.get("Content-Location", "")
+        tag_id = None
+        if content_location:
+            # Extract the numeric ID from the path
+            try:
+                tag_id = int(content_location.rstrip("/").split("/")[-1])
+            except (ValueError, IndexError):
+                pass
+
+        tag_info = {
+            "id": tag_id,
+            "name": name,
+            "userVisible": user_visible,
+            "userAssignable": user_assignable,
+        }
+
+        logger.info(f"Created tag '{name}' with ID {tag_info['id']}")
+        return tag_info
+
+    async def get_or_create_tag(
+        self,
+        name: str,
+        user_visible: bool = True,
+        user_assignable: bool = True,
+    ) -> dict[str, Any]:
+        """Get a tag by name, creating it if it doesn't exist.
+
+        Args:
+            name: Name of the tag
+            user_visible: Whether the tag is visible to users (for creation)
+            user_assignable: Whether users can assign this tag (for creation)
+
+        Returns:
+            Tag dictionary with id, name, userVisible, userAssignable
+        """
+        # First try to get existing tag
+        existing_tag = await self.get_tag_by_name(name)
+        if existing_tag:
+            logger.debug(f"Tag '{name}' already exists with ID {existing_tag['id']}")
+            return existing_tag
+
+        # Create new tag
+        try:
+            return await self.create_tag(name, user_visible, user_assignable)
+        except HTTPStatusError as e:
+            if e.response.status_code == 409:
+                # Tag was created between our check and creation, fetch it
+                existing_tag = await self.get_tag_by_name(name)
+                if existing_tag:
+                    return existing_tag
+            raise
+
+    async def assign_tag_to_file(self, file_id: int, tag_id: int) -> bool:
+        """Assign a system tag to a file.
+
+        Args:
+            file_id: Numeric file ID
+            tag_id: Numeric tag ID
+
+        Returns:
+            True if tag was assigned successfully (or already assigned)
+
+        Raises:
+            HTTPStatusError: If tag assignment fails
+        """
+        response = await self._client.request(
+            "PUT",
+            f"/remote.php/dav/systemtags-relations/files/{file_id}/{tag_id}",
+            headers={"Content-Length": "0"},
+            content=b"",
+        )
+
+        # 201 = Created (new assignment), 409 = Conflict (already assigned)
+        if response.status_code in (201, 409):
+            logger.info(f"Tagged file {file_id} with tag {tag_id}")
+            return True
+
+        response.raise_for_status()
+        return True
+
+    async def remove_tag_from_file(self, file_id: int, tag_id: int) -> bool:
+        """Remove a system tag from a file.
+
+        Args:
+            file_id: Numeric file ID
+            tag_id: Numeric tag ID
+
+        Returns:
+            True if tag was removed successfully (or wasn't assigned)
+
+        Raises:
+            HTTPStatusError: If tag removal fails
+        """
+        response = await self._client.request(
+            "DELETE",
+            f"/remote.php/dav/systemtags-relations/files/{file_id}/{tag_id}",
+        )
+
+        # 204 = No Content (removed), 404 = Not Found (wasn't assigned)
+        if response.status_code in (204, 404):
+            logger.info(f"Removed tag {tag_id} from file {file_id}")
+            return True
+
+        response.raise_for_status()
+        return True

nextcloud_mcp_server/config.py

@@ -2,8 +2,37 @@ import logging
 import logging.config
 import os
 from dataclasses import dataclass
+from enum import Enum
 from typing import Any, Optional
 
+
+class DeploymentMode(Enum):
+    """Deployment mode for the MCP server.
+
+    SELF_HOSTED: Full features, environment-based configuration.
+                 Supports vector sync, semantic search, admin UI.
+
+    SMITHERY_STATELESS: Stateless mode for Smithery hosting.
+                        Session-based configuration, no persistent storage.
+                        Excludes semantic search, vector sync, admin UI.
+    """
+
+    SELF_HOSTED = "self_hosted"
+    SMITHERY_STATELESS = "smithery"
+
+
+def get_deployment_mode() -> DeploymentMode:
+    """Detect deployment mode from environment.
+
+    Returns:
+        DeploymentMode.SMITHERY_STATELESS if SMITHERY_DEPLOYMENT=true,
+        otherwise DeploymentMode.SELF_HOSTED (default).
+    """
+    if os.getenv("SMITHERY_DEPLOYMENT", "false").lower() == "true":
+        return DeploymentMode.SMITHERY_STATELESS
+    return DeploymentMode.SELF_HOSTED
+
+
 LOGGING_CONFIG = {
     "version": 1,
     "disable_existing_loggers": False,
@@ -102,6 +131,14 @@ def get_document_processor_config() -> dict[str, Any]:
             "lang": os.getenv("TESSERACT_LANG", "eng"),
         }
 
+    # PyMuPDF configuration (local PDF processing)
+    if os.getenv("ENABLE_PYMUPDF", "true").lower() == "true":  # Enabled by default
+        config["processors"]["pymupdf"] = {
+            "extract_images": os.getenv("PYMUPDF_EXTRACT_IMAGES", "true").lower()
+            == "true",
+            "image_dir": os.getenv("PYMUPDF_IMAGE_DIR"),  # None = use temp directory
+        }
+
     # Custom processor (via HTTP API)
     if os.getenv("ENABLE_CUSTOM_PROCESSOR", "false").lower() == "true":
         custom_url = os.getenv("CUSTOM_PROCESSOR_URL")
@@ -180,9 +217,14 @@ class Settings:
     ollama_embedding_model: str = "nomic-embed-text"
     ollama_verify_ssl: bool = True
 
+    # OpenAI settings (for embeddings)
+    openai_api_key: Optional[str] = None
+    openai_base_url: Optional[str] = None
+    openai_embedding_model: str = "text-embedding-3-small"
+
     # Document chunking settings (for vector embeddings)
-    document_chunk_size: int = 512  # Words per chunk
-    document_chunk_overlap: int = 50  # Overlapping words between chunks
+    document_chunk_size: int = 2048  # Characters per chunk
+    document_chunk_overlap: int = 200  # Overlapping characters between chunks
 
     # Observability settings
     metrics_enabled: bool = True
@@ -227,10 +269,10 @@ class Settings:
                 f"Overlap should be 10-20% of chunk size for optimal results."
             )
 
-        if self.document_chunk_size < 100:
+        if self.document_chunk_size < 512:
             logger.warning(
-                f"DOCUMENT_CHUNK_SIZE is set to {self.document_chunk_size} words, which is quite small. "
-                f"Smaller chunks may lose context. Consider using at least 256 words."
+                f"DOCUMENT_CHUNK_SIZE is set to {self.document_chunk_size} characters, which is quite small. "
+                f"Smaller chunks may lose context. Consider using at least 1024 characters."
             )
 
         if self.document_chunk_overlap < 0:
@@ -238,6 +280,29 @@ class Settings:
                 f"DOCUMENT_CHUNK_OVERLAP ({self.document_chunk_overlap}) cannot be negative."
             )
 
+    def get_embedding_model_name(self) -> str:
+        """
+        Get the active embedding model name based on provider priority.
+
+        Priority order (same as ProviderRegistry):
+        1. OpenAI - if OPENAI_API_KEY is set
+        2. Ollama - if OLLAMA_BASE_URL is set
+        3. Simple - fallback (returns "simple-384")
+
+        Returns:
+            Active embedding model name
+        """
+        # Check OpenAI first (higher priority than Ollama in registry)
+        if self.openai_api_key:
+            return self.openai_embedding_model
+
+        # Check Ollama
+        if self.ollama_base_url:
+            return self.ollama_embedding_model
+
+        # Fallback to simple provider indicator
+        return "simple-384"
+
     def get_collection_name(self) -> str:
         """
         Get Qdrant collection name.
@@ -253,8 +318,9 @@ class Settings:
         Format: {deployment-id}-{model-name}
 
         Examples:
-            - "my-deployment-nomic-embed-text" (OTEL_SERVICE_NAME set)
-            - "mcp-container-all-minilm" (hostname fallback)
+            - "my-deployment-nomic-embed-text" (Ollama)
+            - "my-deployment-text-embedding-3-small" (OpenAI)
+            - "mcp-container-openai-text-embedding-3-small" (hostname fallback)
 
         Returns:
             Collection name string
@@ -274,7 +340,7 @@ class Settings:
 
         # Sanitize deployment ID and model name
         deployment_id = deployment_id.lower().replace(" ", "-").replace("_", "-")
-        model_name = self.ollama_embedding_model.replace("/", "-").replace(":", "-")
+        model_name = self.get_embedding_model_name().replace("/", "-").replace(":", "-")
 
         return f"{deployment_id}-{model_name}"
 
@@ -334,9 +400,15 @@ def get_settings() -> Settings:
         ollama_base_url=os.getenv("OLLAMA_BASE_URL"),
         ollama_embedding_model=os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text"),
         ollama_verify_ssl=os.getenv("OLLAMA_VERIFY_SSL", "true").lower() == "true",
+        # OpenAI settings
+        openai_api_key=os.getenv("OPENAI_API_KEY"),
+        openai_base_url=os.getenv("OPENAI_BASE_URL"),
+        openai_embedding_model=os.getenv(
+            "OPENAI_EMBEDDING_MODEL", "text-embedding-3-small"
+        ),
         # Document chunking settings
-        document_chunk_size=int(os.getenv("DOCUMENT_CHUNK_SIZE", "512")),
-        document_chunk_overlap=int(os.getenv("DOCUMENT_CHUNK_OVERLAP", "50")),
+        document_chunk_size=int(os.getenv("DOCUMENT_CHUNK_SIZE", "2048")),
+        document_chunk_overlap=int(os.getenv("DOCUMENT_CHUNK_OVERLAP", "200")),
         # Observability settings
         metrics_enabled=os.getenv("METRICS_ENABLED", "true").lower() == "true",
         metrics_port=int(os.getenv("METRICS_PORT", "9090")),

nextcloud_mcp_server/context.py

@@ -1,21 +1,37 @@
 """Helper functions for accessing context in MCP tools."""
 
+import logging
+
+from httpx import BasicAuth
 from mcp.server.fastmcp import Context
 
 from nextcloud_mcp_server.client import NextcloudClient
-from nextcloud_mcp_server.config import get_settings
+from nextcloud_mcp_server.config import (
+    DeploymentMode,
+    get_deployment_mode,
+    get_settings,
+)
+
+logger = logging.getLogger(__name__)
 
 
 async def get_client(ctx: Context) -> NextcloudClient:
     """
     Get the appropriate Nextcloud client based on authentication mode.
 
-    ADR-005 compliant implementation supporting two modes:
-    1. BasicAuth mode: Returns shared client from lifespan context
-    2. Multi-audience mode (ENABLE_TOKEN_EXCHANGE=false, default):
-       Token already contains both MCP and Nextcloud audiences - use directly
-    3. Token exchange mode (ENABLE_TOKEN_EXCHANGE=true):
-       Exchange MCP token for Nextcloud token via RFC 8693
+    ADR-016 compliant implementation supporting three deployment modes:
+
+    1. Smithery stateless mode (SMITHERY_DEPLOYMENT=true):
+       Create client from session configuration (nextcloud_url, username, app_password)
+       No persistent state - client created per-request from Smithery session config.
+
+    2. BasicAuth mode: Returns shared client from lifespan context
+
+    3. OAuth mode:
+       a. Multi-audience mode (ENABLE_TOKEN_EXCHANGE=false, default):
+          Token already contains both MCP and Nextcloud audiences - use directly
+       b. Token exchange mode (ENABLE_TOKEN_EXCHANGE=true):
+          Exchange MCP token for Nextcloud token via RFC 8693
 
     SECURITY: Token passthrough has been REMOVED. All OAuth modes validate
     proper token audiences per MCP Security Best Practices specification.
@@ -24,7 +40,7 @@ async def get_client(ctx: Context) -> NextcloudClient:
     by the MCP server via @require_scopes decorator, not by the IdP.
 
     This function automatically detects the authentication mode by checking
-    the type of the lifespan context.
+    the deployment mode and type of the lifespan context.
 
     Args:
         ctx: MCP request context
@@ -34,6 +50,7 @@ async def get_client(ctx: Context) -> NextcloudClient:
 
     Raises:
         AttributeError: If context doesn't contain expected data
+        ValueError: If Smithery mode but session config is missing required fields
 
     Example:
         ```python
@@ -43,6 +60,12 @@ async def get_client(ctx: Context) -> NextcloudClient:
             return await client.capabilities()
         ```
     """
+    deployment_mode = get_deployment_mode()
+
+    # ADR-016: Smithery stateless mode - create client from session config
+    if deployment_mode == DeploymentMode.SMITHERY_STATELESS:
+        return _get_client_from_session_config(ctx)
+
     settings = get_settings()
     lifespan_ctx = ctx.request_context.lifespan_context
 
@@ -75,3 +98,82 @@ async def get_client(ctx: Context) -> NextcloudClient:
         f"Lifespan context does not have 'client' or 'nextcloud_host' attribute. "
         f"Type: {type(lifespan_ctx)}"
     )
+
+
+def _get_client_from_session_config(ctx: Context) -> NextcloudClient:
+    """
+    Create NextcloudClient from Smithery session configuration.
+
+    ADR-016: In Smithery stateless mode, each request includes session config
+    with the user's Nextcloud credentials. This function creates a fresh client
+    for each request - no state is persisted between requests.
+
+    For container runtime, config is extracted from URL query parameters by
+    SmitheryConfigMiddleware and stored in a context variable.
+
+    Expected session config fields (from Smithery configSchema):
+    - nextcloud_url: str - Nextcloud instance URL (required)
+    - username: str - Nextcloud username (required)
+    - app_password: str - Nextcloud app password (required)
+
+    Args:
+        ctx: MCP request context (not used directly for Smithery config)
+
+    Returns:
+        NextcloudClient configured with session credentials
+
+    Raises:
+        ValueError: If required session config fields are missing
+    """
+    # ADR-016: Get session config from context variable (set by SmitheryConfigMiddleware)
+    from nextcloud_mcp_server.app import get_smithery_session_config
+
+    session_config = get_smithery_session_config()
+
+    if session_config is None:
+        raise ValueError(
+            "Session configuration required in Smithery mode. "
+            "Ensure nextcloud_url, username, and app_password are provided as URL query parameters."
+        )
+
+    # Extract required fields - config is always a dict from SmitheryConfigMiddleware
+    nextcloud_url = session_config.get("nextcloud_url")
+    username = session_config.get("username")
+    app_password = session_config.get("app_password")
+
+    # Validate required fields
+    missing_fields = []
+    if not nextcloud_url:
+        missing_fields.append("nextcloud_url")
+    if not username:
+        missing_fields.append("username")
+    if not app_password:
+        missing_fields.append("app_password")
+
+    if missing_fields:
+        raise ValueError(
+            f"Missing required session config fields: {', '.join(missing_fields)}. "
+            f"Configure these in the Smithery connection settings."
+        )
+
+    # Type assertions after validation (for type checker)
+    # These are guaranteed to be str after the missing_fields check above
+    assert nextcloud_url is not None
+    assert username is not None
+    assert app_password is not None
+
+    # Validate URL format
+    if not nextcloud_url.startswith(("http://", "https://")):
+        raise ValueError(
+            f"Invalid nextcloud_url: {nextcloud_url}. "
+            f"Must start with http:// or https://"
+        )
+
+    logger.debug(f"Creating Smithery client for {nextcloud_url} as {username}")
+
+    # Create client with session credentials using BasicAuth
+    return NextcloudClient(
+        base_url=nextcloud_url,
+        username=username,
+        auth=BasicAuth(username, app_password),
+    )

nextcloud_mcp_server/document_processors/__init__.py

@@ -1,12 +1,18 @@
 """Document processing plugins for extracting text from various file formats."""
 
 from .base import DocumentProcessor, ProcessingResult, ProcessorError
+from .pymupdf import PyMuPDFProcessor
 from .registry import ProcessorRegistry, get_registry
 
+# Register processors at module initialization
+_registry = get_registry()
+_registry.register(PyMuPDFProcessor(), priority=10)
+
 __all__ = [
     "DocumentProcessor",
     "ProcessingResult",
     "ProcessorError",
     "ProcessorRegistry",
     "get_registry",
+    "PyMuPDFProcessor",
 ]

nextcloud_mcp_server/document_processors/pymupdf.py

@@ -0,0 +1,253 @@
+"""Document processor using PyMuPDF (fitz) library."""
+
+import logging
+import pathlib
+import tempfile
+from collections.abc import Awaitable, Callable
+from typing import Any, Optional
+
+# NOTE: Do NOT call pymupdf.layout.activate() here!
+# It changes the behavior of pymupdf4llm.to_markdown() when page_chunks=True,
+# causing it to return a string instead of a list[dict].
+# See: https://github.com/pymupdf/pymupdf4llm/issues/323
+import pymupdf
+import pymupdf4llm
+
+from .base import DocumentProcessor, ProcessingResult, ProcessorError
+
+logger = logging.getLogger(__name__)
+
+
+class PyMuPDFProcessor(DocumentProcessor):
+    """Document processor using PyMuPDF library for PDF processing.
+
+    PyMuPDF (fitz) is a fast, local PDF processing library that extracts text,
+    metadata, and images without requiring external API calls.
+
+    Features:
+    - Fast text extraction with layout preservation
+    - PDF metadata extraction (title, author, creation date, page count)
+    - Image extraction for future multimodal support
+    - Page number tracking for precise citations
+    """
+
+    SUPPORTED_TYPES = {
+        "application/pdf",
+    }
+
+    def __init__(
+        self,
+        extract_images: bool = True,
+        image_dir: Optional[str | pathlib.Path] = None,
+    ):
+        """Initialize PyMuPDF processor.
+
+        Args:
+            extract_images: Whether to extract embedded images from PDFs
+            image_dir: Directory to store extracted images (defaults to temp directory)
+        """
+        self.extract_images = extract_images
+
+        if image_dir is None:
+            self.image_dir = pathlib.Path(tempfile.gettempdir()) / "pdf-images"
+        else:
+            self.image_dir = pathlib.Path(image_dir)
+
+        # Create image directory if it doesn't exist
+        if self.extract_images:
+            self.image_dir.mkdir(exist_ok=True, parents=True)
+            logger.info(
+                f"Initialized PyMuPDFProcessor with image extraction to {self.image_dir}"
+            )
+        else:
+            logger.info("Initialized PyMuPDFProcessor without image extraction")
+
+    @property
+    def name(self) -> str:
+        return "pymupdf"
+
+    @property
+    def supported_mime_types(self) -> set[str]:
+        return self.SUPPORTED_TYPES
+
+    async def process(
+        self,
+        content: bytes,
+        content_type: str,
+        filename: Optional[str] = None,
+        options: Optional[dict[str, Any]] = None,
+        progress_callback: Optional[
+            Callable[[float, Optional[float], Optional[str]], Awaitable[None]]
+        ] = None,
+    ) -> ProcessingResult:
+        """Process a PDF document and extract text, metadata, and images.
+
+        Args:
+            content: PDF document bytes
+            content_type: MIME type (should be application/pdf)
+            filename: Optional filename for better error messages
+            options: Processing options (currently unused)
+            progress_callback: Optional callback for progress updates
+
+        Returns:
+            ProcessingResult with extracted text and metadata
+
+        Raises:
+            ProcessorError: If PDF processing fails
+        """
+        import anyio
+
+        try:
+            if progress_callback:
+                await progress_callback(0, 100, "Opening PDF document")
+
+            # Open document and extract metadata in thread
+            doc = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+                lambda: pymupdf.open("pdf", content)
+            )
+
+            metadata = self._extract_metadata(doc, filename)
+            metadata["file_size"] = len(content)
+            page_count = doc.page_count
+
+            if progress_callback:
+                await progress_callback(10, 100, f"Extracting {page_count} pages")
+
+            # Prepare image directory if needed
+            pdf_image_dir = None
+            if self.extract_images:
+                pdf_id = filename.replace("/", "_") if filename else "unknown"
+                pdf_image_dir = self.image_dir / pdf_id
+                pdf_image_dir.mkdir(exist_ok=True, parents=True)
+
+            # Extract all pages in a single call with page_chunks=True
+            def do_extract() -> list[dict[str, Any]]:
+                # When page_chunks=True, to_markdown returns list[dict] not str
+                return pymupdf4llm.to_markdown(  # type: ignore[return-value]
+                    doc,
+                    write_images=self.extract_images,
+                    image_path=pdf_image_dir if self.extract_images else None,
+                    page_chunks=True,
+                )
+
+            page_chunks: list[dict[str, Any]] = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+                do_extract
+            )
+
+            if progress_callback:
+                await progress_callback(90, 100, "Building result")
+
+            # Extract page texts and build boundaries from chunks
+            page_texts: list[str] = []
+            page_boundaries: list[dict[str, Any]] = []
+            current_offset = 0
+            for chunk in page_chunks:
+                text = chunk.get("text", "")
+                page_num = chunk.get("metadata", {}).get("page", len(page_texts) + 1)
+                page_texts.append(text)
+                page_boundaries.append(
+                    {
+                        "page": page_num,
+                        "start_offset": current_offset,
+                        "end_offset": current_offset + len(text),
+                    }
+                )
+                current_offset += len(text)
+
+            # Collect image paths
+            image_paths = []
+            if pdf_image_dir and pdf_image_dir.exists():
+                image_paths = [str(p) for p in pdf_image_dir.glob("*")]
+
+            # Build final text and metadata
+            md_text = "".join(page_texts)
+            metadata["has_images"] = len(image_paths) > 0
+            if image_paths:
+                metadata["image_count"] = len(image_paths)
+                metadata["image_paths"] = image_paths
+            metadata["page_boundaries"] = page_boundaries
+
+            # Close document
+            doc.close()
+
+            if progress_callback:
+                await progress_callback(100, 100, "Processing complete")
+
+            logger.info(
+                f"Successfully processed PDF {filename or '<bytes>'}: "
+                f"{metadata['page_count']} pages, {len(md_text)} chars, "
+                f"{metadata.get('image_count', 0)} images"
+            )
+
+            return ProcessingResult(
+                text=md_text,
+                metadata=metadata,
+                processor=self.name,
+                success=True,
+            )
+
+        except Exception as e:
+            error_msg = f"Failed to process PDF {filename or '<bytes>'}: {e}"
+            logger.error(error_msg, exc_info=True)
+            raise ProcessorError(error_msg) from e
+
+    def _extract_metadata(
+        self, doc: pymupdf.Document, filename: Optional[str]
+    ) -> dict[str, Any]:
+        """Extract metadata from PDF document.
+
+        Args:
+            doc: Opened PyMuPDF document
+            filename: Optional filename
+
+        Returns:
+            Dictionary with PDF metadata
+        """
+        metadata: dict[str, Any] = {}
+
+        # Basic document info
+        metadata["page_count"] = doc.page_count
+        metadata["format"] = "PDF 1." + str(
+            doc.pdf_version() if hasattr(doc, "pdf_version") else "?"  # type: ignore[call-non-callable]
+        )
+
+        if filename:
+            metadata["filename"] = filename
+
+        # Extract PDF metadata dictionary
+        pdf_metadata = doc.metadata
+        if pdf_metadata:
+            # Standard PDF metadata fields
+            if pdf_metadata.get("title"):
+                metadata["title"] = pdf_metadata["title"]
+            if pdf_metadata.get("author"):
+                metadata["author"] = pdf_metadata["author"]
+            if pdf_metadata.get("subject"):
+                metadata["subject"] = pdf_metadata["subject"]
+            if pdf_metadata.get("keywords"):
+                metadata["keywords"] = pdf_metadata["keywords"]
+            if pdf_metadata.get("creator"):
+                metadata["creator"] = pdf_metadata["creator"]
+            if pdf_metadata.get("producer"):
+                metadata["producer"] = pdf_metadata["producer"]
+            if pdf_metadata.get("creationDate"):
+                metadata["creation_date"] = pdf_metadata["creationDate"]
+            if pdf_metadata.get("modDate"):
+                metadata["modification_date"] = pdf_metadata["modDate"]
+
+        return metadata
+
+    async def health_check(self) -> bool:
+        """Check if PyMuPDF is available and working.
+
+        Returns:
+            True if processor is ready to use
+        """
+        try:
+            # Try to create a simple PDF in memory
+            test_doc = pymupdf.open()
+            test_doc.close()
+            return True
+        except Exception as e:
+            logger.error(f"PyMuPDF health check failed: {e}")
+            return False

nextcloud_mcp_server/embedding/__init__.py

@@ -1,6 +1,13 @@
 """Embedding service package for generating vector embeddings."""
 
-from .service import EmbeddingService, get_embedding_service
+from .bm25_provider import BM25SparseEmbeddingProvider
+from .service import EmbeddingService, get_bm25_service, get_embedding_service
 from .simple_provider import SimpleEmbeddingProvider
 
-__all__ = ["EmbeddingService", "get_embedding_service", "SimpleEmbeddingProvider"]
+__all__ = [
+    "EmbeddingService",
+    "get_embedding_service",
+    "BM25SparseEmbeddingProvider",
+    "get_bm25_service",
+    "SimpleEmbeddingProvider",
+]

nextcloud_mcp_server/embedding/bm25_provider.py

@@ -0,0 +1,98 @@
+"""BM25 sparse embedding provider using FastEmbed."""
+
+import logging
+from typing import Any
+
+from fastembed import SparseTextEmbedding
+
+logger = logging.getLogger(__name__)
+
+
+class BM25SparseEmbeddingProvider:
+    """
+    BM25 sparse embedding provider for hybrid search.
+
+    Uses FastEmbed's BM25 model to generate sparse vectors for keyword-based
+    retrieval. These sparse vectors are combined with dense semantic vectors
+    in Qdrant using Reciprocal Rank Fusion (RRF) for hybrid search.
+
+    Unlike dense embeddings which have fixed dimensions, sparse embeddings
+    have variable-length vectors with (index, value) pairs representing
+    term frequencies in the BM25 vocabulary.
+    """
+
+    def __init__(self, model_name: str = "Qdrant/bm25"):
+        """
+        Initialize BM25 sparse embedding provider.
+
+        Args:
+            model_name: FastEmbed BM25 model name (default: Qdrant/bm25)
+        """
+        self.model_name = model_name
+        logger.info(f"Initializing BM25 sparse embedding provider: {model_name}")
+
+        # Initialize FastEmbed sparse embedding model
+        self.model = SparseTextEmbedding(model_name=model_name)
+        logger.info(f"BM25 sparse embedding model loaded: {model_name}")
+
+    def encode(self, text: str) -> dict[str, Any]:
+        """
+        Generate BM25 sparse embedding for a single text (synchronous).
+
+        Note: For async contexts, prefer encode_async() to avoid blocking the event loop.
+
+        Args:
+            text: Input text to encode
+
+        Returns:
+            Dictionary with 'indices' and 'values' keys for Qdrant sparse vector
+        """
+        # FastEmbed returns a generator, take first result
+        sparse_embedding = next(iter(self.model.embed([text])))
+
+        return {
+            "indices": sparse_embedding.indices.tolist(),
+            "values": sparse_embedding.values.tolist(),
+        }
+
+    async def encode_async(self, text: str) -> dict[str, Any]:
+        """
+        Generate BM25 sparse embedding for a single text (async).
+
+        Runs CPU-bound BM25 encoding in thread pool to avoid blocking the event loop.
+
+        Args:
+            text: Input text to encode
+
+        Returns:
+            Dictionary with 'indices' and 'values' keys for Qdrant sparse vector
+        """
+        import anyio
+
+        # Run CPU-bound BM25 encoding in thread pool
+        return await anyio.to_thread.run_sync(lambda: self.encode(text))  # type: ignore[attr-defined]
+
+    async def encode_batch(self, texts: list[str]) -> list[dict[str, Any]]:
+        """
+        Generate BM25 sparse embeddings for multiple texts (batched).
+
+        Args:
+            texts: List of texts to encode
+
+        Returns:
+            List of dictionaries with 'indices' and 'values' for each text
+        """
+        import anyio
+
+        # Run CPU-bound BM25 encoding in thread pool to avoid blocking event loop
+        sparse_embeddings = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+            lambda: list(self.model.embed(texts))
+        )
+
+        return [
+            {
+                "indices": emb.indices.tolist(),
+                "values": emb.values.tolist(),
+            }
+            for emb in sparse_embeddings
+        ]

nextcloud_mcp_server/embedding/service.py

@@ -1,56 +1,30 @@
-"""Embedding service with provider detection."""
+"""Embedding service with provider detection.
+
+DEPRECATED: This module is maintained for backward compatibility.
+New code should use nextcloud_mcp_server.providers.get_provider() directly.
+"""
 
 import logging
-import os
 
-from .base import EmbeddingProvider
-from .ollama_provider import OllamaEmbeddingProvider
-from .simple_provider import SimpleEmbeddingProvider
+from nextcloud_mcp_server.providers import get_provider
+
+from .bm25_provider import BM25SparseEmbeddingProvider
 
 logger = logging.getLogger(__name__)
 
 
 class EmbeddingService:
-    """Unified embedding service with automatic provider detection."""
+    """
+    Unified embedding service with automatic provider detection.
+
+    DEPRECATED: This class wraps the new unified provider infrastructure
+    for backward compatibility. New code should use
+    nextcloud_mcp_server.providers.get_provider() directly.
+    """
 
     def __init__(self):
         """Initialize embedding service with auto-detected provider."""
-        self.provider = self._detect_provider()
-
-    def _detect_provider(self) -> EmbeddingProvider:
-        """
-        Auto-detect available embedding provider.
-
-        Checks environment variables in order:
-        1. OLLAMA_BASE_URL - Use Ollama provider (production)
-        2. OPENAI_API_KEY - Use OpenAI provider (future)
-        3. Fallback to SimpleEmbeddingProvider (testing/development)
-
-        Returns:
-            Configured embedding provider
-        """
-        # Ollama provider (production)
-        ollama_url = os.getenv("OLLAMA_BASE_URL")
-        if ollama_url:
-            logger.info(f"Using Ollama embedding provider: {ollama_url}")
-            return OllamaEmbeddingProvider(
-                base_url=ollama_url,
-                model=os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text"),
-                verify_ssl=os.getenv("OLLAMA_VERIFY_SSL", "true").lower() == "true",
-            )
-
-        # OpenAI provider (future implementation)
-        # openai_key = os.getenv("OPENAI_API_KEY")
-        # if openai_key:
-        #     return OpenAIEmbeddingProvider(api_key=openai_key)
-
-        # Fallback to simple provider for development/testing
-        logger.warning(
-            "No embedding provider configured (OLLAMA_BASE_URL or OPENAI_API_KEY not set). "
-            "Using SimpleEmbeddingProvider for testing/development. "
-            "For production, configure an external embedding service."
-        )
-        return SimpleEmbeddingProvider(dimension=384)
+        self.provider = get_provider()
 
     async def embed(self, text: str) -> list[float]:
         """
@@ -109,3 +83,20 @@ def get_embedding_service() -> EmbeddingService:
     if _embedding_service is None:
         _embedding_service = EmbeddingService()
     return _embedding_service
+
+
+# BM25 sparse embedding singleton
+_bm25_service: BM25SparseEmbeddingProvider | None = None
+
+
+def get_bm25_service() -> BM25SparseEmbeddingProvider:
+    """
+    Get singleton BM25 sparse embedding service instance.
+
+    Returns:
+        Global BM25SparseEmbeddingProvider instance
+    """
+    global _bm25_service
+    if _bm25_service is None:
+        _bm25_service = BM25SparseEmbeddingProvider()
+    return _bm25_service

nextcloud_mcp_server/models/news.py

@@ -0,0 +1,170 @@
+"""Pydantic models for Nextcloud News app responses."""
+
+from typing import List
+
+from pydantic import BaseModel, ConfigDict, Field
+
+from .base import BaseResponse
+
+
+class NewsFolder(BaseModel):
+    """Model for a News folder."""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    id: int = Field(description="Folder ID")
+    name: str = Field(description="Folder name")
+
+
+class NewsFeed(BaseModel):
+    """Model for a News feed (RSS/Atom subscription)."""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    id: int = Field(description="Feed ID")
+    url: str = Field(description="Feed URL")
+    title: str = Field(description="Feed title")
+    favicon_link: str | None = Field(
+        None, alias="faviconLink", description="Favicon URL"
+    )
+    link: str | None = Field(None, description="Website link")
+    added: int = Field(description="Unix timestamp when feed was added")
+    folder_id: int | None = Field(
+        None, alias="folderId", description="Parent folder ID"
+    )
+    unread_count: int = Field(
+        0, alias="unreadCount", description="Number of unread items"
+    )
+    ordering: int = Field(
+        0, description="Feed ordering (0=default, 1=oldest, 2=newest)"
+    )
+    pinned: bool = Field(False, description="Whether feed is pinned to top")
+    update_error_count: int = Field(
+        0, alias="updateErrorCount", description="Consecutive update failures"
+    )
+    last_update_error: str | None = Field(
+        None, alias="lastUpdateError", description="Last update error message"
+    )
+
+    @property
+    def has_errors(self) -> bool:
+        """Check if feed has update errors."""
+        return self.update_error_count > 0
+
+
+class NewsItem(BaseModel):
+    """Model for a News item (article) with full content."""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    id: int = Field(description="Item ID")
+    guid: str = Field(description="Globally unique identifier")
+    guid_hash: str = Field(alias="guidHash", description="MD5 hash of GUID")
+    url: str | None = Field(None, description="Article URL")
+    title: str = Field(description="Article title")
+    author: str | None = Field(None, description="Article author")
+    pub_date: int | None = Field(
+        None, alias="pubDate", description="Publication timestamp"
+    )
+    body: str | None = Field(None, description="Article content (HTML)")
+    enclosure_mime: str | None = Field(
+        None, alias="enclosureMime", description="Enclosure MIME type"
+    )
+    enclosure_link: str | None = Field(
+        None, alias="enclosureLink", description="Enclosure URL"
+    )
+    media_thumbnail: str | None = Field(
+        None, alias="mediaThumbnail", description="Media thumbnail URL"
+    )
+    media_description: str | None = Field(
+        None, alias="mediaDescription", description="Media description"
+    )
+    feed_id: int = Field(alias="feedId", description="Parent feed ID")
+    unread: bool = Field(True, description="Whether item is unread")
+    starred: bool = Field(False, description="Whether item is starred")
+    rtl: bool = Field(False, description="Right-to-left text")
+    last_modified: int = Field(
+        alias="lastModified", description="Last modification timestamp"
+    )
+    fingerprint: str | None = Field(
+        None, description="Content fingerprint for deduplication"
+    )
+    content_hash: str | None = Field(
+        None, alias="contentHash", description="Content hash"
+    )
+
+
+class NewsItemSummary(BaseModel):
+    """Lightweight model for News item list responses."""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    id: int = Field(description="Item ID")
+    title: str = Field(description="Article title")
+    feed_id: int = Field(alias="feedId", description="Parent feed ID")
+    unread: bool = Field(True, description="Whether item is unread")
+    starred: bool = Field(False, description="Whether item is starred")
+    pub_date: int | None = Field(
+        None, alias="pubDate", description="Publication timestamp"
+    )
+    url: str | None = Field(None, description="Article URL")
+    author: str | None = Field(None, description="Article author")
+
+
+class NewsStatus(BaseModel):
+    """Model for News app status."""
+
+    version: str = Field(description="News app version")
+    warnings: dict = Field(default_factory=dict, description="Configuration warnings")
+
+
+# --- Response Models ---
+
+
+class ListFoldersResponse(BaseResponse):
+    """Response model for listing folders."""
+
+    results: List[NewsFolder] = Field(description="List of folders")
+    total_count: int = Field(description="Total number of folders")
+
+
+class ListFeedsResponse(BaseResponse):
+    """Response model for listing feeds."""
+
+    results: List[NewsFeed] = Field(description="List of feeds")
+    starred_count: int = Field(0, description="Number of starred items")
+    newest_item_id: int | None = Field(None, description="ID of newest item")
+    total_count: int = Field(description="Total number of feeds")
+
+
+class ListItemsResponse(BaseResponse):
+    """Response model for listing items."""
+
+    results: List[NewsItemSummary] = Field(description="List of items")
+    total_count: int = Field(description="Number of items returned")
+    has_more: bool = Field(False, description="Whether more items exist")
+    oldest_id: int | None = Field(None, description="Oldest item ID (for pagination)")
+
+
+class GetItemResponse(BaseResponse):
+    """Response model for getting a single item."""
+
+    item: NewsItem = Field(description="Full item details")
+
+
+class FeedHealthResponse(BaseResponse):
+    """Response model for feed health status."""
+
+    feed_id: int = Field(description="Feed ID")
+    title: str = Field(description="Feed title")
+    url: str = Field(description="Feed URL")
+    has_errors: bool = Field(description="Whether feed has update errors")
+    error_count: int = Field(description="Number of consecutive errors")
+    last_error: str | None = Field(None, description="Last error message")
+
+
+class GetStatusResponse(BaseResponse):
+    """Response model for app status."""
+
+    version: str = Field(description="News app version")
+    warnings: dict = Field(default_factory=dict, description="Configuration warnings")

nextcloud_mcp_server/models/semantic.py

@@ -10,7 +10,7 @@ from .base import BaseResponse
 class SemanticSearchResult(BaseModel):
     """Model for semantic search results with additional metadata."""
 
-    id: int = Field(description="Document ID")
+    id: int = Field(description="Document ID (int for all document types)")
     doc_type: str = Field(
         description="Document type (note, calendar_event, deck_card, etc.)"
     )
@@ -19,9 +19,45 @@ class SemanticSearchResult(BaseModel):
         default="", description="Document category (notes) or location (calendar)"
     )
     excerpt: str = Field(description="Excerpt from matching chunk")
-    score: float = Field(description="Semantic similarity score (0-1)")
+    score: float = Field(
+        description=(
+            "Relevance score (≥ 0.0, higher is better). "
+            "Score range depends on fusion method: "
+            "RRF produces scores in [0.0, 1.0], "
+            "DBSF can exceed 1.0 (sum of normalized scores from multiple systems)"
+        )
+    )
     chunk_index: int = Field(description="Index of matching chunk in document")
     total_chunks: int = Field(description="Total number of chunks in document")
+    chunk_start_offset: Optional[int] = Field(
+        default=None, description="Character position where chunk starts in document"
+    )
+    chunk_end_offset: Optional[int] = Field(
+        default=None, description="Character position where chunk ends in document"
+    )
+    page_number: Optional[int] = Field(
+        default=None, description="Page number for PDF documents"
+    )
+    # Context expansion fields (optional, populated when include_context=True)
+    has_context_expansion: bool = Field(
+        default=False, description="Whether context expansion was performed"
+    )
+    marked_text: Optional[str] = Field(
+        default=None,
+        description="Full text with position markers around matched chunk",
+    )
+    before_context: Optional[str] = Field(
+        default=None, description="Text before the matched chunk"
+    )
+    after_context: Optional[str] = Field(
+        default=None, description="Text after the matched chunk"
+    )
+    has_before_truncation: Optional[bool] = Field(
+        default=None, description="Whether before_context was truncated"
+    )
+    has_after_truncation: Optional[bool] = Field(
+        default=None, description="Whether after_context was truncated"
+    )
 
 
 class SemanticSearchResponse(BaseResponse):

nextcloud_mcp_server/observability/logging_config.py

@@ -37,7 +37,7 @@ class HealthCheckFilter(logging.Filter):
         """
         # Check if the log message contains health check endpoints
         message = record.getMessage()
-        return not any(
+        health_check = any(
             endpoint in message
             for endpoint in [
                 "/health/live",
@@ -47,6 +47,8 @@ class HealthCheckFilter(logging.Filter):
             ]
         )
 
+        return not health_check
+
 
 class TraceContextFormatter(JsonFormatter):
     """
@@ -58,7 +60,7 @@ class TraceContextFormatter(JsonFormatter):
 
     def add_fields(
         self,
-        log_record: dict[str, Any],
+        log_data: dict[str, Any],
         record: logging.LogRecord,
         message_dict: dict[str, Any],
     ) -> None:
@@ -66,28 +68,28 @@ class TraceContextFormatter(JsonFormatter):
         Add custom fields to the log record, including trace context.
 
         Args:
-            log_record: Dictionary to be serialized as JSON
+            log_data: Dictionary to be serialized as JSON
             record: LogRecord instance
             message_dict: Dictionary of extra fields from log call
         """
         # Call parent to add standard fields
-        super().add_fields(log_record, record, message_dict)
+        super().add_fields(log_data, record, message_dict)
 
         # Add trace context if available
         trace_context = get_trace_context()
         if trace_context:
-            log_record["trace_id"] = trace_context.get("trace_id")
-            log_record["span_id"] = trace_context.get("span_id")
+            log_data["trace_id"] = trace_context.get("trace_id")
+            log_data["span_id"] = trace_context.get("span_id")
 
         # Add standard fields with consistent naming
-        log_record["timestamp"] = self.formatTime(record)
-        log_record["level"] = record.levelname
-        log_record["logger"] = record.name
-        log_record["message"] = record.getMessage()
+        log_data["timestamp"] = self.formatTime(record)
+        log_data["level"] = record.levelname
+        log_data["logger"] = record.name
+        log_data["message"] = record.getMessage()
 
         # Include exception info if present
         if record.exc_info:
-            log_record["exception"] = self.formatException(record.exc_info)
+            log_data["exception"] = self.formatException(record.exc_info)
 
 
 class TraceContextTextFormatter(logging.Formatter):

nextcloud_mcp_server/observability/metrics.py

@@ -404,10 +404,11 @@ def update_vector_sync_queue_size(size: int) -> None:
 
 def instrument_tool(func):
     """
-    Decorator to automatically instrument MCP tool functions with metrics.
+    Decorator to automatically instrument MCP tool functions with metrics and tracing.
 
-    Wraps async tool functions to record execution time and success/error status.
-    Compatible with @mcp.tool() and @require_scopes() decorators.
+    Wraps async tool functions to record execution time, success/error status, and
+    create OpenTelemetry trace spans. Compatible with @mcp.tool() and @require_scopes()
+    decorators.
 
     Usage:
         @mcp.tool()
@@ -420,24 +421,46 @@ def instrument_tool(func):
         func: The async function to instrument
 
     Returns:
-        Wrapped function with metrics instrumentation
+        Wrapped function with metrics and tracing instrumentation
     """
     import functools
     import time
 
+    from nextcloud_mcp_server.observability.tracing import trace_operation
+
     @functools.wraps(func)
     async def wrapper(*args, **kwargs):
         tool_name = func.__name__
         start_time = time.time()
-        try:
-            result = await func(*args, **kwargs)
-            duration = time.time() - start_time
-            record_tool_call(tool_name, duration, "success")
-            return result
-        except Exception as e:
-            duration = time.time() - start_time
-            record_tool_call(tool_name, duration, "error")
-            record_tool_error(tool_name, type(e).__name__)
-            raise
+
+        # Extract tool arguments for tracing (sanitize sensitive fields)
+        # kwargs contains the actual arguments passed to the tool
+        tool_args = {
+            k: v
+            for k, v in kwargs.items()
+            if k not in ("password", "token", "secret", "api_key", "etag", "ctx")
+        }
+
+        # Create trace span with metrics collection
+        with trace_operation(
+            f"mcp.tool.{tool_name}",
+            attributes={
+                "mcp.tool.name": tool_name,
+                "mcp.tool.args": str(tool_args)[:500]
+                if tool_args
+                else None,  # Limit to 500 chars
+            },
+            record_exception=True,
+        ):
+            try:
+                result = await func(*args, **kwargs)
+                duration = time.time() - start_time
+                record_tool_call(tool_name, duration, "success")
+                return result
+            except Exception as e:
+                duration = time.time() - start_time
+                record_tool_call(tool_name, duration, "error")
+                record_tool_error(tool_name, type(e).__name__)
+                raise
 
     return wrapper

nextcloud_mcp_server/providers/__init__.py

@@ -0,0 +1,20 @@
+"""Unified provider infrastructure for embeddings and text generation."""
+
+from .anthropic import AnthropicProvider
+from .base import Provider
+from .bedrock import BedrockProvider
+from .ollama import OllamaProvider
+from .openai import OpenAIProvider
+from .registry import get_provider, reset_provider
+from .simple import SimpleProvider
+
+__all__ = [
+    "Provider",
+    "OllamaProvider",
+    "OpenAIProvider",
+    "AnthropicProvider",
+    "SimpleProvider",
+    "BedrockProvider",
+    "get_provider",
+    "reset_provider",
+]

nextcloud_mcp_server/providers/anthropic.py

@@ -0,0 +1,99 @@
+"""Unified Anthropic provider for text generation."""
+
+import logging
+
+from anthropic import AsyncAnthropic
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class AnthropicProvider(Provider):
+    """
+    Anthropic provider for text generation.
+
+    Supports Claude models via the Anthropic API.
+    Note: Anthropic doesn't provide embedding models, only text generation.
+    """
+
+    def __init__(
+        self, api_key: str, generation_model: str = "claude-3-5-sonnet-20241022"
+    ):
+        """
+        Initialize Anthropic provider.
+
+        Args:
+            api_key: Anthropic API key
+            generation_model: Model name (e.g., "claude-3-5-sonnet-20241022")
+        """
+        self.client = AsyncAnthropic(api_key=api_key)
+        self.model = generation_model
+
+        logger.info(f"Initialized Anthropic provider (model={self.model})")
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return False
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return True
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text using Anthropic API.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+        """
+        message = await self.client.messages.create(
+            model=self.model,
+            max_tokens=max_tokens,
+            temperature=0.7,
+            messages=[{"role": "user", "content": prompt}],
+        )
+        return message.content[0].text
+
+    async def close(self) -> None:
+        """Close the client (no-op for Anthropic SDK)."""
+        pass

nextcloud_mcp_server/providers/base.py

@@ -0,0 +1,91 @@
+"""Unified provider interface for embeddings and text generation."""
+
+from abc import ABC, abstractmethod
+
+
+class Provider(ABC):
+    """
+    Unified base class for LLM providers.
+
+    Providers can support embeddings, text generation, or both.
+    Use capability properties to determine what features are available.
+    """
+
+    @property
+    @abstractmethod
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        pass
+
+    @property
+    @abstractmethod
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        pass
+
+    @abstractmethod
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts (optimized).
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension for this provider.
+
+        Returns:
+            Vector dimension (e.g., 768 for nomic-embed-text)
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If provider doesn't support generation
+        """
+        pass
+
+    @abstractmethod
+    async def close(self) -> None:
+        """Close the provider and release resources."""
+        pass

nextcloud_mcp_server/providers/bedrock.py

@@ -0,0 +1,397 @@
+"""Amazon Bedrock provider for embeddings and text generation."""
+
+import json
+import logging
+from typing import Any
+
+try:
+    import boto3
+    from botocore.exceptions import BotoCoreError, ClientError
+
+    BOTO3_AVAILABLE = True
+except ImportError:
+    BOTO3_AVAILABLE = False
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class BedrockProvider(Provider):
+    """
+    Amazon Bedrock provider supporting both embeddings and text generation.
+
+    Uses AWS Bedrock Runtime API with boto3. Supports various model families:
+    - Embeddings: amazon.titan-embed-text-v1, amazon.titan-embed-text-v2, cohere.embed-*
+    - Text Generation: anthropic.claude-*, meta.llama3-*, amazon.titan-text-*, mistral.*, etc.
+
+    Requires AWS credentials configured via:
+    - Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION)
+    - AWS credentials file (~/.aws/credentials)
+    - IAM role (when running on AWS)
+    """
+
+    def __init__(
+        self,
+        region_name: str | None = None,
+        embedding_model: str | None = None,
+        generation_model: str | None = None,
+        aws_access_key_id: str | None = None,
+        aws_secret_access_key: str | None = None,
+    ):
+        """
+        Initialize Bedrock provider.
+
+        Args:
+            region_name: AWS region (e.g., "us-east-1"). Defaults to AWS_REGION env var.
+            embedding_model: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0").
+                None disables embeddings.
+            generation_model: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0").
+                None disables generation.
+            aws_access_key_id: AWS access key (optional, uses default credential chain if not provided)
+            aws_secret_access_key: AWS secret key (optional, uses default credential chain if not provided)
+
+        Raises:
+            ImportError: If boto3 is not installed
+        """
+        if not BOTO3_AVAILABLE:
+            raise ImportError(
+                "boto3 is required for Bedrock provider. Install with: pip install boto3"
+            )
+
+        self.embedding_model = embedding_model
+        self.generation_model = generation_model
+        self._dimension: int | None = None  # Detected dynamically
+
+        # Initialize bedrock-runtime client
+        client_kwargs: dict[str, Any] = {}
+        if region_name:
+            client_kwargs["region_name"] = region_name
+        if aws_access_key_id:
+            client_kwargs["aws_access_key_id"] = aws_access_key_id
+        if aws_secret_access_key:
+            client_kwargs["aws_secret_access_key"] = aws_secret_access_key
+
+        self.client = boto3.client("bedrock-runtime", **client_kwargs)
+
+        logger.info(
+            f"Initialized Bedrock provider in region {region_name or 'default'} "
+            f"(embedding_model={embedding_model}, generation_model={generation_model})"
+        )
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return self.embedding_model is not None
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return self.generation_model is not None
+
+    def _create_embedding_request(self, text: str) -> dict[str, Any]:
+        """
+        Create model-specific embedding request payload.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Request payload dict for the embedding model
+        """
+        if not self.embedding_model:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        # Titan Embed models
+        if self.embedding_model.startswith("amazon.titan-embed"):
+            return {"inputText": text}
+
+        # Cohere Embed models
+        elif self.embedding_model.startswith("cohere.embed"):
+            return {"texts": [text], "input_type": "search_document"}
+
+        # Unknown model - try Titan format as default
+        else:
+            logger.warning(
+                f"Unknown embedding model format for {self.embedding_model}, "
+                "using Titan format as default"
+            )
+            return {"inputText": text}
+
+    def _parse_embedding_response(self, response: dict[str, Any]) -> list[float]:
+        """
+        Parse model-specific embedding response.
+
+        Args:
+            response: Raw response from Bedrock
+
+        Returns:
+            Embedding vector as list of floats
+        """
+        # Titan Embed models
+        if self.embedding_model and self.embedding_model.startswith(
+            "amazon.titan-embed"
+        ):
+            return response["embedding"]
+
+        # Cohere Embed models
+        elif self.embedding_model and self.embedding_model.startswith("cohere.embed"):
+            return response["embeddings"][0]
+
+        # Unknown model - try Titan format as default
+        else:
+            logger.warning(
+                f"Unknown embedding response format for {self.embedding_model}, "
+                "trying Titan format"
+            )
+            return response.get("embedding", response.get("embeddings", [None])[0])
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        try:
+            request_body = self._create_embedding_request(text)
+
+            response = self.client.invoke_model(
+                modelId=self.embedding_model,
+                body=json.dumps(request_body),
+                accept="application/json",
+                contentType="application/json",
+            )
+
+            response_body = json.loads(response["body"].read())
+            embedding = self._parse_embedding_response(response_body)
+
+            return embedding
+
+        except (BotoCoreError, ClientError) as e:
+            logger.error(f"Bedrock embedding error: {e}")
+            raise
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts.
+
+        Note: Current implementation sends requests sequentially.
+        Future optimization could use asyncio for concurrent requests.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        embeddings = []
+        for text in texts:
+            embedding = await self.embed(text)
+            embeddings.append(embedding)
+        return embeddings
+
+    async def _detect_dimension(self):
+        """
+        Detect embedding dimension by generating a test embedding.
+        """
+        if self._dimension is None and self.supports_embeddings:
+            logger.debug(
+                f"Detecting embedding dimension for model {self.embedding_model}..."
+            )
+            test_embedding = await self.embed("test")
+            self._dimension = len(test_embedding)
+            logger.info(
+                f"Detected embedding dimension: {self._dimension} "
+                f"for model {self.embedding_model}"
+            )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Returns:
+            Vector dimension for the configured embedding model
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            RuntimeError: If dimension not detected yet (call _detect_dimension first)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if self._dimension is None:
+            raise RuntimeError(
+                f"Embedding dimension not detected yet for model {self.embedding_model}. "
+                "Call _detect_dimension() first or generate an embedding."
+            )
+        return self._dimension
+
+    def _create_generation_request(
+        self, prompt: str, max_tokens: int
+    ) -> dict[str, Any]:
+        """
+        Create model-specific text generation request payload.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Request payload dict for the generation model
+        """
+        if not self.generation_model:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        # Anthropic Claude models
+        if self.generation_model.startswith("anthropic.claude"):
+            return {
+                "anthropic_version": "bedrock-2023-05-31",
+                "max_tokens": max_tokens,
+                "temperature": 0.7,
+                "messages": [{"role": "user", "content": prompt}],
+            }
+
+        # Meta Llama models
+        elif self.generation_model.startswith("meta.llama"):
+            return {"prompt": prompt, "max_gen_len": max_tokens, "temperature": 0.7}
+
+        # Amazon Titan Text models
+        elif self.generation_model.startswith("amazon.titan-text"):
+            return {
+                "inputText": prompt,
+                "textGenerationConfig": {
+                    "maxTokenCount": max_tokens,
+                    "temperature": 0.7,
+                },
+            }
+
+        # Mistral models
+        elif self.generation_model.startswith("mistral"):
+            return {"prompt": prompt, "max_tokens": max_tokens, "temperature": 0.7}
+
+        # Unknown model - try Claude format as default
+        else:
+            logger.warning(
+                f"Unknown generation model format for {self.generation_model}, "
+                "using Claude format as default"
+            )
+            return {
+                "anthropic_version": "bedrock-2023-05-31",
+                "max_tokens": max_tokens,
+                "temperature": 0.7,
+                "messages": [{"role": "user", "content": prompt}],
+            }
+
+    def _parse_generation_response(self, response: dict[str, Any]) -> str:
+        """
+        Parse model-specific text generation response.
+
+        Args:
+            response: Raw response from Bedrock
+
+        Returns:
+            Generated text
+        """
+        # Anthropic Claude models
+        if self.generation_model and self.generation_model.startswith(
+            "anthropic.claude"
+        ):
+            return response["content"][0]["text"]
+
+        # Meta Llama models
+        elif self.generation_model and self.generation_model.startswith("meta.llama"):
+            return response["generation"]
+
+        # Amazon Titan Text models
+        elif self.generation_model and self.generation_model.startswith(
+            "amazon.titan-text"
+        ):
+            return response["results"][0]["outputText"]
+
+        # Mistral models
+        elif self.generation_model and self.generation_model.startswith("mistral"):
+            return response["outputs"][0]["text"]
+
+        # Unknown model - try common response fields
+        else:
+            logger.warning(
+                f"Unknown generation response format for {self.generation_model}, "
+                "trying common fields"
+            )
+            # Try common response field names
+            for field in ["text", "generation", "outputText", "completion"]:
+                if field in response:
+                    return response[field]
+            # Last resort: return JSON string
+            return json.dumps(response)
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If generation not enabled (no generation_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_generation:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        try:
+            request_body = self._create_generation_request(prompt, max_tokens)
+
+            response = self.client.invoke_model(
+                modelId=self.generation_model,
+                body=json.dumps(request_body),
+                accept="application/json",
+                contentType="application/json",
+            )
+
+            response_body = json.loads(response["body"].read())
+            text = self._parse_generation_response(response_body)
+
+            return text
+
+        except (BotoCoreError, ClientError) as e:
+            logger.error(f"Bedrock generation error: {e}")
+            raise
+
+    async def close(self) -> None:
+        """Close the client (no-op for boto3 clients)."""
+        pass

nextcloud_mcp_server/providers/ollama.py

@@ -0,0 +1,234 @@
+"""Unified Ollama provider for embeddings and text generation."""
+
+import logging
+
+import httpx
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class OllamaProvider(Provider):
+    """
+    Ollama provider supporting both embeddings and text generation.
+
+    Supports TLS, SSL verification, and automatic model loading.
+    """
+
+    def __init__(
+        self,
+        base_url: str,
+        embedding_model: str | None = None,
+        generation_model: str | None = None,
+        verify_ssl: bool = True,
+        timeout: httpx.Timeout | None = None,
+    ):
+        """
+        Initialize Ollama provider.
+
+        Args:
+            base_url: Ollama API base URL (e.g., https://ollama.internal.example.com:443)
+            embedding_model: Model for embeddings (e.g., "nomic-embed-text"). None disables embeddings.
+            generation_model: Model for text generation (e.g., "llama3.2:1b"). None disables generation.
+            verify_ssl: Verify SSL certificates (default: True)
+            timeout: HTTP timeout configuration
+        """
+        self.base_url = base_url.rstrip("/")
+        self.embedding_model = embedding_model
+        self.generation_model = generation_model
+        self.verify_ssl = verify_ssl
+
+        if timeout is None:
+            timeout = httpx.Timeout(timeout=120, connect=5)
+
+        self.client = httpx.AsyncClient(verify=verify_ssl, timeout=timeout)
+        self._dimension: int | None = None  # Detected dynamically for embeddings
+
+        logger.info(
+            f"Initialized Ollama provider: {base_url} "
+            f"(embedding_model={embedding_model}, generation_model={generation_model}, "
+            f"verify_ssl={verify_ssl})"
+        )
+
+        # Pre-check and auto-load models
+        if embedding_model:
+            self._check_model_is_loaded(embedding_model, autoload=True)
+        if generation_model:
+            self._check_model_is_loaded(generation_model, autoload=True)
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return self.embedding_model is not None
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return self.generation_model is not None
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        response = await self.client.post(
+            f"{self.base_url}/api/embeddings",
+            json={"model": self.embedding_model, "prompt": text},
+        )
+        response.raise_for_status()
+        return response.json()["embedding"]
+
+    async def embed_batch(
+        self, texts: list[str], batch_size: int = 32
+    ) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts using Ollama's batch API.
+
+        Uses /api/embed endpoint with array input for efficient batch processing.
+        Conservative batch size (32) prevents quality degradation observed in
+        Ollama issue #6262 with larger batches.
+
+        Note: Ollama processes batches serially, not in parallel.
+
+        Args:
+            texts: List of texts to embed
+            batch_size: Maximum texts per batch (default: 32)
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        all_embeddings = []
+        for i in range(0, len(texts), batch_size):
+            batch = texts[i : i + batch_size]
+            response = await self.client.post(
+                f"{self.base_url}/api/embed",
+                json={"model": self.embedding_model, "input": batch},
+            )
+            response.raise_for_status()
+            all_embeddings.extend(response.json()["embeddings"])
+
+        return all_embeddings
+
+    async def _detect_dimension(self):
+        """
+        Detect embedding dimension by generating a test embedding.
+
+        This method queries the model to determine the actual dimension
+        instead of relying on hardcoded values.
+        """
+        if self._dimension is None and self.supports_embeddings:
+            logger.debug(
+                f"Detecting embedding dimension for model {self.embedding_model}..."
+            )
+            test_embedding = await self.embed("test")
+            self._dimension = len(test_embedding)
+            logger.info(
+                f"Detected embedding dimension: {self._dimension} "
+                f"for model {self.embedding_model}"
+            )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Returns:
+            Vector dimension for the configured embedding model
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            RuntimeError: If dimension not detected yet (call _detect_dimension first)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if self._dimension is None:
+            raise RuntimeError(
+                f"Embedding dimension not detected yet for model {self.embedding_model}. "
+                "Call _detect_dimension() first or generate an embedding."
+            )
+        return self._dimension
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If generation not enabled (no generation_model)
+        """
+        if not self.supports_generation:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        response = await self.client.post(
+            f"{self.base_url}/api/generate",
+            json={
+                "model": self.generation_model,
+                "prompt": prompt,
+                "stream": False,
+                "options": {
+                    "num_predict": max_tokens,
+                    "temperature": 0.7,
+                },
+            },
+        )
+        response.raise_for_status()
+        data = response.json()
+        return data["response"]
+
+    def _check_model_is_loaded(self, model: str, autoload: bool = True):
+        """
+        Check if model is loaded in Ollama, optionally auto-loading it.
+
+        Args:
+            model: Model name to check
+            autoload: Whether to automatically pull the model if not loaded
+        """
+        response = httpx.get(f"{self.base_url}/api/tags")
+        response.raise_for_status()
+
+        models = [m["name"] for m in response.json().get("models", [])]
+        logger.info("Ollama has following models pre-loaded: %s", models)
+
+        if (model not in models) and autoload:
+            logger.warning(
+                "Model '%s' not yet available in ollama, attempting to pull now...",
+                model,
+            )
+            response = httpx.post(f"{self.base_url}/api/pull", json={"model": model})
+            response.raise_for_status()
+
+    async def close(self) -> None:
+        """Close HTTP client."""
+        await self.client.aclose()

nextcloud_mcp_server/providers/openai.py

@@ -0,0 +1,271 @@
+"""Unified OpenAI provider for embeddings and text generation.
+
+Supports:
+- OpenAI's standard API
+- GitHub Models API (models.github.ai)
+- Any OpenAI-compatible API via base_url override
+"""
+
+import logging
+from functools import wraps
+
+import anyio
+from openai import AsyncOpenAI, RateLimitError
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+# Rate limit retry configuration
+MAX_RETRIES = 5
+INITIAL_RETRY_DELAY = 2.0  # seconds
+MAX_RETRY_DELAY = 60.0  # seconds
+
+
+def retry_on_rate_limit(func):
+    """Decorator to retry on OpenAI rate limit errors with exponential backoff."""
+
+    @wraps(func)
+    async def wrapper(*args, **kwargs):
+        retry_delay = INITIAL_RETRY_DELAY
+        last_error: Exception | None = None
+
+        for attempt in range(1, MAX_RETRIES + 1):
+            try:
+                return await func(*args, **kwargs)
+            except RateLimitError as e:
+                last_error = e
+                if attempt < MAX_RETRIES:
+                    logger.warning(
+                        f"Rate limit hit (attempt {attempt}/{MAX_RETRIES}), "
+                        f"retrying in {retry_delay:.1f}s..."
+                    )
+                    await anyio.sleep(retry_delay)
+                    retry_delay = min(retry_delay * 2, MAX_RETRY_DELAY)
+
+        logger.error(f"Rate limit exceeded after {MAX_RETRIES} attempts")
+        raise last_error  # type: ignore[misc]
+
+    return wrapper
+
+
+# Well-known embedding dimensions for OpenAI models
+OPENAI_EMBEDDING_DIMENSIONS: dict[str, int] = {
+    "text-embedding-3-small": 1536,
+    "text-embedding-3-large": 3072,
+    "text-embedding-ada-002": 1536,
+    # GitHub Models API uses openai/ prefix
+    "openai/text-embedding-3-small": 1536,
+    "openai/text-embedding-3-large": 3072,
+}
+
+
+class OpenAIProvider(Provider):
+    """
+    OpenAI provider supporting both embeddings and text generation.
+
+    Works with:
+    - OpenAI's standard API (api.openai.com)
+    - GitHub Models API (models.github.ai)
+    - Any OpenAI-compatible API (via base_url)
+    """
+
+    def __init__(
+        self,
+        api_key: str,
+        base_url: str | None = None,
+        embedding_model: str | None = None,
+        generation_model: str | None = None,
+        timeout: float = 120.0,
+    ):
+        """
+        Initialize OpenAI provider.
+
+        Args:
+            api_key: OpenAI API key (or GITHUB_TOKEN for GitHub Models)
+            base_url: Base URL override (e.g., "https://models.github.ai/inference")
+            embedding_model: Model for embeddings (e.g., "text-embedding-3-small").
+                            None disables embeddings.
+            generation_model: Model for text generation (e.g., "gpt-4o-mini").
+                             None disables generation.
+            timeout: HTTP timeout in seconds (default: 120)
+        """
+        self.embedding_model = embedding_model
+        self.generation_model = generation_model
+        self._dimension: int | None = None
+
+        # Initialize async client
+        self.client = AsyncOpenAI(
+            api_key=api_key,
+            base_url=base_url,
+            timeout=timeout,
+        )
+
+        # Try to get known dimension without API call
+        if embedding_model and embedding_model in OPENAI_EMBEDDING_DIMENSIONS:
+            self._dimension = OPENAI_EMBEDDING_DIMENSIONS[embedding_model]
+
+        logger.info(
+            f"Initialized OpenAI provider: base_url={base_url or 'default'} "
+            f"(embedding_model={embedding_model}, generation_model={generation_model}, "
+            f"dimension={self._dimension})"
+        )
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return self.embedding_model is not None
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return self.generation_model is not None
+
+    @retry_on_rate_limit
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        assert self.embedding_model is not None  # Type narrowing
+        response = await self.client.embeddings.create(
+            input=text,
+            model=self.embedding_model,
+        )
+
+        embedding = response.data[0].embedding
+
+        # Update dimension if not set
+        if self._dimension is None:
+            self._dimension = len(embedding)
+            logger.info(
+                f"Detected embedding dimension: {self._dimension} "
+                f"for model {self.embedding_model}"
+            )
+
+        return embedding
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts using OpenAI's batch API.
+
+        OpenAI supports up to 2048 inputs per request.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if not texts:
+            return []
+
+        # OpenAI supports batches up to 2048, but use smaller batches for safety
+        batch_size = 100
+        all_embeddings: list[list[float]] = []
+
+        for i in range(0, len(texts), batch_size):
+            batch = texts[i : i + batch_size]
+
+            # Use helper method with retry logic for each batch
+            batch_embeddings = await self._embed_batch_request(batch)
+            all_embeddings.extend(batch_embeddings)
+
+            # Update dimension if not set
+            if self._dimension is None and batch_embeddings:
+                self._dimension = len(batch_embeddings[0])
+                logger.info(
+                    f"Detected embedding dimension: {self._dimension} "
+                    f"for model {self.embedding_model}"
+                )
+
+        return all_embeddings
+
+    @retry_on_rate_limit
+    async def _embed_batch_request(self, batch: list[str]) -> list[list[float]]:
+        """Make a single batch embedding request with retry logic."""
+        assert self.embedding_model is not None  # Type narrowing
+        response = await self.client.embeddings.create(
+            input=batch,
+            model=self.embedding_model,
+        )
+        # Sort by index to maintain order
+        sorted_data = sorted(response.data, key=lambda x: x.index)
+        return [item.embedding for item in sorted_data]
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Returns:
+            Vector dimension for the configured embedding model
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            RuntimeError: If dimension not detected yet (call embed first)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if self._dimension is None:
+            raise RuntimeError(
+                f"Embedding dimension not detected yet for model {self.embedding_model}. "
+                "Call embed() first or use a known model."
+            )
+        return self._dimension
+
+    @retry_on_rate_limit
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If generation not enabled (no generation_model)
+        """
+        if not self.supports_generation:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        response = await self.client.chat.completions.create(
+            model=self.generation_model,
+            messages=[{"role": "user", "content": prompt}],
+            max_tokens=max_tokens,
+            temperature=0.7,
+        )
+
+        return response.choices[0].message.content or ""
+
+    async def close(self) -> None:
+        """Close HTTP client."""
+        await self.client.close()

nextcloud_mcp_server/providers/registry.py

@@ -0,0 +1,156 @@
+"""Provider registry and factory for auto-detection and instantiation."""
+
+import logging
+import os
+
+from .base import Provider
+from .bedrock import BedrockProvider
+from .ollama import OllamaProvider
+from .openai import OpenAIProvider
+from .simple import SimpleProvider
+
+logger = logging.getLogger(__name__)
+
+
+class ProviderRegistry:
+    """
+    Registry for provider auto-detection and instantiation.
+
+    Checks environment variables in priority order and creates appropriate provider:
+    1. Bedrock (AWS_REGION + BEDROCK_*_MODEL)
+    2. OpenAI (OPENAI_API_KEY)
+    3. Ollama (OLLAMA_BASE_URL)
+    4. Simple (fallback for testing/development)
+    """
+
+    @staticmethod
+    def create_provider() -> Provider:
+        """
+        Auto-detect and create provider based on environment variables.
+
+        Priority order:
+        1. Bedrock - if AWS_REGION or BEDROCK_EMBEDDING_MODEL is set
+        2. OpenAI - if OPENAI_API_KEY is set
+        3. Ollama - if OLLAMA_BASE_URL is set
+        4. Simple - fallback for testing/development
+
+        Returns:
+            Provider instance
+
+        Environment Variables:
+            Bedrock:
+                - AWS_REGION: AWS region (e.g., "us-east-1")
+                - AWS_ACCESS_KEY_ID: AWS access key (optional, uses credential chain)
+                - AWS_SECRET_ACCESS_KEY: AWS secret key (optional)
+                - BEDROCK_EMBEDDING_MODEL: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0")
+                - BEDROCK_GENERATION_MODEL: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+
+            OpenAI:
+                - OPENAI_API_KEY: OpenAI API key (or GITHUB_TOKEN for GitHub Models)
+                - OPENAI_BASE_URL: Base URL override (e.g., "https://models.github.ai/inference")
+                - OPENAI_EMBEDDING_MODEL: Model for embeddings (default: "text-embedding-3-small")
+                - OPENAI_GENERATION_MODEL: Model for text generation (e.g., "gpt-4o-mini")
+
+            Ollama:
+                - OLLAMA_BASE_URL: Ollama API base URL (e.g., "http://localhost:11434")
+                - OLLAMA_EMBEDDING_MODEL: Model for embeddings (default: "nomic-embed-text")
+                - OLLAMA_GENERATION_MODEL: Model for text generation (e.g., "llama3.2:1b")
+                - OLLAMA_VERIFY_SSL: Verify SSL certificates (default: "true")
+
+            Simple (no configuration needed, fallback):
+                - SIMPLE_EMBEDDING_DIMENSION: Embedding dimension (default: 384)
+        """
+        # 1. Check for Bedrock
+        aws_region = os.getenv("AWS_REGION")
+        bedrock_embedding_model = os.getenv("BEDROCK_EMBEDDING_MODEL")
+        bedrock_generation_model = os.getenv("BEDROCK_GENERATION_MODEL")
+
+        if aws_region or bedrock_embedding_model or bedrock_generation_model:
+            logger.info(
+                f"Using Bedrock provider: region={aws_region}, "
+                f"embedding_model={bedrock_embedding_model}, "
+                f"generation_model={bedrock_generation_model}"
+            )
+            return BedrockProvider(
+                region_name=aws_region,
+                embedding_model=bedrock_embedding_model,
+                generation_model=bedrock_generation_model,
+                aws_access_key_id=os.getenv("AWS_ACCESS_KEY_ID"),
+                aws_secret_access_key=os.getenv("AWS_SECRET_ACCESS_KEY"),
+            )
+
+        # 2. Check for OpenAI
+        openai_api_key = os.getenv("OPENAI_API_KEY")
+        if openai_api_key:
+            base_url = os.getenv("OPENAI_BASE_URL")
+            embedding_model = os.getenv(
+                "OPENAI_EMBEDDING_MODEL", "text-embedding-3-small"
+            )
+            generation_model = os.getenv("OPENAI_GENERATION_MODEL")
+
+            logger.info(
+                f"Using OpenAI provider: base_url={base_url or 'default'}, "
+                f"embedding_model={embedding_model}, "
+                f"generation_model={generation_model}"
+            )
+            return OpenAIProvider(
+                api_key=openai_api_key,
+                base_url=base_url,
+                embedding_model=embedding_model,
+                generation_model=generation_model,
+            )
+
+        # 3. Check for Ollama (local LLM)
+        ollama_url = os.getenv("OLLAMA_BASE_URL")
+        if ollama_url:
+            embedding_model = os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text")
+            generation_model = os.getenv("OLLAMA_GENERATION_MODEL")
+            verify_ssl = os.getenv("OLLAMA_VERIFY_SSL", "true").lower() == "true"
+
+            logger.info(
+                f"Using Ollama provider: {ollama_url}, "
+                f"embedding_model={embedding_model}, "
+                f"generation_model={generation_model}"
+            )
+            return OllamaProvider(
+                base_url=ollama_url,
+                embedding_model=embedding_model,
+                generation_model=generation_model,
+                verify_ssl=verify_ssl,
+            )
+
+        # 4. Fallback to Simple provider for development/testing
+        dimension = int(os.getenv("SIMPLE_EMBEDDING_DIMENSION", "384"))
+        logger.warning(
+            "No provider configured (AWS_REGION, OPENAI_API_KEY, OLLAMA_BASE_URL not set). "
+            "Using SimpleProvider for testing/development. "
+            "For production, configure Bedrock, OpenAI, or Ollama."
+        )
+        return SimpleProvider(dimension=dimension)
+
+
+# Singleton instance
+_provider: Provider | None = None
+
+
+def get_provider() -> Provider:
+    """
+    Get singleton provider instance.
+
+    Returns:
+        Global Provider instance (auto-detected on first call)
+    """
+    global _provider
+    if _provider is None:
+        _provider = ProviderRegistry.create_provider()
+    return _provider
+
+
+def reset_provider():
+    """
+    Reset singleton provider instance.
+
+    Useful for testing or reconfiguration.
+    """
+    global _provider
+    _provider = None

nextcloud_mcp_server/providers/simple.py

@@ -0,0 +1,149 @@
+"""Simple in-process embedding provider for testing.
+
+This provider uses a basic TF-IDF-like approach with feature hashing to generate
+deterministic embeddings without requiring external services. Suitable for testing
+but not for production use.
+"""
+
+import hashlib
+import math
+import re
+from collections import Counter
+
+from .base import Provider
+
+
+class SimpleProvider(Provider):
+    """Simple deterministic embedding provider using feature hashing.
+
+    This implementation:
+    - Tokenizes text into words
+    - Uses feature hashing to map words to fixed-size vectors
+    - Applies TF-IDF-like weighting
+    - Normalizes vectors to unit length
+
+    Not suitable for production but good for testing semantic search infrastructure.
+    Only supports embeddings, not text generation.
+    """
+
+    def __init__(self, dimension: int = 384):
+        """Initialize simple embedding provider.
+
+        Args:
+            dimension: Embedding dimension (default: 384)
+        """
+        self.dimension = dimension
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return True
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return False
+
+    def _tokenize(self, text: str) -> list[str]:
+        """Tokenize text into lowercase words.
+
+        Args:
+            text: Input text
+
+        Returns:
+            List of lowercase word tokens
+        """
+        # Simple word tokenization
+        text = text.lower()
+        words = re.findall(r"\b\w+\b", text)
+        return words
+
+    def _hash_word(self, word: str) -> int:
+        """Hash word to dimension index.
+
+        Args:
+            word: Word to hash
+
+        Returns:
+            Index in range [0, dimension)
+        """
+        hash_bytes = hashlib.md5(word.encode()).digest()
+        hash_int = int.from_bytes(hash_bytes[:4], byteorder="big")
+        return hash_int % self.dimension
+
+    def _embed_single(self, text: str) -> list[float]:
+        """Generate embedding for single text.
+
+        Args:
+            text: Input text
+
+        Returns:
+            Normalized embedding vector
+        """
+        tokens = self._tokenize(text)
+        if not tokens:
+            return [0.0] * self.dimension
+
+        # Count term frequencies
+        term_freq = Counter(tokens)
+
+        # Initialize vector
+        vector = [0.0] * self.dimension
+
+        # Apply TF weighting with feature hashing
+        for word, count in term_freq.items():
+            idx = self._hash_word(word)
+            # Simple TF weighting: log(1 + count)
+            vector[idx] += math.log1p(count)
+
+        # Normalize to unit length
+        norm = math.sqrt(sum(x * x for x in vector))
+        if norm > 0:
+            vector = [x / norm for x in vector]
+
+        return vector
+
+    async def embed(self, text: str) -> list[float]:
+        """Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+        """
+        return self._embed_single(text)
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """Generate embeddings for multiple texts.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+        """
+        return [self._embed_single(text) for text in texts]
+
+    def get_dimension(self) -> int:
+        """Get embedding dimension.
+
+        Returns:
+            Vector dimension
+        """
+        return self.dimension
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Raises:
+            NotImplementedError: Simple provider doesn't support text generation
+        """
+        raise NotImplementedError(
+            "Text generation not supported by Simple provider - use Ollama, Anthropic, or Bedrock"
+        )
+
+    async def close(self) -> None:
+        """Close the provider (no-op for simple provider)."""
+        pass

nextcloud_mcp_server/search/__init__.py

@@ -1,13 +1,11 @@
-"""Search algorithms module for unified multi-algorithm search.
+"""Search algorithms module for BM25 hybrid search.
 
-This module provides a unified interface for different search algorithms:
-- Semantic search (vector similarity)
-- Keyword search (token-based matching)
-- Fuzzy search (character overlap)
-- Hybrid search (RRF fusion of multiple algorithms)
+This module provides BM25 hybrid search combining:
+- Dense semantic vectors (vector similarity via embeddings)
+- Sparse BM25 vectors (keyword-based retrieval)
 
-All algorithms share the same interface and can be used interchangeably by both
-MCP tools and the visualization pane.
+Results are fused using Qdrant's native Reciprocal Rank Fusion (RRF) for
+optimal relevance across both semantic and keyword queries.
 """
 
 from nextcloud_mcp_server.search.algorithms import (
@@ -16,9 +14,7 @@ from nextcloud_mcp_server.search.algorithms import (
     SearchResult,
     get_indexed_doc_types,
 )
-from nextcloud_mcp_server.search.fuzzy import FuzzySearchAlgorithm
-from nextcloud_mcp_server.search.hybrid import HybridSearchAlgorithm
-from nextcloud_mcp_server.search.keyword import KeywordSearchAlgorithm
+from nextcloud_mcp_server.search.bm25_hybrid import BM25HybridSearchAlgorithm
 from nextcloud_mcp_server.search.semantic import SemanticSearchAlgorithm
 
 __all__ = [
@@ -27,7 +23,5 @@ __all__ = [
     "SearchResult",
     "get_indexed_doc_types",
     "SemanticSearchAlgorithm",
-    "KeywordSearchAlgorithm",
-    "FuzzySearchAlgorithm",
-    "HybridSearchAlgorithm",
+    "BM25HybridSearchAlgorithm",
 ]

nextcloud_mcp_server/search/algorithms.py

@@ -83,6 +83,7 @@ async def get_indexed_doc_types(user_id: str) -> set[str]:
     from qdrant_client.models import FieldCondition, Filter, MatchValue
 
     from nextcloud_mcp_server.config import get_settings
+    from nextcloud_mcp_server.vector.placeholder import get_placeholder_filter
     from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
     logger = logging.getLogger(__name__)
@@ -97,15 +98,18 @@ async def get_indexed_doc_types(user_id: str) -> set[str]:
         scroll_results, _next_offset = await qdrant_client.scroll(
             collection_name=collection,
             scroll_filter=Filter(
-                must=[FieldCondition(key="user_id", match=MatchValue(value=user_id))]
+                must=[
+                    get_placeholder_filter(),  # Exclude placeholders from doc_type discovery
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                ]
             ),
             limit=1000,  # Sample size to discover types
             with_payload=["doc_type"],
             with_vectors=False,  # Don't need vectors for type discovery
         )
 
-        doc_types = {
-            point.payload.get("doc_type")
+        doc_types: set[str] = {
+            str(point.payload.get("doc_type"))
             for point in scroll_results
             if point.payload.get("doc_type")
         }
@@ -123,12 +127,20 @@ class SearchResult:
     """A single search result with metadata and score.
 
     Attributes:
-        id: Document ID
+        id: Document ID (int for all document types)
         doc_type: Document type (note, file, calendar, contact, etc.)
         title: Document title
         excerpt: Content excerpt showing match context
-        score: Relevance score (0.0-1.0, higher is better)
+        score: Relevance score (≥ 0.0, higher is better)
+            - RRF fusion: scores in [0.0, 1.0]
+            - DBSF fusion: scores can exceed 1.0 (sum of normalized scores)
         metadata: Additional algorithm-specific metadata
+        chunk_start_offset: Character position where chunk starts (None if not available)
+        chunk_end_offset: Character position where chunk ends (None if not available)
+        page_number: Page number for PDF documents (None for other doc types)
+        chunk_index: Zero-based index of this chunk in the document
+        total_chunks: Total number of chunks in the document
+        point_id: Qdrant point ID for batch vector retrieval (None if not from Qdrant)
     """
 
     id: int
@@ -137,11 +149,24 @@ class SearchResult:
     excerpt: str
     score: float
     metadata: dict[str, Any] | None = None
+    chunk_start_offset: int | None = None
+    chunk_end_offset: int | None = None
+    page_number: int | None = None
+    chunk_index: int = 0
+    total_chunks: int = 1
+    point_id: str | None = None
 
     def __post_init__(self):
-        """Validate score is in valid range."""
-        if not 0.0 <= self.score <= 1.0:
-            raise ValueError(f"Score must be between 0.0 and 1.0, got {self.score}")
+        """Validate score is non-negative.
+
+        Note: Different fusion methods produce different score ranges:
+        - RRF (Reciprocal Rank Fusion): Bounded to [0.0, 1.0]
+        - DBSF (Distribution-Based Score Fusion): Unbounded (can exceed 1.0)
+          DBSF sums normalized scores from multiple systems, so scores can be
+          1.5, 2.0, etc. when multiple systems agree a document is highly relevant.
+        """
+        if self.score < 0.0:
+            raise ValueError(f"Score must be non-negative, got {self.score}")
 
 
 class SearchAlgorithm(ABC):
@@ -149,8 +174,15 @@ class SearchAlgorithm(ABC):
 
     All search algorithms must implement the search() method with consistent
     interface, allowing them to be used interchangeably.
+
+    Attributes:
+        query_embedding: The query embedding generated during the last search.
+            Available after search() completes for algorithms that use embeddings.
+            Can be reused by callers to avoid redundant embedding generation.
     """
 
+    query_embedding: list[float] | None = None
+
     @abstractmethod
     async def search(
         self,

nextcloud_mcp_server/search/bm25_hybrid.py

@@ -0,0 +1,255 @@
+"""BM25 hybrid search algorithm using Qdrant native RRF fusion."""
+
+import logging
+from typing import Any
+
+from qdrant_client import models
+from qdrant_client.models import FieldCondition, Filter, MatchValue
+
+from nextcloud_mcp_server.config import get_settings
+from nextcloud_mcp_server.embedding import get_bm25_service, get_embedding_service
+from nextcloud_mcp_server.observability.metrics import record_qdrant_operation
+from nextcloud_mcp_server.observability.tracing import trace_operation
+from nextcloud_mcp_server.search.algorithms import SearchAlgorithm, SearchResult
+from nextcloud_mcp_server.vector.placeholder import get_placeholder_filter
+from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+
+logger = logging.getLogger(__name__)
+
+
+class BM25HybridSearchAlgorithm(SearchAlgorithm):
+    """
+    Hybrid search combining dense semantic vectors with BM25 sparse vectors.
+
+    Uses Qdrant's native Reciprocal Rank Fusion (RRF) to automatically merge
+    results from both dense (semantic) and sparse (BM25 keyword) searches.
+    This provides the best of both worlds: semantic understanding for conceptual
+    queries and precise keyword matching for specific terms, acronyms, and codes.
+
+    The fusion happens efficiently in the database using the prefetch mechanism,
+    eliminating the need for application-layer result merging.
+    """
+
+    def __init__(self, score_threshold: float = 0.0, fusion: str = "rrf"):
+        """
+        Initialize BM25 hybrid search algorithm.
+
+        Args:
+            score_threshold: Minimum fusion score (0-1, default: 0.0 to allow fusion scoring)
+                           Note: Both RRF and DBSF produce normalized scores
+            fusion: Fusion algorithm to use: "rrf" (Reciprocal Rank Fusion, default)
+                   or "dbsf" (Distribution-Based Score Fusion)
+
+        Raises:
+            ValueError: If fusion is not "rrf" or "dbsf"
+        """
+        if fusion not in ("rrf", "dbsf"):
+            raise ValueError(
+                f"Invalid fusion algorithm '{fusion}'. Must be 'rrf' or 'dbsf'"
+            )
+
+        self.score_threshold = score_threshold
+        self.fusion = models.Fusion.RRF if fusion == "rrf" else models.Fusion.DBSF
+        self.fusion_name = fusion
+
+    @property
+    def name(self) -> str:
+        return "bm25_hybrid"
+
+    @property
+    def requires_vector_db(self) -> bool:
+        return True
+
+    async def search(
+        self,
+        query: str,
+        user_id: str,
+        limit: int = 10,
+        doc_type: str | None = None,
+        **kwargs: Any,
+    ) -> list[SearchResult]:
+        """
+        Execute hybrid search using dense + sparse vectors with native RRF fusion.
+
+        Returns unverified results from Qdrant. Access verification should be
+        performed separately at the final output stage using verify_search_results().
+
+        Deduplicates by (doc_id, doc_type, chunk_start_offset, chunk_end_offset)
+        to show multiple chunks from the same document while avoiding duplicate chunks.
+
+        Args:
+            query: Natural language or keyword search query
+            user_id: User ID for filtering
+            limit: Maximum results to return
+            doc_type: Optional document type filter
+            **kwargs: Additional parameters (score_threshold override)
+
+        Returns:
+            List of unverified SearchResult objects ranked by RRF fusion score
+
+        Raises:
+            McpError: If vector sync is not enabled or search fails
+        """
+        settings = get_settings()
+        score_threshold = kwargs.get("score_threshold", self.score_threshold)
+
+        logger.info(
+            f"BM25 hybrid search: query='{query}', user={user_id}, "
+            f"limit={limit}, score_threshold={score_threshold}, doc_type={doc_type}, "
+            f"fusion={self.fusion_name}"
+        )
+
+        # Generate dense embedding for semantic search
+        with trace_operation("search.get_embedding_service"):
+            embedding_service = get_embedding_service()
+        with trace_operation("search.dense_embedding"):
+            dense_embedding = await embedding_service.embed(query)
+        # Store for reuse by callers (e.g., viz_routes PCA visualization)
+        self.query_embedding = dense_embedding
+        logger.debug(f"Generated dense embedding (dimension={len(dense_embedding)})")
+
+        # Generate sparse embedding for BM25 keyword search
+        with trace_operation("search.get_bm25_service"):
+            bm25_service = get_bm25_service()
+        with trace_operation("search.sparse_embedding_bm25"):
+            sparse_embedding = await bm25_service.encode_async(query)
+        logger.debug(
+            f"Generated sparse embedding "
+            f"({len(sparse_embedding['indices'])} non-zero terms)"
+        )
+
+        # Build Qdrant filter
+        filter_conditions = [
+            get_placeholder_filter(),  # Always exclude placeholders from user-facing queries
+            FieldCondition(
+                key="user_id",
+                match=MatchValue(value=user_id),
+            ),
+        ]
+
+        # Add doc_type filter if specified
+        if doc_type:
+            filter_conditions.append(
+                FieldCondition(
+                    key="doc_type",
+                    match=MatchValue(value=doc_type),
+                )
+            )
+
+        query_filter = Filter(must=filter_conditions)
+
+        # Execute hybrid search with Qdrant native RRF fusion
+        with trace_operation("search.get_qdrant_client"):
+            qdrant_client = await get_qdrant_client()
+
+        try:
+            # Use prefetch to run both dense and sparse searches
+            # Qdrant will automatically merge results using RRF
+            with trace_operation(
+                "search.qdrant_query",
+                attributes={"query.limit": limit * 2, "query.fusion": self.fusion_name},
+            ):
+                search_response = await qdrant_client.query_points(
+                    collection_name=settings.get_collection_name(),
+                    prefetch=[
+                        # Dense semantic search
+                        models.Prefetch(
+                            query=dense_embedding,
+                            using="dense",
+                            limit=limit * 2,  # Get extra for deduplication
+                            filter=query_filter,
+                        ),
+                        # Sparse BM25 search
+                        models.Prefetch(
+                            query=models.SparseVector(
+                                indices=sparse_embedding["indices"],
+                                values=sparse_embedding["values"],
+                            ),
+                            using="sparse",
+                            limit=limit * 2,  # Get extra for deduplication
+                            filter=query_filter,
+                        ),
+                    ],
+                    # Fusion query (RRF or DBSF based on initialization)
+                    query=models.FusionQuery(fusion=self.fusion),
+                    limit=limit * 2,  # Get extra for deduplication
+                    score_threshold=score_threshold,
+                    with_payload=True,
+                    with_vectors=False,  # Don't return vectors to save bandwidth
+                )
+            record_qdrant_operation("search", "success")
+        except Exception:
+            record_qdrant_operation("search", "error")
+            raise
+
+        logger.info(
+            f"Qdrant {self.fusion_name.upper()} fusion returned {len(search_response.points)} results "
+            f"(before deduplication)"
+        )
+
+        if search_response.points:
+            # Log top 3 fusion scores to help with threshold tuning
+            top_scores = [p.score for p in search_response.points[:3]]
+            logger.debug(
+                f"Top 3 {self.fusion_name.upper()} fusion scores: {top_scores}"
+            )
+
+        # Deduplicate by (doc_id, doc_type, chunk_start, chunk_end)
+        # This allows multiple chunks from same doc, but removes duplicate chunks
+        with trace_operation(
+            "search.deduplicate",
+            attributes={"dedupe.num_points": len(search_response.points)},
+        ):
+            seen_chunks = set()
+            results = []
+
+            for result in search_response.points:
+                if result.payload is None:
+                    continue
+                # doc_id can be int (notes) or str (files - file paths)
+                doc_id = result.payload["doc_id"]
+                doc_type = result.payload.get("doc_type", "note")
+                chunk_start = result.payload.get("chunk_start_offset")
+                chunk_end = result.payload.get("chunk_end_offset")
+                chunk_key = (doc_id, doc_type, chunk_start, chunk_end)
+
+                # Skip if we've already seen this exact chunk
+                if chunk_key in seen_chunks:
+                    continue
+
+                seen_chunks.add(chunk_key)
+
+                # Return unverified results (verification happens at output stage)
+                results.append(
+                    SearchResult(
+                        id=doc_id,
+                        doc_type=doc_type,
+                        title=result.payload.get("title", "Untitled"),
+                        excerpt=result.payload.get("excerpt", ""),
+                        score=result.score,  # Fusion score (RRF or DBSF)
+                        metadata={
+                            "chunk_index": result.payload.get("chunk_index"),
+                            "total_chunks": result.payload.get("total_chunks"),
+                            "search_method": f"bm25_hybrid_{self.fusion_name}",
+                        },
+                        chunk_start_offset=result.payload.get("chunk_start_offset"),
+                        chunk_end_offset=result.payload.get("chunk_end_offset"),
+                        page_number=result.payload.get("page_number"),
+                        chunk_index=result.payload.get("chunk_index", 0),
+                        total_chunks=result.payload.get("total_chunks", 1),
+                        point_id=str(result.id),  # Qdrant point ID for batch retrieval
+                    )
+                )
+
+                if len(results) >= limit:
+                    break
+
+        logger.info(f"Returning {len(results)} unverified results after deduplication")
+        if results:
+            result_details = [
+                f"{r.doc_type}_{r.id} (score={r.score:.3f}, title='{r.title}')"
+                for r in results[:5]  # Show top 5
+            ]
+            logger.debug(f"Top results: {', '.join(result_details)}")
+
+        return results

nextcloud_mcp_server/search/context.py

@@ -0,0 +1,598 @@
+"""Context expansion for search results.
+
+Provides utilities to expand matched chunks with surrounding context and
+position markers for better visualization and understanding of search results.
+"""
+
+import logging
+from dataclasses import dataclass
+
+from nextcloud_mcp_server.client import NextcloudClient
+
+logger = logging.getLogger(__name__)
+
+
+async def _get_chunk_from_qdrant(
+    user_id: str, doc_id: int, doc_type: str, chunk_start: int, chunk_end: int
+) -> str | None:
+    """Retrieve full chunk text from Qdrant payload.
+
+    This avoids re-fetching and re-parsing documents by using the cached
+    chunk content already stored in Qdrant.
+
+    Args:
+        user_id: User ID who owns the document
+        doc_id: Document ID
+        doc_type: Document type (e.g., "note", "file")
+        chunk_start: Character offset where chunk starts
+        chunk_end: Character offset where chunk ends
+
+    Returns:
+        Full chunk text from Qdrant excerpt field, or None if not found
+    """
+    try:
+        from qdrant_client.models import FieldCondition, Filter, MatchValue
+
+        from nextcloud_mcp_server.config import get_settings
+        from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Query for the specific chunk
+        scroll_result = await qdrant_client.scroll(
+            collection_name=settings.get_collection_name(),
+            scroll_filter=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=doc_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value=doc_type)),
+                    FieldCondition(
+                        key="chunk_start_offset", match=MatchValue(value=chunk_start)
+                    ),
+                    FieldCondition(
+                        key="chunk_end_offset", match=MatchValue(value=chunk_end)
+                    ),
+                ]
+            ),
+            limit=1,
+            with_payload=["excerpt"],
+            with_vectors=False,
+        )
+
+        if scroll_result[0]:
+            point = scroll_result[0][0]
+            excerpt = point.payload.get("excerpt")
+            if excerpt:
+                logger.debug(
+                    f"Retrieved chunk from Qdrant for {doc_type} {doc_id}: "
+                    f"{len(excerpt)} chars"
+                )
+                return str(excerpt)
+
+        logger.debug(
+            f"Chunk not found in Qdrant for {doc_type} {doc_id}, "
+            f"chunk [{chunk_start}:{chunk_end}]. Will fall back to document fetch."
+        )
+        return None
+
+    except Exception as e:
+        logger.error(
+            f"Error querying Qdrant for chunk: {e}. Falling back to document fetch.",
+            exc_info=True,
+        )
+        return None
+
+
+async def _get_chunk_by_index_from_qdrant(
+    user_id: str, doc_id: int, doc_type: str, chunk_index: int
+) -> str | None:
+    """Retrieve chunk text by chunk_index from Qdrant payload.
+
+    Used to fetch adjacent chunks for context expansion.
+
+    Args:
+        user_id: User ID who owns the document
+        doc_id: Document ID
+        doc_type: Document type (e.g., "note", "file")
+        chunk_index: Zero-based chunk index in document
+
+    Returns:
+        Full chunk text from Qdrant excerpt field, or None if not found
+    """
+    try:
+        from qdrant_client.models import FieldCondition, Filter, MatchValue
+
+        from nextcloud_mcp_server.config import get_settings
+        from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Query for chunk by index
+        scroll_result = await qdrant_client.scroll(
+            collection_name=settings.get_collection_name(),
+            scroll_filter=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=doc_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value=doc_type)),
+                    FieldCondition(
+                        key="chunk_index", match=MatchValue(value=chunk_index)
+                    ),
+                ]
+            ),
+            limit=1,
+            with_payload=["excerpt"],
+            with_vectors=False,
+        )
+
+        if scroll_result[0]:
+            point = scroll_result[0][0]
+            excerpt = point.payload.get("excerpt")
+            if excerpt:
+                logger.debug(
+                    f"Retrieved adjacent chunk {chunk_index} from Qdrant for "
+                    f"{doc_type} {doc_id}: {len(excerpt)} chars"
+                )
+                return str(excerpt)
+
+        return None
+
+    except Exception as e:
+        logger.debug(
+            f"Could not retrieve adjacent chunk {chunk_index} for "
+            f"{doc_type} {doc_id}: {e}"
+        )
+        return None
+
+
+async def _get_file_path_from_qdrant(
+    user_id: str, file_id: int, chunk_start: int, chunk_end: int
+) -> str | None:
+    """Resolve file_id to file_path by querying Qdrant payload.
+
+    Args:
+        user_id: User ID who owns the file
+        file_id: Numeric file ID
+        chunk_start: Character offset where chunk starts
+        chunk_end: Character offset where chunk ends
+
+    Returns:
+        File path string, or None if not found in Qdrant
+    """
+    try:
+        from qdrant_client.models import FieldCondition, Filter, MatchValue
+
+        from nextcloud_mcp_server.config import get_settings
+        from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Query for the specific chunk
+        scroll_result = await qdrant_client.scroll(
+            collection_name=settings.get_collection_name(),
+            scroll_filter=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=file_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value="file")),
+                    FieldCondition(
+                        key="chunk_start_offset", match=MatchValue(value=chunk_start)
+                    ),
+                    FieldCondition(
+                        key="chunk_end_offset", match=MatchValue(value=chunk_end)
+                    ),
+                ]
+            ),
+            limit=1,
+            with_payload=["file_path"],
+            with_vectors=False,
+        )
+
+        if scroll_result[0]:
+            point = scroll_result[0][0]
+            file_path = point.payload.get("file_path")
+            if file_path:
+                logger.debug(f"Resolved file_id {file_id} to file_path {file_path}")
+                return str(file_path)
+
+        logger.warning(
+            f"Could not find file_path in Qdrant for file_id {file_id}, "
+            f"chunk [{chunk_start}:{chunk_end}]"
+        )
+        return None
+
+    except Exception as e:
+        logger.error(f"Error querying Qdrant for file_path: {e}", exc_info=True)
+        return None
+
+
+@dataclass
+class ChunkContext:
+    """Expanded chunk with surrounding context and position markers.
+
+    Attributes:
+        chunk_text: The matched chunk text
+        before_context: Text before the chunk (up to context_chars)
+        after_context: Text after the chunk (up to context_chars)
+        chunk_start_offset: Character position where chunk starts in document
+        chunk_end_offset: Character position where chunk ends in document
+        page_number: Page number for PDFs (None for other doc types)
+        chunk_index: Zero-based chunk index (N in "chunk N of M")
+        total_chunks: Total number of chunks in document
+        marked_text: Full text with position markers around the chunk
+        has_before_truncation: True if before_context was truncated
+        has_after_truncation: True if after_context was truncated
+    """
+
+    chunk_text: str
+    before_context: str
+    after_context: str
+    chunk_start_offset: int
+    chunk_end_offset: int
+    page_number: int | None
+    chunk_index: int
+    total_chunks: int
+    marked_text: str
+    has_before_truncation: bool
+    has_after_truncation: bool
+
+
+async def get_chunk_with_context(
+    nc_client: NextcloudClient,
+    user_id: str,
+    doc_id: str | int,
+    doc_type: str,
+    chunk_start: int,
+    chunk_end: int,
+    page_number: int | None = None,
+    chunk_index: int = 0,
+    total_chunks: int = 1,
+    context_chars: int = 300,
+) -> ChunkContext | None:
+    """Fetch chunk with surrounding context.
+
+    First tries to retrieve the chunk from Qdrant (fast, cached). If that fails
+    (e.g., legacy data with truncated excerpts), falls back to fetching and
+    parsing the full document (slower, for PDFs especially).
+
+    Args:
+        nc_client: Authenticated Nextcloud client
+        user_id: User ID who owns the document
+        doc_id: Document ID (int for notes/files)
+        doc_type: Type of document ("note", "file", etc.)
+        chunk_start: Character offset where chunk starts
+        chunk_end: Character offset where chunk ends
+        page_number: Optional page number for PDFs
+        chunk_index: Zero-based chunk index in document
+        total_chunks: Total number of chunks in document
+        context_chars: Number of characters to include before/after chunk
+
+    Returns:
+        ChunkContext with expanded context and markers, or None if document
+        cannot be retrieved
+    """
+    # Convert doc_id to int for Qdrant query
+    doc_id_int = (
+        int(doc_id)
+        if isinstance(doc_id, str) and doc_id.isdigit()
+        else (doc_id if isinstance(doc_id, int) else None)
+    )
+
+    # Try to get chunk from Qdrant first (fast path)
+    if doc_id_int is not None:
+        chunk_text = await _get_chunk_from_qdrant(
+            user_id, doc_id_int, doc_type, chunk_start, chunk_end
+        )
+        if chunk_text:
+            logger.info(
+                f"Retrieved chunk from Qdrant cache for {doc_type} {doc_id} "
+                f"(avoids document re-fetch/re-parse)"
+            )
+
+            # Fetch adjacent chunks for context expansion
+            # Get chunk overlap from config to remove duplicate text
+            from nextcloud_mcp_server.config import get_settings
+
+            settings = get_settings()
+            chunk_overlap = settings.document_chunk_overlap
+
+            before_context = ""
+            after_context = ""
+            has_before_truncation = False
+            has_after_truncation = False
+
+            # Fetch previous chunk if not first chunk
+            if chunk_index > 0:
+                before_chunk = await _get_chunk_by_index_from_qdrant(
+                    user_id, doc_id_int, doc_type, chunk_index - 1
+                )
+                if before_chunk:
+                    # Remove overlap: the last chunk_overlap chars of previous chunk
+                    # overlap with the first chunk_overlap chars of current chunk
+                    before_context = (
+                        before_chunk[:-chunk_overlap]
+                        if len(before_chunk) > chunk_overlap
+                        else ""
+                    )
+                    # Truncate if requested context_chars < remaining length
+                    if before_context and len(before_context) > context_chars:
+                        before_context = before_context[-context_chars:]
+                        has_before_truncation = True
+                else:
+                    # Could not fetch previous chunk, but we're not at start
+                    has_before_truncation = True
+
+            # Fetch next chunk if not last chunk
+            if chunk_index < total_chunks - 1:
+                after_chunk = await _get_chunk_by_index_from_qdrant(
+                    user_id, doc_id_int, doc_type, chunk_index + 1
+                )
+                if after_chunk:
+                    # Remove overlap: the first chunk_overlap chars of next chunk
+                    # overlap with the last chunk_overlap chars of current chunk
+                    after_context = (
+                        after_chunk[chunk_overlap:]
+                        if len(after_chunk) > chunk_overlap
+                        else ""
+                    )
+                    # Truncate if requested context_chars < remaining length
+                    if after_context and len(after_context) > context_chars:
+                        after_context = after_context[:context_chars]
+                        has_after_truncation = True
+                else:
+                    # Could not fetch next chunk, but we're not at end
+                    has_after_truncation = True
+
+            marked_text = _insert_position_markers(
+                before_context=before_context,
+                chunk_text=chunk_text,
+                after_context=after_context,
+                page_number=page_number,
+                chunk_index=chunk_index,
+                total_chunks=total_chunks,
+                has_before_truncation=has_before_truncation,
+                has_after_truncation=has_after_truncation,
+            )
+            return ChunkContext(
+                chunk_text=chunk_text,
+                before_context=before_context,
+                after_context=after_context,
+                chunk_start_offset=chunk_start,
+                chunk_end_offset=chunk_end,
+                page_number=page_number,
+                chunk_index=chunk_index,
+                total_chunks=total_chunks,
+                marked_text=marked_text,
+                has_before_truncation=has_before_truncation,
+                has_after_truncation=has_after_truncation,
+            )
+
+    # Fallback: Fetch full document and extract chunk with context
+    # This path is taken for:
+    # 1. Legacy data with truncated excerpts in Qdrant
+    # 2. Failed Qdrant queries
+    logger.info(
+        f"Falling back to document fetch for {doc_type} {doc_id} "
+        f"(Qdrant cache miss, possibly legacy data)"
+    )
+
+    # For files, retrieve file_path from Qdrant payload
+    resolved_doc_id = doc_id
+    if doc_type == "file" and isinstance(doc_id, int):
+        file_path = await _get_file_path_from_qdrant(
+            user_id, doc_id, chunk_start, chunk_end
+        )
+        if not file_path:
+            logger.warning(
+                f"Could not resolve file_id {doc_id} to file_path from Qdrant"
+            )
+            return None
+        resolved_doc_id = file_path
+        logger.debug(f"Resolved file_id {doc_id} to file_path {file_path}")
+
+    # Fetch full document text
+    full_text = await _fetch_document_text(nc_client, resolved_doc_id, doc_type)
+    if full_text is None:
+        logger.warning(
+            f"Could not fetch document text for {doc_type} {doc_id}, "
+            "skipping context expansion"
+        )
+        return None
+
+    # Validate offsets
+    if chunk_start < 0 or chunk_end > len(full_text) or chunk_start >= chunk_end:
+        logger.warning(
+            f"Invalid chunk offsets for {doc_type} {doc_id}: "
+            f"start={chunk_start}, end={chunk_end}, doc_len={len(full_text)}"
+        )
+        return None
+
+    # Extract chunk text
+    chunk_text = full_text[chunk_start:chunk_end]
+
+    # Calculate context boundaries
+    context_start = max(0, chunk_start - context_chars)
+    context_end = min(len(full_text), chunk_end + context_chars)
+
+    # Extract context
+    before_context = full_text[context_start:chunk_start]
+    after_context = full_text[chunk_end:context_end]
+
+    # Check for truncation
+    has_before_truncation = context_start > 0
+    has_after_truncation = context_end < len(full_text)
+
+    # Create marked text with position markers
+    marked_text = _insert_position_markers(
+        before_context=before_context,
+        chunk_text=chunk_text,
+        after_context=after_context,
+        page_number=page_number,
+        chunk_index=chunk_index,
+        total_chunks=total_chunks,
+        has_before_truncation=has_before_truncation,
+        has_after_truncation=has_after_truncation,
+    )
+
+    return ChunkContext(
+        chunk_text=chunk_text,
+        before_context=before_context,
+        after_context=after_context,
+        chunk_start_offset=chunk_start,
+        chunk_end_offset=chunk_end,
+        page_number=page_number,
+        chunk_index=chunk_index,
+        total_chunks=total_chunks,
+        marked_text=marked_text,
+        has_before_truncation=has_before_truncation,
+        has_after_truncation=has_after_truncation,
+    )
+
+
+async def _fetch_document_text(
+    nc_client: NextcloudClient, doc_id: str | int, doc_type: str
+) -> str | None:
+    """Fetch full text content of a document.
+
+    Args:
+        nc_client: Authenticated Nextcloud client
+        doc_id: Document ID (note ID or file path)
+        doc_type: Type of document ("note", "file", etc.)
+
+    Returns:
+        Full document text, or None if document cannot be retrieved
+    """
+    try:
+        if doc_type == "note":
+            # Fetch note by ID
+            note = await nc_client.notes.get_note(note_id=int(doc_id))
+            # Reconstruct full content as indexed: title + "\n\n" + content
+            # This ensures chunk offsets align with indexed content structure
+            title = note.get("title", "")
+            content = note.get("content", "")
+            return f"{title}\n\n{content}"
+        elif doc_type == "file":
+            # Fetch file content via WebDAV
+            try:
+                file_path = str(doc_id)
+                file_content, content_type = await nc_client.webdav.read_file(file_path)
+
+                # Check if it's a PDF (by content type or file extension)
+                is_pdf = (
+                    content_type and "pdf" in content_type.lower()
+                ) or file_path.lower().endswith(".pdf")
+
+                if is_pdf:
+                    # Extract text from PDF using PyMuPDF
+                    # IMPORTANT: Use pymupdf4llm.to_markdown() to match indexing extraction
+                    # This ensures character offsets align between indexed chunks and retrieval
+                    import pymupdf
+                    import pymupdf4llm
+
+                    logger.debug(f"Extracting text from PDF: {file_path}")
+                    pdf_doc = pymupdf.open(stream=file_content, filetype="pdf")
+                    text_parts = []
+
+                    # Extract each page as markdown (same as indexing)
+                    for page_num in range(pdf_doc.page_count):
+                        page_md = pymupdf4llm.to_markdown(
+                            pdf_doc,
+                            pages=[page_num],
+                            write_images=False,  # Don't need images for context
+                            page_chunks=False,
+                        )
+                        text_parts.append(page_md)
+
+                    pdf_doc.close()
+
+                    # Join pages (no separator - matches indexing)
+                    full_text = "".join(text_parts)
+                    logger.debug(
+                        f"Extracted {len(full_text)} characters from "
+                        f"{pdf_doc.page_count} pages in {file_path}"
+                    )
+                    return full_text
+                else:
+                    # Assume it's a text file, decode to string
+                    logger.debug(f"Decoding text file: {file_path}")
+                    return file_content.decode("utf-8", errors="replace")
+            except Exception as e:
+                logger.error(
+                    f"Error fetching file content for {doc_id}: {e}", exc_info=True
+                )
+                return None
+        else:
+            logger.warning(f"Unsupported doc_type for context expansion: {doc_type}")
+            return None
+    except Exception as e:
+        logger.error(f"Error fetching document {doc_type} {doc_id}: {e}", exc_info=True)
+        return None
+
+
+def _insert_position_markers(
+    before_context: str,
+    chunk_text: str,
+    after_context: str,
+    page_number: int | None,
+    chunk_index: int,
+    total_chunks: int,
+    has_before_truncation: bool,
+    has_after_truncation: bool,
+) -> str:
+    """Insert position markers around matched chunk.
+
+    Creates markdown-formatted text with visual markers indicating chunk
+    boundaries and metadata.
+
+    Args:
+        before_context: Text before chunk
+        chunk_text: The matched chunk
+        after_context: Text after chunk
+        page_number: Optional page number
+        chunk_index: Zero-based chunk index
+        total_chunks: Total chunks in document
+        has_before_truncation: Whether before_context is truncated
+        has_after_truncation: Whether after_context is truncated
+
+    Returns:
+        Formatted text with position markers
+    """
+    # Build position metadata
+    position_parts = []
+    if page_number is not None:
+        position_parts.append(f"Page {page_number}")
+    position_parts.append(f"Chunk {chunk_index + 1} of {total_chunks}")
+    position_metadata = ", ".join(position_parts)
+
+    # Build marked text
+    parts = []
+
+    # Add truncation indicator for before context
+    if has_before_truncation:
+        parts.append("**[...]**\n\n")
+
+    # Add before context if present
+    if before_context:
+        parts.append(before_context)
+
+    # Add chunk start marker
+    parts.append(f"\n\n🔍 **MATCHED CHUNK START** ({position_metadata})\n\n")
+
+    # Add chunk text
+    parts.append(chunk_text)
+
+    # Add chunk end marker
+    parts.append("\n\n🔍 **MATCHED CHUNK END**\n\n")
+
+    # Add after context if present
+    if after_context:
+        parts.append(after_context)
+
+    # Add truncation indicator for after context
+    if has_after_truncation:
+        parts.append("\n\n**[...]**")
+
+    return "".join(parts)

nextcloud_mcp_server/search/fuzzy.py

@@ -1,219 +0,0 @@
-"""Fuzzy search algorithm using character overlap matching on Qdrant payload."""
-
-import logging
-from typing import Any
-
-from qdrant_client.models import FieldCondition, Filter, MatchValue
-
-from nextcloud_mcp_server.config import get_settings
-from nextcloud_mcp_server.search.algorithms import SearchAlgorithm, SearchResult
-from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
-
-logger = logging.getLogger(__name__)
-
-
-class FuzzySearchAlgorithm(SearchAlgorithm):
-    """Fuzzy search using simple character-based similarity.
-
-    Implements character overlap matching with configurable threshold:
-    - Compares character sets between query and text
-    - Requires configurable % character overlap to match (default: 70%)
-    - Tolerant to typos and minor variations
-    """
-
-    def __init__(self, threshold: float = 0.7):
-        """Initialize fuzzy search algorithm.
-
-        Args:
-            threshold: Minimum character overlap ratio (0-1, default: 0.7)
-        """
-        if not 0.0 <= threshold <= 1.0:
-            raise ValueError(f"Threshold must be between 0.0 and 1.0, got {threshold}")
-        self.threshold = threshold
-
-    @property
-    def name(self) -> str:
-        return "fuzzy"
-
-    async def search(
-        self,
-        query: str,
-        user_id: str,
-        limit: int = 10,
-        doc_type: str | None = None,
-        **kwargs: Any,
-    ) -> list[SearchResult]:
-        """Execute fuzzy search using character overlap on Qdrant payload.
-
-        Queries Qdrant for all indexed documents, then scores based on character
-        overlap in title and excerpt fields. Returns unverified results - access
-        verification should be performed separately at the final output stage.
-
-        Args:
-            query: Search query
-            user_id: User ID for filtering
-            limit: Maximum results to return
-            doc_type: Optional document type filter (None = all types)
-            **kwargs: Additional parameters (threshold override)
-
-        Returns:
-            List of unverified SearchResult objects ranked by character overlap score
-        """
-        settings = get_settings()
-        threshold = kwargs.get("threshold", self.threshold)
-
-        logger.info(
-            f"Fuzzy search: query='{query}', user={user_id}, "
-            f"limit={limit}, threshold={threshold}, doc_type={doc_type}"
-        )
-
-        # Build Qdrant filter
-        filter_conditions = [
-            FieldCondition(key="user_id", match=MatchValue(value=user_id))
-        ]
-        if doc_type:
-            filter_conditions.append(
-                FieldCondition(key="doc_type", match=MatchValue(value=doc_type))
-            )
-
-        # Scroll through Qdrant to get all matching documents
-        qdrant_client = await get_qdrant_client()
-        collection = settings.get_collection_name()
-
-        all_points = []
-        offset = None
-
-        # Scroll through all points matching filter
-        while True:
-            scroll_result, next_offset = await qdrant_client.scroll(
-                collection_name=collection,
-                scroll_filter=Filter(must=filter_conditions),
-                limit=100,  # Batch size
-                offset=offset,
-                with_payload=["doc_id", "doc_type", "title", "excerpt", "chunk_index"],
-                with_vectors=False,  # Don't need vectors
-            )
-
-            all_points.extend(scroll_result)
-
-            if next_offset is None:
-                break
-            offset = next_offset
-
-        logger.debug(f"Retrieved {len(all_points)} points from Qdrant for fuzzy search")
-
-        # Deduplicate by (doc_id, doc_type) - keep first chunk
-        seen_docs = {}
-        for point in all_points:
-            doc_id = int(point.payload["doc_id"])
-            dtype = point.payload.get("doc_type", "note")
-            doc_key = (doc_id, dtype)
-
-            chunk_idx = point.payload.get("chunk_index", 0)
-            if doc_key not in seen_docs or chunk_idx == 0:
-                seen_docs[doc_key] = point
-
-        logger.debug(f"Deduplicated to {len(seen_docs)} unique documents")
-
-        # Score each document based on fuzzy matches
-        scored_results = []
-        query_lower = query.lower()
-
-        for doc_key, point in seen_docs.items():
-            doc_id, dtype = doc_key
-            title = point.payload.get("title", "")
-            excerpt = point.payload.get("excerpt", "")
-
-            # Check title match
-            title_score = self._calculate_char_overlap(query_lower, title.lower())
-
-            # Check excerpt match
-            excerpt_score = self._calculate_char_overlap(query_lower, excerpt.lower())
-
-            # Use best score
-            best_score = max(title_score, excerpt_score)
-
-            if best_score >= threshold:
-                match_location = "title" if title_score >= excerpt_score else "excerpt"
-                scored_results.append(
-                    {
-                        "doc_id": doc_id,
-                        "doc_type": dtype,
-                        "title": title,
-                        "excerpt": excerpt
-                        if excerpt_score >= title_score
-                        else f"Title match: {title}",
-                        "score": best_score,
-                        "match_location": match_location,
-                    }
-                )
-
-        # Sort by score (descending) and limit
-        scored_results.sort(key=lambda x: x["score"], reverse=True)
-        top_results = scored_results[:limit]
-
-        # Return unverified results (verification happens at output stage)
-        final_results = []
-        for result in top_results:
-            final_results.append(
-                SearchResult(
-                    id=result["doc_id"],
-                    doc_type=result["doc_type"],
-                    title=result["title"],
-                    excerpt=result["excerpt"],
-                    score=result["score"],
-                    metadata={"match_location": result["match_location"]},
-                )
-            )
-
-        logger.info(f"Fuzzy search returned {len(final_results)} unverified results")
-        if final_results:
-            result_details = [
-                f"{r.doc_type}_{r.id} (score={r.score:.3f}, title='{r.title}')"
-                for r in final_results[:5]
-            ]
-            logger.debug(f"Top fuzzy results: {', '.join(result_details)}")
-
-        return final_results
-
-    def _calculate_char_overlap(self, query: str, text: str) -> float:
-        """Calculate character overlap ratio between query and text.
-
-        Args:
-            query: Query string (normalized)
-            text: Text to compare (normalized)
-
-        Returns:
-            Overlap ratio (0.0-1.0)
-        """
-        if not query or not text:
-            return 0.0
-
-        # Convert to character sets
-        query_chars = set(query)
-        text_chars = set(text)
-
-        # Calculate overlap
-        overlap = query_chars & text_chars
-        overlap_ratio = len(overlap) / len(query_chars)
-
-        return overlap_ratio
-
-    def _extract_excerpt(self, content: str, max_length: int = 200) -> str:
-        """Extract excerpt from content.
-
-        Args:
-            content: Full document content
-            max_length: Maximum excerpt length
-
-        Returns:
-            Excerpt string
-        """
-        if not content:
-            return ""
-
-        excerpt = content[:max_length].strip()
-        if len(content) > max_length:
-            excerpt += "..."
-
-        return excerpt

nextcloud_mcp_server/search/hybrid.py

@@ -1,278 +0,0 @@
-"""Hybrid search algorithm using Reciprocal Rank Fusion (RRF)."""
-
-import logging
-from collections import defaultdict
-from typing import Any
-
-import anyio
-
-from nextcloud_mcp_server.search.algorithms import SearchAlgorithm, SearchResult
-from nextcloud_mcp_server.search.fuzzy import FuzzySearchAlgorithm
-from nextcloud_mcp_server.search.keyword import KeywordSearchAlgorithm
-from nextcloud_mcp_server.search.semantic import SemanticSearchAlgorithm
-
-logger = logging.getLogger(__name__)
-
-
-class HybridSearchAlgorithm(SearchAlgorithm):
-    """Hybrid search combining multiple algorithms using Reciprocal Rank Fusion.
-
-    Implements RRF from ADR-003 to combine results from:
-    - Semantic search (vector similarity)
-    - Keyword search (token matching)
-    - Fuzzy search (character overlap)
-
-    RRF formula: score = weight / (k + rank)
-    where k=60 (standard value) and rank is 1-indexed position.
-    """
-
-    DEFAULT_RRF_K = 60  # Standard RRF constant
-
-    def __init__(
-        self,
-        semantic_weight: float = 0.5,
-        keyword_weight: float = 0.3,
-        fuzzy_weight: float = 0.2,
-        rrf_k: int = DEFAULT_RRF_K,
-    ):
-        """Initialize hybrid search with algorithm weights.
-
-        Args:
-            semantic_weight: Weight for semantic results (default: 0.5)
-            keyword_weight: Weight for keyword results (default: 0.3)
-            fuzzy_weight: Weight for fuzzy results (default: 0.2)
-            rrf_k: RRF constant for rank decay (default: 60)
-
-        Raises:
-            ValueError: If weights are invalid
-        """
-        # Validate weights
-        if semantic_weight < 0 or keyword_weight < 0 or fuzzy_weight < 0:
-            raise ValueError("Weights must be non-negative")
-
-        total_weight = semantic_weight + keyword_weight + fuzzy_weight
-        if total_weight > 1.0:
-            raise ValueError(f"Weights sum to {total_weight:.2f}, must be ≤1.0")
-
-        if total_weight == 0.0:
-            raise ValueError("At least one weight must be > 0")
-
-        self.semantic_weight = semantic_weight
-        self.keyword_weight = keyword_weight
-        self.fuzzy_weight = fuzzy_weight
-        self.rrf_k = rrf_k
-        self.total_weight = total_weight
-
-        # Initialize sub-algorithms
-        self.semantic = SemanticSearchAlgorithm()
-        self.keyword = KeywordSearchAlgorithm()
-        self.fuzzy = FuzzySearchAlgorithm()
-
-    @property
-    def name(self) -> str:
-        return "hybrid"
-
-    @property
-    def requires_vector_db(self) -> bool:
-        # Requires vector DB if semantic search has non-zero weight
-        return self.semantic_weight > 0
-
-    async def search(
-        self,
-        query: str,
-        user_id: str,
-        limit: int = 10,
-        doc_type: str | None = None,
-        **kwargs: Any,
-    ) -> list[SearchResult]:
-        """Execute hybrid search using RRF to combine algorithms.
-
-        Returns unverified results from combined algorithms. Access verification
-        should be performed separately at the final output stage.
-
-        Args:
-            query: Search query
-            user_id: User ID for filtering
-            limit: Maximum results to return
-            doc_type: Optional document type filter
-            **kwargs: Additional parameters passed to sub-algorithms
-
-        Returns:
-            List of unverified SearchResult objects ranked by RRF combined score
-        """
-        logger.info(
-            f"Hybrid search: query='{query}', user={user_id}, limit={limit}, "
-            f"weights=(semantic={self.semantic_weight}, keyword={self.keyword_weight}, "
-            f"fuzzy={self.fuzzy_weight})"
-        )
-
-        # Prepare algorithm configurations for parallel execution
-        algo_configs = []
-        if self.semantic_weight > 0:
-            algo_configs.append(
-                (
-                    "semantic",
-                    self.semantic.search,
-                    query,
-                    user_id,
-                    limit * 2,
-                    doc_type,
-                    kwargs,
-                )
-            )
-        if self.keyword_weight > 0:
-            algo_configs.append(
-                (
-                    "keyword",
-                    self.keyword.search,
-                    query,
-                    user_id,
-                    limit * 2,
-                    doc_type,
-                    kwargs,
-                )
-            )
-        if self.fuzzy_weight > 0:
-            algo_configs.append(
-                (
-                    "fuzzy",
-                    self.fuzzy.search,
-                    query,
-                    user_id,
-                    limit * 2,
-                    doc_type,
-                    kwargs,
-                )
-            )
-
-        # Pre-allocate results list and extract algorithm names
-        results_list = [None] * len(algo_configs)
-        algo_names = [name for name, *_ in algo_configs]
-
-        async def search_one(
-            index: int,
-            search_func,
-            query_arg: str,
-            user_id_arg: str,
-            limit_arg: int,
-            doc_type_arg: str | None,
-            kwargs_arg: dict,
-        ):
-            """Execute one search algorithm and store result at index."""
-            result = await search_func(
-                query_arg, user_id_arg, limit_arg, doc_type_arg, **kwargs_arg
-            )
-            results_list[index] = result
-
-        # Execute searches in parallel using anyio task group
-        async with anyio.create_task_group() as tg:
-            for idx, (name, search_func, q, uid, lim, dt, kw) in enumerate(
-                algo_configs
-            ):
-                tg.start_soon(search_one, idx, search_func, q, uid, lim, dt, kw)
-
-        # Build results dict
-        algo_results = {}
-        for algo_name, results in zip(algo_names, results_list):
-            algo_results[algo_name] = results
-            logger.debug(f"{algo_name} returned {len(results)} results")
-
-        # Combine using RRF
-        combined_results = self._reciprocal_rank_fusion(
-            algo_results,
-            {
-                "semantic": self.semantic_weight,
-                "keyword": self.keyword_weight,
-                "fuzzy": self.fuzzy_weight,
-            },
-            limit,
-        )
-
-        logger.info(f"Hybrid search returned {len(combined_results)} combined results")
-        if combined_results:
-            result_details = [
-                f"{r.doc_type}_{r.id} (score={r.score:.3f}, title='{r.title}')"
-                for r in combined_results[:5]
-            ]
-            logger.debug(f"Top hybrid results: {', '.join(result_details)}")
-
-        return combined_results
-
-    def _reciprocal_rank_fusion(
-        self,
-        algo_results: dict[str, list[SearchResult]],
-        weights: dict[str, float],
-        limit: int,
-    ) -> list[SearchResult]:
-        """Combine multiple ranked result lists using RRF.
-
-        Args:
-            algo_results: Dict of algorithm_name -> ranked results
-            weights: Dict of algorithm_name -> weight (0-1)
-            limit: Maximum results to return
-
-        Returns:
-            Combined and re-ranked results
-        """
-        # Track RRF scores per document
-        rrf_scores: dict[tuple[int, str], float] = defaultdict(float)
-        # Track best result object for each document
-        best_results: dict[tuple[int, str], SearchResult] = {}
-
-        for algo_name, results in algo_results.items():
-            weight = weights.get(algo_name, 0.0)
-            if weight == 0:
-                continue
-
-            for rank, result in enumerate(results, start=1):
-                doc_key = (result.id, result.doc_type)
-
-                # RRF formula: weight / (k + rank)
-                rrf_score = weight / (self.rrf_k + rank)
-                rrf_scores[doc_key] += rrf_score
-
-                # Track best result object (prefer higher original scores)
-                if doc_key not in best_results:
-                    best_results[doc_key] = result
-                elif result.score > best_results[doc_key].score:
-                    best_results[doc_key] = result
-
-        # Sort by combined RRF score
-        sorted_docs = sorted(
-            rrf_scores.items(),
-            key=lambda x: x[1],
-            reverse=True,
-        )[:limit]
-
-        # Calculate normalization factor to scale RRF scores to 0-1 range
-        # Theoretical max RRF score = total_weight / (rrf_k + 1)
-        # Normalization factor = (rrf_k + 1) / total_weight
-        normalization_factor = (self.rrf_k + 1) / self.total_weight
-
-        # Build final results with normalized RRF scores
-        final_results = []
-        for doc_key, rrf_score in sorted_docs:
-            result = best_results[doc_key]
-
-            # Normalize RRF score to 0-1 range for better user comprehension
-            normalized_score = rrf_score * normalization_factor
-
-            # Create new result with normalized score
-            # Keep original metadata but add RRF details
-            metadata = result.metadata or {}
-            metadata["rrf_score_raw"] = rrf_score  # Original RRF score
-            metadata["original_score"] = result.score  # Original algorithm score
-            metadata["normalization_factor"] = normalization_factor
-
-            final_results.append(
-                SearchResult(
-                    id=result.id,
-                    doc_type=result.doc_type,
-                    title=result.title,
-                    excerpt=result.excerpt,
-                    score=normalized_score,  # Use normalized score (0-1 range)
-                    metadata=metadata,
-                )
-            )
-
-        return final_results

nextcloud_mcp_server/search/keyword.py

@@ -1,277 +0,0 @@
-"""Keyword search algorithm using token-based matching on Qdrant payload (ADR-001)."""
-
-import logging
-from typing import Any
-
-from qdrant_client.models import FieldCondition, Filter, MatchValue
-
-from nextcloud_mcp_server.config import get_settings
-from nextcloud_mcp_server.search.algorithms import SearchAlgorithm, SearchResult
-from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
-
-logger = logging.getLogger(__name__)
-
-
-class KeywordSearchAlgorithm(SearchAlgorithm):
-    """Keyword search using token-based matching with weighted scoring.
-
-    Implements token-based search from ADR-001:
-    - Title matches weighted 3x higher than content matches
-    - Case-insensitive token matching
-    - Relevance scoring based on match frequency and location
-    """
-
-    # Weighting constants from ADR-001
-    TITLE_WEIGHT = 3.0
-    CONTENT_WEIGHT = 1.0
-
-    @property
-    def name(self) -> str:
-        return "keyword"
-
-    async def search(
-        self,
-        query: str,
-        user_id: str,
-        limit: int = 10,
-        doc_type: str | None = None,
-        **kwargs: Any,
-    ) -> list[SearchResult]:
-        """Execute keyword search using token matching on Qdrant payload.
-
-        Queries Qdrant for all indexed documents, then scores based on token
-        matches in title and excerpt fields. Returns unverified results - access
-        verification should be performed separately at the final output stage.
-
-        Args:
-            query: Search query to tokenize and match
-            user_id: User ID for filtering
-            limit: Maximum results to return
-            doc_type: Optional document type filter (None = all types)
-            **kwargs: Additional parameters (unused)
-
-        Returns:
-            List of unverified SearchResult objects ranked by keyword match score
-        """
-        settings = get_settings()
-
-        logger.info(
-            f"Keyword search: query='{query}', user={user_id}, "
-            f"limit={limit}, doc_type={doc_type}"
-        )
-
-        # Tokenize query
-        query_tokens = self._process_query(query)
-        logger.debug(f"Query tokens: {query_tokens}")
-
-        # Build Qdrant filter
-        filter_conditions = [
-            FieldCondition(key="user_id", match=MatchValue(value=user_id))
-        ]
-        if doc_type:
-            filter_conditions.append(
-                FieldCondition(key="doc_type", match=MatchValue(value=doc_type))
-            )
-
-        # Scroll through Qdrant to get all matching documents
-        # We need title and excerpt from payload for token matching
-        qdrant_client = await get_qdrant_client()
-        collection = settings.get_collection_name()
-
-        all_points = []
-        offset = None
-
-        # Scroll through all points matching filter
-        while True:
-            scroll_result, next_offset = await qdrant_client.scroll(
-                collection_name=collection,
-                scroll_filter=Filter(must=filter_conditions),
-                limit=100,  # Batch size
-                offset=offset,
-                with_payload=[
-                    "doc_id",
-                    "doc_type",
-                    "title",
-                    "excerpt",
-                    "chunk_index",
-                    "total_chunks",
-                ],
-                with_vectors=False,  # Don't need vectors for keyword search
-            )
-
-            all_points.extend(scroll_result)
-
-            if next_offset is None:
-                break
-            offset = next_offset
-
-        logger.debug(
-            f"Retrieved {len(all_points)} points from Qdrant for keyword search"
-        )
-
-        # Deduplicate by (doc_id, doc_type) - keep best chunk per document
-        seen_docs = {}
-        for point in all_points:
-            doc_id = int(point.payload["doc_id"])
-            dtype = point.payload.get("doc_type", "note")
-            doc_key = (doc_id, dtype)
-
-            # Keep first chunk (chunk_index=0) as it has the most relevant content
-            chunk_idx = point.payload.get("chunk_index", 0)
-            if doc_key not in seen_docs or chunk_idx == 0:
-                seen_docs[doc_key] = point
-
-        logger.debug(f"Deduplicated to {len(seen_docs)} unique documents")
-
-        # Score each document based on keyword matches
-        scored_results = []
-        for doc_key, point in seen_docs.items():
-            doc_id, dtype = doc_key
-            title = point.payload.get("title", "")
-            excerpt = point.payload.get("excerpt", "")
-
-            # Calculate keyword match score
-            score = self._calculate_score(query_tokens, title, excerpt)
-
-            if score > 0:  # Only include matches
-                scored_results.append(
-                    {
-                        "doc_id": doc_id,
-                        "doc_type": dtype,
-                        "title": title,
-                        "excerpt": excerpt,
-                        "score": score,
-                    }
-                )
-
-        # Sort by score (descending) and limit
-        scored_results.sort(key=lambda x: x["score"], reverse=True)
-        top_results = scored_results[:limit]
-
-        # Return unverified results (verification happens at output stage)
-        final_results = []
-        for result in top_results:
-            final_results.append(
-                SearchResult(
-                    id=result["doc_id"],
-                    doc_type=result["doc_type"],
-                    title=result["title"],
-                    excerpt=result["excerpt"],
-                    score=result["score"],
-                    metadata={},
-                )
-            )
-
-        logger.info(f"Keyword search returned {len(final_results)} unverified results")
-        if final_results:
-            result_details = [
-                f"{r.doc_type}_{r.id} (score={r.score:.3f}, title='{r.title}')"
-                for r in final_results[:5]
-            ]
-            logger.debug(f"Top keyword results: {', '.join(result_details)}")
-
-        return final_results
-
-    def _process_query(self, query: str) -> list[str]:
-        """Tokenize and normalize query.
-
-        Args:
-            query: Raw query string
-
-        Returns:
-            List of normalized tokens
-        """
-        # Convert to lowercase and split into tokens
-        tokens = query.lower().split()
-
-        # Filter out very short tokens (optional)
-        tokens = [token for token in tokens if len(token) > 1]
-
-        return tokens
-
-    def _calculate_score(
-        self, query_tokens: list[str], title: str, content: str
-    ) -> float:
-        """Calculate relevance score based on token matches.
-
-        Args:
-            query_tokens: List of query tokens
-            title: Document title
-            content: Document content
-
-        Returns:
-            Relevance score (0.0-1.0)
-        """
-        if not query_tokens:
-            return 0.0
-
-        # Process title and content
-        title_tokens = title.lower().split()
-        content_tokens = content.lower().split()
-
-        score = 0.0
-
-        # Count matches in title
-        title_matches = sum(1 for qt in query_tokens if qt in title_tokens)
-        if query_tokens:  # Avoid division by zero
-            title_match_ratio = title_matches / len(query_tokens)
-            score += self.TITLE_WEIGHT * title_match_ratio
-
-        # Count matches in content
-        content_matches = sum(1 for qt in query_tokens if qt in content_tokens)
-        if query_tokens:
-            content_match_ratio = content_matches / len(query_tokens)
-            score += self.CONTENT_WEIGHT * content_match_ratio
-
-        # Normalize score to 0-1 range
-        # Max score would be TITLE_WEIGHT + CONTENT_WEIGHT if all tokens match everywhere
-        max_score = self.TITLE_WEIGHT + self.CONTENT_WEIGHT
-        normalized_score = min(score / max_score, 1.0)
-
-        return normalized_score
-
-    def _extract_excerpt(
-        self, content: str, query_tokens: list[str], max_length: int = 200
-    ) -> str:
-        """Extract excerpt showing match context.
-
-        Args:
-            content: Full document content
-            query_tokens: Query tokens to find
-            max_length: Maximum excerpt length in characters
-
-        Returns:
-            Excerpt string with context around matches
-        """
-        if not content:
-            return ""
-
-        content_lower = content.lower()
-
-        # Find first occurrence of any query token
-        first_match_pos = -1
-        for token in query_tokens:
-            pos = content_lower.find(token)
-            if pos != -1:
-                if first_match_pos == -1 or pos < first_match_pos:
-                    first_match_pos = pos
-
-        if first_match_pos == -1:
-            # No matches found, return beginning
-            return content[:max_length].strip() + (
-                "..." if len(content) > max_length else ""
-            )
-
-        # Extract context around match
-        start = max(0, first_match_pos - max_length // 2)
-        end = min(len(content), first_match_pos + max_length // 2)
-
-        excerpt = content[start:end].strip()
-
-        # Add ellipsis if truncated
-        if start > 0:
-            excerpt = "..." + excerpt
-        if end < len(content):
-            excerpt = excerpt + "..."
-
-        return excerpt

nextcloud_mcp_server/search/pdf_highlighter.py

@@ -0,0 +1,907 @@
+"""PDF chunk highlighting utilities for vector visualization.
+
+This module provides utilities to generate highlighted page images showing
+matched chunks and their context from semantic search results.
+
+The highlighting uses character offsets to precisely locate chunks within
+PDF documents, ensuring accurate highlighting even when text formatting
+varies between indexing and rendering.
+"""
+
+import logging
+import re
+from typing import Optional
+
+import pymupdf
+import pymupdf4llm
+
+logger = logging.getLogger(__name__)
+
+
+class PDFHighlighter:
+    """Generate highlighted page images from PDF chunks."""
+
+    # Color definitions (RGB, 0-1 range)
+    COLORS = {
+        "yellow": [1, 1, 0],
+        "red": [1, 0, 0],
+        "green": [0, 1, 0],
+        "blue": [0, 0, 1],
+        "orange": [1, 0.5, 0],
+        "pink": [1, 0, 1],
+        "gray": [0.7, 0.7, 0.7],
+        "light_blue": [0.7, 0.9, 1.0],
+        "light_green": [0.7, 1.0, 0.7],
+    }
+
+    @staticmethod
+    def strip_markdown(text: str) -> str:
+        """Remove markdown formatting to improve search accuracy.
+
+        Args:
+            text: Text with potential markdown formatting
+
+        Returns:
+            Plain text with markdown removed
+        """
+        # Remove bold/italic markers
+        text = re.sub(r"\*\*(.+?)\*\*", r"\1", text)
+        text = re.sub(r"\*(.+?)\*", r"\1", text)
+        text = re.sub(r"__(.+?)__", r"\1", text)
+        text = re.sub(r"_(.+?)_", r"\1", text)
+
+        # Remove headers
+        text = re.sub(r"^#+\s+", "", text, flags=re.MULTILINE)
+
+        # Remove inline code
+        text = re.sub(r"`(.+?)`", r"\1", text)
+
+        return text.strip()
+
+    @staticmethod
+    def extract_pdf_text_with_boundaries(
+        pdf_doc: pymupdf.Document,
+    ) -> tuple[str, list[dict]]:
+        """Extract full document text with page boundary tracking.
+
+        Uses pymupdf4llm.to_markdown() for consistency with indexing.
+
+        IMPORTANT: Must use write_images=True to match PyMuPDFProcessor behavior!
+        Even though we don't need the images, we need the image references in the
+        markdown text to maintain consistent character offsets with indexing.
+
+        Args:
+            pdf_doc: Open PyMuPDF document
+
+        Returns:
+            Tuple of (full_text, page_boundaries) where page_boundaries is a list of:
+            {"page": 1, "start_offset": 0, "end_offset": 1234}
+        """
+        import tempfile
+        from pathlib import Path
+
+        page_boundaries = []
+        text_parts = []
+        current_offset = 0
+
+        # Use temp directory for image output (images are discarded after extraction)
+        temp_dir = Path(tempfile.mkdtemp(prefix="pdf_highlight_"))
+
+        for page_idx in range(pdf_doc.page_count):
+            page_md = pymupdf4llm.to_markdown(
+                pdf_doc,
+                pages=[page_idx],
+                write_images=True,  # Must match indexing! Otherwise offsets misalign
+                image_path=temp_dir,
+                page_chunks=False,
+            )
+
+            page_boundaries.append(
+                {
+                    "page": page_idx + 1,  # 1-indexed
+                    "start_offset": current_offset,
+                    "end_offset": current_offset + len(page_md),
+                }
+            )
+
+            text_parts.append(page_md)
+            current_offset += len(page_md)
+
+        full_text = "".join(text_parts)
+
+        # Clean up temp directory and extracted images
+        import shutil
+
+        try:
+            shutil.rmtree(temp_dir)
+        except Exception as e:
+            logger.warning(f"Failed to clean up temp directory {temp_dir}: {e}")
+
+        return full_text, page_boundaries
+
+    @staticmethod
+    def find_chunk_page(
+        chunk_start_offset: int,
+        chunk_end_offset: int,
+        page_boundaries: list[dict],
+    ) -> Optional[dict]:
+        """Find which page contains the most of a given chunk.
+
+        Args:
+            chunk_start_offset: Chunk start position in full document
+            chunk_end_offset: Chunk end position in full document
+            page_boundaries: Page boundary list from extract_pdf_text_with_boundaries()
+
+        Returns:
+            Dict with keys: page_num, overlap_chars, page_relative_start, page_relative_end
+            or None if chunk not found on any page
+        """
+        chunk_pages = []
+
+        for boundary in page_boundaries:
+            page_start = boundary["start_offset"]
+            page_end = boundary["end_offset"]
+
+            # Check if chunk overlaps with this page
+            if chunk_start_offset < page_end and chunk_end_offset > page_start:
+                overlap_start = max(chunk_start_offset, page_start)
+                overlap_end = min(chunk_end_offset, page_end)
+                overlap_chars = overlap_end - overlap_start
+
+                chunk_pages.append(
+                    {
+                        "page_num": boundary["page"],
+                        "overlap_chars": overlap_chars,
+                        "page_relative_start": overlap_start - page_start,
+                        "page_relative_end": overlap_end - page_start,
+                    }
+                )
+
+        if not chunk_pages:
+            return None
+
+        # Return page with maximum overlap
+        return max(chunk_pages, key=lambda p: p["overlap_chars"])
+
+    @staticmethod
+    def highlight_chunk_by_word_positions(
+        page: pymupdf.Page,
+        chunk_text: str,
+        color: str = "yellow",
+        search_region: tuple[float, float, float, float] | None = None,
+    ) -> int:
+        """Highlight chunk using word-position matching.
+
+        This method matches words from the chunk to their positions on the PDF page,
+        avoiding text search mismatches between markdown-formatted text and raw PDF text.
+
+        Args:
+            page: PyMuPDF page object
+            chunk_text: Text to highlight (may contain markdown)
+            color: Color name from COLORS dict
+            search_region: Optional (x0, y0, x1, y1) bounding box to constrain search.
+                          If provided, only words within this region are considered.
+
+        Returns:
+            Number of highlight rectangles added
+        """
+        # Tokenize chunk into words (alphanumeric only, lowercase)
+        chunk_words = re.findall(
+            r"\w+", PDFHighlighter.strip_markdown(chunk_text).lower()
+        )
+
+        if not chunk_words:
+            logger.warning("No words found in chunk text")
+            return 0
+
+        # Get all words from page with positions
+        # Format: (x0, y0, x1, y1, "word", block_no, line_no, word_no)
+        try:
+            page_words = page.get_text("words")
+        except Exception as e:
+            logger.error(f"Failed to extract words from page: {e}")
+            return 0
+
+        if not page_words:
+            logger.warning("No words found on page")
+            return 0
+
+        # Filter words by search region if provided
+        if search_region:
+            rx0, ry0, rx1, ry1 = search_region
+            # Allow some tolerance (10 points) for words near region boundary
+            tolerance = 10
+            page_words = [
+                w
+                for w in page_words
+                if (
+                    w[0] >= rx0 - tolerance
+                    and w[2] <= rx1 + tolerance
+                    and w[1] >= ry0 - tolerance
+                    and w[3] <= ry1 + tolerance
+                )
+            ]
+            logger.debug(
+                f"Filtered to {len(page_words)} words in region "
+                f"({rx0:.0f}, {ry0:.0f}, {rx1:.0f}, {ry1:.0f})"
+            )
+
+        if not page_words:
+            logger.warning("No words found in search region")
+            return 0
+
+        # Find matching word sequence - use FIRST match, not longest
+        # This ensures we highlight the actual chunk location, not similar text elsewhere
+        matches = []
+
+        # Build a simple word-to-positions index for the first few chunk words
+        # to find candidate starting positions
+        first_chunk_word = chunk_words[0] if chunk_words else ""
+        candidate_starts = []
+
+        for i, pw in enumerate(page_words):
+            page_word = pw[4].lower()
+            # Check if this could be the start of the chunk
+            if (
+                first_chunk_word == page_word
+                or first_chunk_word in page_word
+                or page_word in first_chunk_word
+            ):
+                candidate_starts.append(i)
+
+        # Try each candidate start position and take the FIRST good match
+        for start_pos in candidate_starts:
+            current_matches = []
+            chunk_idx = 0
+            skip_count = 0
+            max_skips = 3  # Allow some formatting differences
+
+            for page_idx in range(start_pos, len(page_words)):
+                if chunk_idx >= len(chunk_words):
+                    break
+
+                page_word = page_words[page_idx][4].lower()
+                chunk_word = chunk_words[chunk_idx]
+
+                # Check for match (allow partial matches for flexibility)
+                if (
+                    chunk_word == page_word
+                    or chunk_word in page_word
+                    or page_word in chunk_word
+                ):
+                    current_matches.append(page_words[page_idx])
+                    chunk_idx += 1
+                    skip_count = 0
+                elif skip_count < max_skips:
+                    # Allow skipping some words (formatting, punctuation)
+                    skip_count += 1
+                    continue
+                else:
+                    break
+
+            # Accept if we matched at least 50% of chunk words
+            if len(current_matches) >= len(chunk_words) * 0.5:
+                matches = current_matches
+                logger.debug(
+                    f"Found match at position {start_pos}: "
+                    f"{len(matches)}/{len(chunk_words)} words"
+                )
+                break  # Take FIRST match, not best/longest
+
+        if not matches:
+            logger.debug(f"No word matches found (chunk has {len(chunk_words)} words)")
+            return 0
+
+        logger.debug(
+            f"Matched {len(matches)} words out of {len(chunk_words)} chunk words"
+        )
+
+        # Build rectangles from matched words
+        rects = [pymupdf.Rect(w[0], w[1], w[2], w[3]) for w in matches]
+
+        # Check if matches are contiguous (not scattered across the page)
+        # Scattered matches indicate false positives from common words
+        if len(rects) > 1:
+            # Sort by vertical position then horizontal
+            sorted_matches = sorted(matches, key=lambda w: (round(w[1]), w[0]))
+
+            # Check for large vertical gaps (more than ~2 lines apart)
+            # A typical line height is 12-20 points
+            max_line_gap = 50  # Points - allows for ~2-3 lines gap
+            prev_y = sorted_matches[0][1]
+            large_gaps = 0
+
+            for match in sorted_matches[1:]:
+                y_gap = match[1] - prev_y
+                if y_gap > max_line_gap:
+                    large_gaps += 1
+                prev_y = match[1]
+
+            # If matches are scattered (many large gaps), reject this match
+            # A chunk should be mostly contiguous text
+            if large_gaps > len(matches) * 0.3:  # More than 30% have gaps
+                logger.debug(
+                    f"Rejecting scattered matches: {large_gaps} large gaps "
+                    f"out of {len(matches)} matches"
+                )
+                return 0
+
+        # Merge adjacent rectangles on the same line for cleaner highlighting
+        merged_rects = []
+        sorted_rects = sorted(rects, key=lambda r: (round(r.y0), r.x0))
+
+        current_rect = None
+        for rect in sorted_rects:
+            if current_rect is None:
+                current_rect = rect
+            elif abs(rect.y0 - current_rect.y0) < 5:  # Same line (within 5 points)
+                current_rect = current_rect | rect  # Union
+            else:
+                merged_rects.append(current_rect)
+                current_rect = rect
+
+        if current_rect:
+            merged_rects.append(current_rect)
+
+        # Add highlights
+        rgb = PDFHighlighter.COLORS.get(color, PDFHighlighter.COLORS["yellow"])
+        for rect in merged_rects:
+            highlight = page.add_highlight_annot(rect)
+            highlight.set_colors({"stroke": rgb})
+            highlight.set_info(
+                content="Chunk from semantic search",
+                title="PDF Highlighter (word-position)",
+            )
+            highlight.update()
+
+        return len(merged_rects)
+
+    @staticmethod
+    def find_unique_phrase(
+        text: str, min_len: int = 30, max_len: int = 80
+    ) -> str | None:
+        """Find a relatively unique phrase from text for location search.
+
+        Looks for phrases that are likely to be unique on the page:
+        - Prefers phrases with numbers or special terms
+        - Avoids very common words
+
+        Args:
+            text: Source text to extract phrase from
+            min_len: Minimum phrase length
+            max_len: Maximum phrase length
+
+        Returns:
+            A phrase likely to be unique, or None if not found
+        """
+        clean_text = PDFHighlighter.strip_markdown(text).strip()
+        if not clean_text:
+            return None
+
+        # Try first sentence (often unique due to context)
+        sentences = re.split(r"[.!?]\s+", clean_text)
+        for sentence in sentences:
+            sentence = sentence.strip()
+            if min_len <= len(sentence) <= max_len:
+                return sentence
+            elif len(sentence) > max_len:
+                return sentence[:max_len]
+
+        # Fallback: first N chars
+        if len(clean_text) >= min_len:
+            return clean_text[:max_len]
+
+        return clean_text if clean_text else None
+
+    @staticmethod
+    def _find_chunk_bbox(
+        page: pymupdf.Page,
+        chunk_text: str,
+        page_relative_start: int,
+        page_relative_end: int,
+        page_text_length: int,
+    ) -> tuple[float, float, float, float] | None:
+        """Find bounding box for a chunk without modifying the page.
+
+        Returns (x0, y0, x1, y1) in page coordinates, or None if not found.
+        """
+        page_rect = page.rect
+
+        # Strip markdown for searching
+        search_text = PDFHighlighter.strip_markdown(chunk_text)
+
+        # Try to find chunk location using text search
+        anchor_rect = None
+        search_phrases = []
+
+        # Build search phrases from chunk text
+        sentences = re.split(r"[.!?]\s+", search_text)
+        for sentence in sentences[:3]:
+            sentence = sentence.strip()
+            if len(sentence) >= 20:
+                search_phrases.append(sentence[:80])
+                if len(sentence) >= 40:
+                    search_phrases.append(sentence[:40])
+
+        # Also try first N characters
+        if len(search_text) >= 30:
+            search_phrases.append(search_text[:60])
+            search_phrases.append(search_text[:30])
+
+        for phrase in search_phrases:
+            if not phrase:
+                continue
+            rects = page.search_for(phrase.strip())
+            if rects:
+                anchor_rect = rects[0]
+                break
+
+        if not anchor_rect:
+            return None
+
+        # Calculate chunk height based on character count
+        chunk_chars = len(search_text)
+        estimated_lines = max(1, chunk_chars / 60)
+        estimated_height = estimated_lines * 14
+
+        # Build bounding box
+        return (
+            page_rect.x0 + 30,  # Left margin
+            anchor_rect.y0 - 5,  # Start slightly above anchor
+            page_rect.x1 - 30,  # Right margin
+            min(anchor_rect.y0 + estimated_height + 10, page_rect.y1 - 30),
+        )
+
+    @staticmethod
+    def highlight_chunk_on_page(
+        page: pymupdf.Page,
+        chunk_text: str,
+        color: str = "yellow",
+        page_relative_start: int | None = None,
+        page_relative_end: int | None = None,
+        page_text_length: int | None = None,
+    ) -> int:
+        """Add bounding box highlight to a PDF page for the given chunk text.
+
+        Uses text search to find the chunk's location on the page, then draws
+        a bounding box around that region. Falls back to character offset estimation
+        if text search fails.
+
+        Args:
+            page: PyMuPDF page object
+            chunk_text: Text to highlight (may contain markdown)
+            color: Color name from COLORS dict
+            page_relative_start: Character offset where chunk starts on page (optional)
+            page_relative_end: Character offset where chunk ends on page (optional)
+            page_text_length: Total character length of page text (optional)
+
+        Returns:
+            Number of highlights added (1 for bounding box, 0 if failed)
+        """
+        page_rect = page.rect
+        rgb = PDFHighlighter.COLORS.get(color, PDFHighlighter.COLORS["yellow"])
+
+        # Strip markdown for searching
+        search_text = PDFHighlighter.strip_markdown(chunk_text)
+
+        # Try to find chunk location using text search
+        # Search for progressively shorter phrases until we find a match
+        anchor_rect = None
+        search_phrases = []
+
+        # Build search phrases from chunk text
+        sentences = re.split(r"[.!?]\s+", search_text)
+        for sentence in sentences[:3]:  # Try first 3 sentences
+            sentence = sentence.strip()
+            if len(sentence) >= 20:
+                search_phrases.append(sentence[:80])
+                if len(sentence) >= 40:
+                    search_phrases.append(sentence[:40])
+
+        # Also try first N characters
+        if len(search_text) >= 30:
+            search_phrases.append(search_text[:60])
+            search_phrases.append(search_text[:30])
+
+        for phrase in search_phrases:
+            if not phrase:
+                continue
+            rects = page.search_for(phrase.strip())
+            if rects:
+                anchor_rect = rects[0]  # Use first match
+                logger.debug(f"Found chunk anchor using phrase: '{phrase[:30]}...'")
+                break
+
+        if not anchor_rect:
+            page_num = page.number + 1 if page.number is not None else "unknown"
+            logger.warning(f"Could not find chunk text on page {page_num}")
+            return 0
+
+        # Calculate chunk height based on character count
+        # Estimate ~15 chars per line, ~12pt line height
+        chunk_chars = len(search_text)
+        estimated_lines = max(1, chunk_chars / 60)  # ~60 chars per line typical
+        estimated_height = estimated_lines * 14  # ~14pt per line
+
+        # Build bounding box starting from anchor
+        chunk_rect = pymupdf.Rect(
+            page_rect.x0 + 30,  # Left margin
+            anchor_rect.y0 - 5,  # Start slightly above anchor
+            page_rect.x1 - 30,  # Right margin
+            min(
+                anchor_rect.y0 + estimated_height + 10, page_rect.y1 - 30
+            ),  # Estimated bottom
+        )
+
+        # Draw a visible rectangle around the chunk region
+        shape = page.new_shape()
+        shape.draw_rect(chunk_rect)
+        shape.finish(
+            color=rgb,  # Border color
+            fill=None,  # No fill (transparent)
+            width=2.5,  # Border width
+            dashes="[4 2]",  # Dashed line
+        )
+        shape.commit()
+
+        # Add semi-transparent fill for visibility
+        fill_shape = page.new_shape()
+        fill_shape.draw_rect(chunk_rect)
+        fill_shape.finish(
+            color=None,  # No border
+            fill=[1, 1, 0.7],  # Light yellow fill
+            fill_opacity=0.15,  # Very transparent
+        )
+        fill_shape.commit()
+
+        logger.debug(
+            f"Added bounding box at y={chunk_rect.y0:.0f}-{chunk_rect.y1:.0f} "
+            f"(estimated {estimated_lines:.1f} lines)"
+        )
+
+        return 1
+
+    @staticmethod
+    def highlight_chunk(
+        pdf_bytes: bytes,
+        chunk_start_offset: int,
+        chunk_end_offset: int,
+        stored_page_number: Optional[int] = None,
+        color: str = "yellow",
+        zoom: float = 2.0,
+    ) -> Optional[tuple[bytes, int, int]]:
+        """Generate PNG image of PDF page with highlighted chunk.
+
+        This is the main entry point for highlighting. It:
+        1. Extracts document text with page boundaries
+        2. Finds which page contains the chunk
+        3. Extracts chunk text using character offsets
+        4. Highlights the chunk on the page
+        5. Renders page to PNG
+
+        Args:
+            pdf_bytes: PDF file bytes
+            chunk_start_offset: Chunk start position (document-level)
+            chunk_end_offset: Chunk end position (document-level)
+            stored_page_number: Page number from metadata (optional, for validation)
+            color: Highlight color name
+            zoom: Rendering zoom factor (2.0 = 144 DPI)
+
+        Returns:
+            Tuple of (png_bytes, page_number, highlight_count) or None if failed
+        """
+        import tempfile
+        from pathlib import Path
+
+        temp_pdf_path = None
+        try:
+            # Write PDF to temp file with consistent name "pdf.pdf"
+            # This ensures image references match indexing (e.g., pdf-0001.png)
+            # Different temp filenames would cause different markdown text lengths!
+            temp_dir = Path(tempfile.mkdtemp(prefix="pdf_highlight_"))
+            temp_pdf_path = temp_dir / "pdf.pdf"
+            temp_pdf_path.write_bytes(pdf_bytes)
+
+            # Open PDF from temp file
+            doc = pymupdf.open(temp_pdf_path)
+
+            # Extract text with page boundaries
+            full_text, page_boundaries = (
+                PDFHighlighter.extract_pdf_text_with_boundaries(doc)
+            )
+
+            # Find which page contains the chunk
+            chunk_page_info = PDFHighlighter.find_chunk_page(
+                chunk_start_offset, chunk_end_offset, page_boundaries
+            )
+
+            if not chunk_page_info:
+                logger.error("Chunk not found on any page")
+                doc.close()
+                return None
+
+            page_num = chunk_page_info["page_num"]
+
+            # Log if page differs from stored metadata
+            if stored_page_number and stored_page_number != page_num:
+                logger.info(
+                    f"Chunk primarily on page {page_num}, metadata says {stored_page_number}"
+                )
+
+            # Extract page text
+            page_boundary = page_boundaries[page_num - 1]
+            page_start = page_boundary["start_offset"]
+            page_end = page_boundary["end_offset"]
+            page_text = full_text[page_start:page_end]
+
+            # Extract chunk text using page-relative offsets
+            page_relative_start = chunk_page_info["page_relative_start"]
+            page_relative_end = chunk_page_info["page_relative_end"]
+            chunk_text = page_text[page_relative_start:page_relative_end]
+
+            # Calculate page text length for region estimation
+            page_text_length = page_end - page_start
+
+            logger.debug(
+                f"Extracted {len(chunk_text)} chars on page {page_num} "
+                f"(offsets {page_relative_start}-{page_relative_end} of {page_text_length})"
+            )
+
+            # Get page and add highlights
+            page = doc[page_num - 1]
+            highlight_count = PDFHighlighter.highlight_chunk_on_page(
+                page,
+                chunk_text,
+                color,
+                page_relative_start=page_relative_start,
+                page_relative_end=page_relative_end,
+                page_text_length=page_text_length,
+            )
+
+            if highlight_count == 0:
+                logger.warning("No highlights added")
+                doc.close()
+                return None
+
+            # Render page to PNG
+            mat = pymupdf.Matrix(zoom, zoom)
+            pix = page.get_pixmap(matrix=mat, alpha=False)
+            png_bytes = pix.tobytes("png")
+
+            doc.close()
+
+            logger.info(
+                f"Generated {len(png_bytes):,} byte image with {highlight_count} highlights"
+            )
+
+            return (png_bytes, page_num, highlight_count)
+
+        except Exception as e:
+            logger.error(f"Error highlighting chunk: {e}", exc_info=True)
+            return None
+
+        finally:
+            # Clean up temp directory and PDF file
+            if temp_pdf_path and temp_pdf_path.parent.exists():
+                try:
+                    import shutil
+
+                    shutil.rmtree(temp_pdf_path.parent)
+                except Exception as e:
+                    logger.warning(
+                        f"Failed to delete temp directory {temp_pdf_path.parent}: {e}"
+                    )
+
+    @staticmethod
+    def highlight_chunks_batch(
+        pdf_bytes: bytes,
+        chunks: list[tuple[int, int, int, int | None, str]],
+        page_boundaries: list[dict],
+        full_text: str,
+        color: str = "yellow",
+        zoom: float = 2.0,
+    ) -> dict[int, tuple[bytes, int, int]]:
+        """Generate highlighted images for multiple chunks.
+
+        Opens PDF once for rendering, uses pre-computed page boundaries from the
+        document processor. This ensures consistent character offsets between
+        chunking and highlighting.
+
+        Args:
+            pdf_bytes: PDF file bytes
+            chunks: List of (chunk_index, start_offset, end_offset, stored_page_number, chunk_text)
+                    The chunk_index is used as the key in the returned dict.
+                    chunk_text is the actual text content of the chunk.
+            page_boundaries: Pre-computed page boundaries from document processor.
+                            Each entry: {"page": 1, "start_offset": 0, "end_offset": 1234}
+            full_text: Full document text for extracting page-relative portions.
+            color: Highlight color name
+            zoom: Rendering zoom factor (2.0 = 144 DPI)
+
+        Returns:
+            Dict mapping chunk_index to (png_bytes, page_number, highlight_count)
+            Chunks that fail to highlight are omitted from the result.
+        """
+        import shutil
+        import tempfile
+        from collections import defaultdict
+        from pathlib import Path
+
+        results: dict[int, tuple[bytes, int, int]] = {}
+
+        if not chunks:
+            return results
+
+        temp_pdf_path = None
+        try:
+            # Write PDF to temp file
+            temp_dir = Path(tempfile.mkdtemp(prefix="pdf_highlight_batch_"))
+            temp_pdf_path = temp_dir / "pdf.pdf"
+            temp_pdf_path.write_bytes(pdf_bytes)
+
+            # Open PDF once (only for rendering, not text extraction)
+            doc = pymupdf.open(temp_pdf_path)
+
+            logger.debug(
+                f"Batch highlighting: {len(chunks)} chunks, "
+                f"{len(page_boundaries)} pages"
+            )
+
+            # Group chunks by their target page for efficient rendering
+            # We'll render each page only once with all its highlights
+            chunks_by_page: dict[int, list[tuple[int, dict, str]]] = defaultdict(list)
+
+            for chunk_tuple in chunks:
+                # Unpack chunk tuple - chunk_text is now passed directly
+                chunk_index, start_offset, end_offset, stored_page_num, chunk_text = (
+                    chunk_tuple
+                )
+
+                # Find which page contains this chunk
+                chunk_page_info = PDFHighlighter.find_chunk_page(
+                    start_offset, end_offset, page_boundaries
+                )
+
+                if not chunk_page_info:
+                    logger.warning(f"Chunk {chunk_index}: not found on any page")
+                    continue
+
+                page_num = chunk_page_info["page_num"]
+
+                # Log if page differs from stored metadata
+                if stored_page_num and stored_page_num != page_num:
+                    logger.debug(
+                        f"Chunk {chunk_index}: found on page {page_num}, "
+                        f"metadata says {stored_page_num}"
+                    )
+
+                # Extract page-relative portion of chunk text
+                # This is critical for cross-page chunks where the start
+                # of the chunk might be on a different page
+                page_boundary = page_boundaries[page_num - 1]
+                page_start = page_boundary["start_offset"]
+                page_end = page_boundary["end_offset"]
+                page_text_length = page_end - page_start
+
+                # Calculate what portion of the chunk appears on this page
+                chunk_start_on_page = max(start_offset, page_start)
+                chunk_end_on_page = min(end_offset, page_end)
+
+                # Extract just the text that appears on this page
+                page_relative_text = full_text[chunk_start_on_page:chunk_end_on_page]
+
+                chunks_by_page[page_num].append(
+                    (chunk_index, chunk_page_info, page_relative_text, page_text_length)
+                )
+
+            logger.debug(
+                f"Chunks distributed across {len(chunks_by_page)} unique pages"
+            )
+
+            # OPTIMIZATION: Render each page ONCE, then draw highlights using PIL
+            # This avoids expensive page.get_pixmap() calls per chunk
+            from io import BytesIO
+
+            from PIL import Image, ImageDraw
+
+            # PIL color for bounding box (RGB tuple)
+            rgb = PDFHighlighter.COLORS.get(color, PDFHighlighter.COLORS["yellow"])
+            pil_color = tuple(int(c * 255) for c in rgb)
+            fill_color = (255, 255, 178, 38)  # Light yellow with alpha
+
+            for page_num, page_chunks in chunks_by_page.items():
+                page = doc[page_num - 1]
+
+                # Render page ONCE to get base image (most expensive operation)
+                mat = pymupdf.Matrix(zoom, zoom)
+                base_pix = page.get_pixmap(matrix=mat, alpha=False)
+                base_png = base_pix.tobytes("png")
+
+                # Convert to PIL Image for fast highlight drawing
+                base_image = Image.open(BytesIO(base_png)).convert("RGBA")
+                page_rect = page.rect
+
+                logger.debug(
+                    f"Page {page_num}: rendered once, processing {len(page_chunks)} chunks"
+                )
+
+                for (
+                    chunk_index,
+                    chunk_page_info,
+                    chunk_text,
+                    page_text_length,
+                ) in page_chunks:
+                    try:
+                        # Find chunk bounding box using text search
+                        bbox = PDFHighlighter._find_chunk_bbox(
+                            page,
+                            chunk_text,
+                            chunk_page_info["page_relative_start"],
+                            chunk_page_info["page_relative_end"],
+                            page_text_length,
+                        )
+
+                        if bbox is None:
+                            logger.warning(f"Chunk {chunk_index}: could not find bbox")
+                            continue
+
+                        # Copy base image for this chunk
+                        chunk_image = base_image.copy()
+
+                        # Scale bbox coordinates to pixmap coordinates
+                        scale_x = base_pix.width / page_rect.width
+                        scale_y = base_pix.height / page_rect.height
+                        pil_bbox = (
+                            int(bbox[0] * scale_x),
+                            int(bbox[1] * scale_y),
+                            int(bbox[2] * scale_x),
+                            int(bbox[3] * scale_y),
+                        )
+
+                        # Create transparent overlay for fill (proper alpha blending)
+                        overlay = Image.new("RGBA", chunk_image.size, (0, 0, 0, 0))
+                        overlay_draw = ImageDraw.Draw(overlay)
+                        overlay_draw.rectangle(pil_bbox, fill=fill_color)
+
+                        # Alpha composite the overlay onto the chunk image
+                        chunk_image = Image.alpha_composite(chunk_image, overlay)
+
+                        # Draw border on top (solid, not transparent)
+                        border_draw = ImageDraw.Draw(chunk_image)
+                        border_draw.rectangle(pil_bbox, outline=pil_color, width=3)
+
+                        # Convert back to PNG bytes
+                        output = BytesIO()
+                        chunk_image.convert("RGB").save(output, format="PNG")
+                        png_bytes = output.getvalue()
+
+                        results[chunk_index] = (png_bytes, page_num, 1)
+
+                        logger.debug(
+                            f"Chunk {chunk_index}: {len(png_bytes):,} bytes, "
+                            f"page {page_num}, bbox {pil_bbox}"
+                        )
+
+                    except Exception as e:
+                        logger.error(f"Chunk {chunk_index}: error - {e}")
+                        continue
+
+            doc.close()
+
+            logger.info(
+                f"Batch highlighted {len(results)}/{len(chunks)} chunks successfully"
+            )
+
+            return results
+
+        except Exception as e:
+            logger.error(f"Error in batch highlighting: {e}", exc_info=True)
+            return results
+
+        finally:
+            # Clean up temp directory
+            if temp_pdf_path and temp_pdf_path.parent.exists():
+                try:
+                    shutil.rmtree(temp_pdf_path.parent)
+                except Exception as e:
+                    logger.warning(f"Failed to clean up temp dir: {e}")

nextcloud_mcp_server/search/semantic.py

@@ -9,6 +9,7 @@ from nextcloud_mcp_server.config import get_settings
 from nextcloud_mcp_server.embedding import get_embedding_service
 from nextcloud_mcp_server.observability.metrics import record_qdrant_operation
 from nextcloud_mcp_server.search.algorithms import SearchAlgorithm, SearchResult
+from nextcloud_mcp_server.vector.placeholder import get_placeholder_filter
 from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
 logger = logging.getLogger(__name__)
@@ -50,6 +51,9 @@ class SemanticSearchAlgorithm(SearchAlgorithm):
         Returns unverified results from Qdrant. Access verification should be
         performed separately at the final output stage using verify_search_results().
 
+        Deduplicates by (doc_id, doc_type, chunk_start_offset, chunk_end_offset)
+        to show multiple chunks from the same document while avoiding duplicate chunks.
+
         Args:
             query: Natural language search query
             user_id: User ID for filtering
@@ -74,16 +78,19 @@ class SemanticSearchAlgorithm(SearchAlgorithm):
         # Generate embedding for query
         embedding_service = get_embedding_service()
         query_embedding = await embedding_service.embed(query)
+        # Store for reuse by callers (e.g., viz_routes PCA visualization)
+        self.query_embedding = query_embedding
         logger.debug(
             f"Generated embedding for query (dimension={len(query_embedding)})"
         )
 
         # Build Qdrant filter
         filter_conditions = [
+            get_placeholder_filter(),  # Always exclude placeholders from user-facing queries
             FieldCondition(
                 key="user_id",
                 match=MatchValue(value=user_id),
-            )
+            ),
         ]
 
         # Add doc_type filter if specified
@@ -101,6 +108,7 @@ class SemanticSearchAlgorithm(SearchAlgorithm):
             search_response = await qdrant_client.query_points(
                 collection_name=settings.get_collection_name(),
                 query=query_embedding,
+                using="dense",  # Use named dense vector (BM25 hybrid collections)
                 query_filter=Filter(must=filter_conditions),
                 limit=limit * 2,  # Get extra for deduplication
                 score_threshold=score_threshold,
@@ -122,20 +130,26 @@ class SemanticSearchAlgorithm(SearchAlgorithm):
             top_scores = [p.score for p in search_response.points[:3]]
             logger.debug(f"Top 3 similarity scores: {top_scores}")
 
-        # Deduplicate by (doc_id, doc_type) - multiple chunks per document
-        seen_docs = set()
+        # Deduplicate by (doc_id, doc_type, chunk_start, chunk_end)
+        # This allows multiple chunks from same doc, but removes duplicate chunks
+        seen_chunks = set()
         results = []
 
         for result in search_response.points:
-            doc_id = int(result.payload["doc_id"])
+            if result.payload is None:
+                continue
+            # doc_id can be int (notes) or str (files - file paths)
+            doc_id = result.payload["doc_id"]
             doc_type = result.payload.get("doc_type", "note")
-            doc_key = (doc_id, doc_type)
+            chunk_start = result.payload.get("chunk_start_offset")
+            chunk_end = result.payload.get("chunk_end_offset")
+            chunk_key = (doc_id, doc_type, chunk_start, chunk_end)
 
-            # Skip if we've already seen this document
-            if doc_key in seen_docs:
+            # Skip if we've already seen this exact chunk
+            if chunk_key in seen_chunks:
                 continue
 
-            seen_docs.add(doc_key)
+            seen_chunks.add(chunk_key)
 
             # Return unverified results (verification happens at output stage)
             results.append(
@@ -149,6 +163,12 @@ class SemanticSearchAlgorithm(SearchAlgorithm):
                         "chunk_index": result.payload.get("chunk_index"),
                         "total_chunks": result.payload.get("total_chunks"),
                     },
+                    chunk_start_offset=result.payload.get("chunk_start_offset"),
+                    chunk_end_offset=result.payload.get("chunk_end_offset"),
+                    page_number=result.payload.get("page_number"),
+                    chunk_index=result.payload.get("chunk_index", 0),
+                    total_chunks=result.payload.get("total_chunks", 1),
+                    point_id=str(result.id),  # Qdrant point ID for batch retrieval
                 )
             )
 

nextcloud_mcp_server/server/__init__.py

@@ -2,6 +2,7 @@ from .calendar import configure_calendar_tools
 from .contacts import configure_contacts_tools
 from .cookbook import configure_cookbook_tools
 from .deck import configure_deck_tools
+from .news import configure_news_tools
 from .notes import configure_notes_tools
 from .semantic import configure_semantic_tools
 from .sharing import configure_sharing_tools
@@ -13,6 +14,7 @@ __all__ = [
     "configure_contacts_tools",
     "configure_cookbook_tools",
     "configure_deck_tools",
+    "configure_news_tools",
     "configure_notes_tools",
     "configure_semantic_tools",
     "configure_sharing_tools",

nextcloud_mcp_server/server/news.py

@@ -0,0 +1,360 @@
+"""MCP tools for Nextcloud News app."""
+
+import logging
+
+from httpx import HTTPStatusError, RequestError
+from mcp.server.fastmcp import Context, FastMCP
+from mcp.shared.exceptions import McpError
+from mcp.types import ErrorData
+
+from nextcloud_mcp_server.auth import require_scopes
+from nextcloud_mcp_server.client.news import NewsItemType
+from nextcloud_mcp_server.context import get_client
+from nextcloud_mcp_server.models.news import (
+    FeedHealthResponse,
+    GetItemResponse,
+    GetStatusResponse,
+    ListFeedsResponse,
+    ListFoldersResponse,
+    ListItemsResponse,
+    NewsFeed,
+    NewsFolder,
+    NewsItem,
+    NewsItemSummary,
+)
+from nextcloud_mcp_server.observability.metrics import instrument_tool
+
+logger = logging.getLogger(__name__)
+
+
+def configure_news_tools(mcp: FastMCP):
+    """Configure News app MCP tools."""
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_list_folders(ctx: Context) -> ListFoldersResponse:
+        """List all News folders (requires news:read scope)."""
+        client = await get_client(ctx)
+        try:
+            folders_data = await client.news.get_folders()
+            folders = [NewsFolder(**f) for f in folders_data]
+            return ListFoldersResponse(results=folders, total_count=len(folders))
+        except RequestError as e:
+            raise McpError(
+                ErrorData(code=-1, message=f"Network error listing folders: {str(e)}")
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to list folders: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_list_feeds(ctx: Context) -> ListFeedsResponse:
+        """List all News feeds with metadata (requires news:read scope).
+
+        Returns feeds with unread counts, error status, and overall starred count.
+        """
+        client = await get_client(ctx)
+        try:
+            data = await client.news.get_feeds()
+            feeds = [NewsFeed(**f) for f in data.get("feeds", [])]
+            return ListFeedsResponse(
+                results=feeds,
+                starred_count=data.get("starredCount", 0),
+                newest_item_id=data.get("newestItemId"),
+                total_count=len(feeds),
+            )
+        except RequestError as e:
+            raise McpError(
+                ErrorData(code=-1, message=f"Network error listing feeds: {str(e)}")
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to list feeds: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_list_items(
+        ctx: Context,
+        feed_id: int | None = None,
+        folder_id: int | None = None,
+        starred_only: bool = False,
+        unread_only: bool = False,
+        limit: int = 50,
+        offset: int = 0,
+    ) -> ListItemsResponse:
+        """List News items (articles) with optional filtering (requires news:read scope).
+
+        Args:
+            feed_id: Filter by specific feed ID
+            folder_id: Filter by specific folder ID
+            starred_only: Return only starred items
+            unread_only: Return only unread items
+            limit: Maximum number of items to return (default 50, -1 for all)
+            offset: Item ID to start after (for pagination)
+
+        Returns:
+            ListItemsResponse with items, count, and pagination info
+        """
+        client = await get_client(ctx)
+
+        # Determine item type filter
+        type_ = NewsItemType.ALL
+        id_ = 0
+        if starred_only:
+            type_ = NewsItemType.STARRED
+        elif feed_id is not None:
+            type_ = NewsItemType.FEED
+            id_ = feed_id
+        elif folder_id is not None:
+            type_ = NewsItemType.FOLDER
+            id_ = folder_id
+
+        try:
+            items_data = await client.news.get_items(
+                batch_size=limit,
+                offset=offset,
+                type_=type_,
+                id_=id_,
+                get_read=not unread_only,
+            )
+            items = [NewsItemSummary(**i) for i in items_data]
+
+            # Determine pagination info
+            oldest_id = min((i.id for i in items), default=None) if items else None
+            has_more = len(items) == limit and limit > 0
+
+            return ListItemsResponse(
+                results=items,
+                total_count=len(items),
+                has_more=has_more,
+                oldest_id=oldest_id,
+            )
+        except RequestError as e:
+            raise McpError(
+                ErrorData(code=-1, message=f"Network error listing items: {str(e)}")
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to list items: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_get_item(item_id: int, ctx: Context) -> GetItemResponse:
+        """Get a specific News item by ID with full content (requires news:read scope).
+
+        Args:
+            item_id: Item ID
+
+        Returns:
+            GetItemResponse with full item details including HTML body
+        """
+        client = await get_client(ctx)
+        try:
+            item_data = await client.news.get_item(item_id)
+            item = NewsItem(**item_data)
+            return GetItemResponse(item=item)
+        except ValueError as e:
+            raise McpError(ErrorData(code=-1, message=str(e)))
+        except RequestError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1, message=f"Network error getting item {item_id}: {str(e)}"
+                )
+            )
+        except HTTPStatusError as e:
+            if e.response.status_code == 404:
+                raise McpError(ErrorData(code=-1, message=f"Item {item_id} not found"))
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to get item {item_id}: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_get_starred_items(
+        ctx: Context, limit: int = 50, offset: int = 0
+    ) -> ListItemsResponse:
+        """Get starred (favorited) News items (requires news:read scope).
+
+        Convenience method for retrieving user's starred articles.
+
+        Args:
+            limit: Maximum number of items to return (default 50, -1 for all)
+            offset: Item ID to start after (for pagination)
+
+        Returns:
+            ListItemsResponse with starred items
+        """
+        client = await get_client(ctx)
+        try:
+            items_data = await client.news.get_items(
+                batch_size=limit,
+                offset=offset,
+                type_=NewsItemType.STARRED,
+                get_read=True,  # Include read starred items
+            )
+            items = [NewsItemSummary(**i) for i in items_data]
+
+            oldest_id = min((i.id for i in items), default=None) if items else None
+            has_more = len(items) == limit and limit > 0
+
+            return ListItemsResponse(
+                results=items,
+                total_count=len(items),
+                has_more=has_more,
+                oldest_id=oldest_id,
+            )
+        except RequestError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1, message=f"Network error getting starred items: {str(e)}"
+                )
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to get starred items: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_get_unread_items(
+        ctx: Context, limit: int = 50, offset: int = 0
+    ) -> ListItemsResponse:
+        """Get unread News items (requires news:read scope).
+
+        Convenience method for retrieving unread articles across all feeds.
+
+        Args:
+            limit: Maximum number of items to return (default 50, -1 for all)
+            offset: Item ID to start after (for pagination)
+
+        Returns:
+            ListItemsResponse with unread items
+        """
+        client = await get_client(ctx)
+        try:
+            items_data = await client.news.get_items(
+                batch_size=limit,
+                offset=offset,
+                type_=NewsItemType.ALL,
+                get_read=False,  # Only unread items
+            )
+            items = [NewsItemSummary(**i) for i in items_data]
+
+            oldest_id = min((i.id for i in items), default=None) if items else None
+            has_more = len(items) == limit and limit > 0
+
+            return ListItemsResponse(
+                results=items,
+                total_count=len(items),
+                has_more=has_more,
+                oldest_id=oldest_id,
+            )
+        except RequestError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1, message=f"Network error getting unread items: {str(e)}"
+                )
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to get unread items: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_get_feed_health(feed_id: int, ctx: Context) -> FeedHealthResponse:
+        """Get health status for a specific feed (requires news:read scope).
+
+        Returns error count and last error message if the feed has update issues.
+
+        Args:
+            feed_id: Feed ID to check
+
+        Returns:
+            FeedHealthResponse with error status
+        """
+        client = await get_client(ctx)
+        try:
+            data = await client.news.get_feeds()
+            for feed_data in data.get("feeds", []):
+                if feed_data.get("id") == feed_id:
+                    feed = NewsFeed(**feed_data)
+                    return FeedHealthResponse(
+                        feed_id=feed.id,
+                        title=feed.title,
+                        url=feed.url,
+                        has_errors=feed.has_errors,
+                        error_count=feed.update_error_count,
+                        last_error=feed.last_update_error,
+                    )
+            raise McpError(ErrorData(code=-1, message=f"Feed {feed_id} not found"))
+        except RequestError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Network error getting feed health: {str(e)}",
+                )
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to get feed health: {e.response.status_code}",
+                )
+            )
+
+    @mcp.tool()
+    @require_scopes("news:read")
+    @instrument_tool
+    async def nc_news_get_status(ctx: Context) -> GetStatusResponse:
+        """Get News app status and version (requires news:read scope).
+
+        Returns version information and any configuration warnings.
+        """
+        client = await get_client(ctx)
+        try:
+            status_data = await client.news.get_status()
+            return GetStatusResponse(
+                version=status_data.get("version", "unknown"),
+                warnings=status_data.get("warnings", {}),
+            )
+        except RequestError as e:
+            raise McpError(
+                ErrorData(code=-1, message=f"Network error getting status: {str(e)}")
+            )
+        except HTTPStatusError as e:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message=f"Failed to get status: {e.response.status_code}",
+                )
+            )

nextcloud_mcp_server/server/semantic.py

@@ -1,7 +1,6 @@
 """Semantic search MCP tools using vector database."""
 
 import logging
-from typing import Literal
 
 import anyio
 from httpx import RequestError
@@ -26,12 +25,8 @@ from nextcloud_mcp_server.models.semantic import (
 from nextcloud_mcp_server.observability.metrics import (
     instrument_tool,
 )
-from nextcloud_mcp_server.search import (
-    FuzzySearchAlgorithm,
-    HybridSearchAlgorithm,
-    KeywordSearchAlgorithm,
-    SemanticSearchAlgorithm,
-)
+from nextcloud_mcp_server.search.bm25_hybrid import BM25HybridSearchAlgorithm
+from nextcloud_mcp_server.search.context import get_chunk_with_context
 
 logger = logging.getLogger(__name__)
 
@@ -47,36 +42,38 @@ def configure_semantic_tools(mcp: FastMCP):
         ctx: Context,
         limit: int = 10,
         doc_types: list[str] | None = None,
-        score_threshold: float = 0.7,
-        algorithm: Literal["semantic", "keyword", "fuzzy", "hybrid"] = "hybrid",
-        semantic_weight: float = 0.5,
-        keyword_weight: float = 0.3,
-        fuzzy_weight: float = 0.2,
+        score_threshold: float = 0.0,
+        fusion: str = "rrf",
+        include_context: bool = False,
+        context_chars: int = 300,
     ) -> SemanticSearchResponse:
         """
-        Search Nextcloud content using configurable algorithms with cross-app support.
+        Search Nextcloud content using BM25 hybrid search with cross-app support.
 
-        Supports multiple search algorithms with client-configurable weighting:
-        - semantic: Vector similarity search (requires VECTOR_SYNC_ENABLED=true)
-        - keyword: Token-based matching (title matches weighted 3x)
-        - fuzzy: Character overlap matching (typo-tolerant)
-        - hybrid: Combines all algorithms using Reciprocal Rank Fusion (default)
+        Uses Qdrant's native hybrid search combining:
+        - Dense semantic vectors: For conceptual similarity and natural language queries
+        - BM25 sparse vectors: For precise keyword matching, acronyms, and specific terms
 
-        Document types are queried from the vector database to determine what's
-        actually indexed. Currently only "note" documents are fully supported.
+        Results are automatically fused using the selected fusion algorithm in the
+        database for optimal relevance. This provides the best of both semantic
+        understanding and keyword precision.
+
+        Requires VECTOR_SYNC_ENABLED=true. Currently only "note" documents are
+        fully supported for indexing.
 
         Args:
-            query: Natural language search query
+            query: Natural language or keyword search query
             limit: Maximum number of results to return (default: 10)
             doc_types: Document types to search (e.g., ["note", "file"]). None = search all indexed types (default)
-            score_threshold: Minimum similarity score for semantic/hybrid (0-1, default: 0.7)
-            algorithm: Search algorithm to use (default: "hybrid")
-            semantic_weight: Weight for semantic results in hybrid mode (default: 0.5)
-            keyword_weight: Weight for keyword results in hybrid mode (default: 0.3)
-            fuzzy_weight: Weight for fuzzy results in hybrid mode (default: 0.2)
+            score_threshold: Minimum fusion score (0-1, default: 0.0)
+            fusion: Fusion algorithm: "rrf" (Reciprocal Rank Fusion, default) or "dbsf" (Distribution-Based Score Fusion)
+                   RRF: Good general-purpose fusion using reciprocal ranks
+                   DBSF: Uses distribution-based normalization, may better balance different score ranges
+            include_context: Whether to expand results with surrounding context (default: False)
+            context_chars: Number of characters to include before/after matched chunk (default: 300)
 
         Returns:
-            SemanticSearchResponse with matching documents and relevance scores
+            SemanticSearchResponse with matching documents ranked by fusion scores
         """
         from nextcloud_mcp_server.config import get_settings
 
@@ -85,42 +82,24 @@ def configure_semantic_tools(mcp: FastMCP):
         username = client.username
 
         logger.info(
-            f"Search: query='{query}', user={username}, algorithm={algorithm}, "
-            f"limit={limit}, score_threshold={score_threshold}"
+            f"BM25 hybrid search: query='{query}', user={username}, "
+            f"limit={limit}, score_threshold={score_threshold}, fusion={fusion}"
         )
 
+        # Check that vector sync is enabled
+        if not settings.vector_sync_enabled:
+            raise McpError(
+                ErrorData(
+                    code=-1,
+                    message="BM25 hybrid search requires VECTOR_SYNC_ENABLED=true",
+                )
+            )
+
         try:
-            # Create appropriate algorithm instance
-            if algorithm == "semantic":
-                if not settings.vector_sync_enabled:
-                    raise McpError(
-                        ErrorData(
-                            code=-1,
-                            message="Semantic search requires VECTOR_SYNC_ENABLED=true",
-                        )
-                    )
-                search_algo = SemanticSearchAlgorithm(score_threshold=score_threshold)
-            elif algorithm == "keyword":
-                search_algo = KeywordSearchAlgorithm()
-            elif algorithm == "fuzzy":
-                search_algo = FuzzySearchAlgorithm()
-            elif algorithm == "hybrid":
-                if semantic_weight > 0 and not settings.vector_sync_enabled:
-                    raise McpError(
-                        ErrorData(
-                            code=-1,
-                            message="Hybrid search with semantic component requires VECTOR_SYNC_ENABLED=true",
-                        )
-                    )
-                search_algo = HybridSearchAlgorithm(
-                    semantic_weight=semantic_weight,
-                    keyword_weight=keyword_weight,
-                    fuzzy_weight=fuzzy_weight,
-                )
-            else:
-                raise McpError(
-                    ErrorData(code=-1, message=f"Unknown algorithm: {algorithm}")
-                )
+            # Create BM25 hybrid search algorithm with specified fusion
+            search_algo = BM25HybridSearchAlgorithm(
+                score_threshold=score_threshold, fusion=fusion
+            )
 
             # Execute search across requested document types
             # If doc_types is None, search all indexed types (cross-app search)
@@ -154,18 +133,16 @@ def configure_semantic_tools(mcp: FastMCP):
                 # Sort combined results by score
                 all_results.sort(key=lambda r: r.score, reverse=True)
 
-            # Deduplicate results (hybrid search may return same doc from dense + sparse)
-            # Qdrant already filters by user_id for multi-tenant isolation
-            # Sampling tool will verify access when fetching full content
-            seen = set()
-            unique_results = []
-            for result in all_results:
-                key = (result.id, result.doc_type)
-                if key not in seen:
-                    seen.add(key)
-                    unique_results.append(result)
-
-            search_results = unique_results[:limit]  # Final limit after deduplication
+            # Note: BM25HybridSearchAlgorithm already deduplicates at chunk level
+            # (doc_id, doc_type, chunk_start, chunk_end), which allows multiple
+            # chunks from the same document while preventing duplicate chunks.
+            # No additional deduplication needed here - multiple chunks per document
+            # are valuable for RAG contexts.
+            # Qdrant already filters by user_id for multi-tenant isolation.
+            # Sampling tool will verify access when fetching full content.
+            search_results = all_results[
+                :limit
+            ]  # Final limit after chunk-level dedup in algorithm
 
             # Convert SearchResult objects to SemanticSearchResult for response
             results = []
@@ -184,16 +161,108 @@ def configure_semantic_tools(mcp: FastMCP):
                         total_chunks=r.metadata.get("total_chunks", 1)
                         if r.metadata
                         else 1,
+                        chunk_start_offset=r.chunk_start_offset,
+                        chunk_end_offset=r.chunk_end_offset,
+                        page_number=r.page_number,
                     )
                 )
 
-            logger.info(f"Returning {len(results)} results from {algorithm} search")
+            # Expand results with surrounding context if requested
+            if include_context and results:
+                logger.info(
+                    f"Expanding {len(results)} results with context "
+                    f"(context_chars={context_chars})"
+                )
+
+                # Fetch context for all results in parallel
+                # Limit concurrent requests to prevent connection pool exhaustion
+                max_concurrent = 20
+                semaphore = anyio.Semaphore(max_concurrent)
+                expanded_results = [None] * len(results)
+
+                async def fetch_context(index: int, result: SemanticSearchResult):
+                    """Fetch context for a single result (parallel with semaphore)."""
+                    async with semaphore:
+                        # Only expand if we have valid chunk offsets
+                        if (
+                            result.chunk_start_offset is None
+                            or result.chunk_end_offset is None
+                        ):
+                            # Keep result as-is without context expansion
+                            expanded_results[index] = result
+                            return
+
+                        try:
+                            chunk_context = await get_chunk_with_context(
+                                nc_client=client,
+                                user_id=username,
+                                doc_id=result.id,
+                                doc_type=result.doc_type,
+                                chunk_start=result.chunk_start_offset,
+                                chunk_end=result.chunk_end_offset,
+                                page_number=result.page_number,
+                                chunk_index=result.chunk_index,
+                                total_chunks=result.total_chunks,
+                                context_chars=context_chars,
+                            )
+
+                            if chunk_context:
+                                # Create new result with context fields populated
+                                expanded_results[index] = SemanticSearchResult(
+                                    id=result.id,
+                                    doc_type=result.doc_type,
+                                    title=result.title,
+                                    category=result.category,
+                                    excerpt=result.excerpt,
+                                    score=result.score,
+                                    chunk_index=result.chunk_index,
+                                    total_chunks=result.total_chunks,
+                                    chunk_start_offset=result.chunk_start_offset,
+                                    chunk_end_offset=result.chunk_end_offset,
+                                    page_number=result.page_number,
+                                    # Context expansion fields
+                                    has_context_expansion=True,
+                                    marked_text=chunk_context.marked_text,
+                                    before_context=chunk_context.before_context,
+                                    after_context=chunk_context.after_context,
+                                    has_before_truncation=chunk_context.has_before_truncation,
+                                    has_after_truncation=chunk_context.has_after_truncation,
+                                )
+                                logger.debug(
+                                    f"Expanded context for {result.doc_type} {result.id}"
+                                )
+                            else:
+                                # Context expansion failed, keep original result
+                                expanded_results[index] = result
+                                logger.debug(
+                                    f"Failed to expand context for {result.doc_type} {result.id}, "
+                                    "keeping original result"
+                                )
+                        except Exception as e:
+                            # Context expansion failed, keep original result
+                            expanded_results[index] = result
+                            logger.warning(
+                                f"Error expanding context for {result.doc_type} {result.id}: {e}"
+                            )
+
+                # Run all context fetches in parallel using anyio task group
+                async with anyio.create_task_group() as tg:
+                    for idx, result in enumerate(results):
+                        tg.start_soon(fetch_context, idx, result)
+
+                # Replace results with expanded versions
+                results = [r for r in expanded_results if r is not None]
+                logger.info(
+                    f"Context expansion completed: {len(results)} results with context"
+                )
+
+            logger.info(f"Returning {len(results)} results from BM25 hybrid search")
 
             return SemanticSearchResponse(
                 results=results,
                 query=query,
                 total_found=len(results),
-                search_method=algorithm,
+                search_method=f"bm25_hybrid_{fusion}",
             )
 
         except ValueError as e:
@@ -225,6 +294,9 @@ def configure_semantic_tools(mcp: FastMCP):
         limit: int = 5,
         score_threshold: float = 0.7,
         max_answer_tokens: int = 500,
+        fusion: str = "rrf",
+        include_context: bool = False,
+        context_chars: int = 300,
     ) -> SamplingSearchResponse:
         """
         Semantic search with LLM-generated answer using MCP sampling.
@@ -249,6 +321,9 @@ def configure_semantic_tools(mcp: FastMCP):
             limit: Maximum number of documents to retrieve (default: 5)
             score_threshold: Minimum similarity score 0-1 (default: 0.7)
             max_answer_tokens: Maximum tokens for generated answer (default: 500)
+            fusion: Fusion algorithm: "rrf" (Reciprocal Rank Fusion, default) or "dbsf" (Distribution-Based Score Fusion)
+            include_context: Whether to expand results with surrounding context (default: False)
+            context_chars: Number of characters to include before/after matched chunk (default: 300)
 
         Returns:
             SamplingSearchResponse containing:
@@ -260,27 +335,6 @@ def configure_semantic_tools(mcp: FastMCP):
         Note: Requires MCP client to support sampling. If sampling is unavailable,
         the tool gracefully degrades to returning documents with an explanation.
         The client may prompt the user to approve the sampling request.
-
-        Examples:
-            >>> # Query about objectives across multiple apps
-            >>> result = await nc_semantic_search_answer(
-            ...     query="What are my Q1 2025 project goals?",
-            ...     ctx=ctx
-            ... )
-            >>> print(result.generated_answer)
-            "Based on Document 1 (note: Project Kickoff), Document 2 (calendar event:
-            Q1 Planning Meeting), and Document 3 (deck card: Implement semantic search),
-            your main goals are: 1) Improve semantic search accuracy by 20%,
-            2) Deploy new embedding model, 3) Reduce indexing latency..."
-
-            >>> # Query about appointments
-            >>> result = await nc_semantic_search_answer(
-            ...     query="When is my next dentist appointment?",
-            ...     ctx=ctx,
-            ...     limit=10
-            ... )
-            >>> len(result.sources)  # Calendar events and related notes
-            3
         """
         # 1. Retrieve relevant documents via existing semantic search
         search_response = await nc_semantic_search(
@@ -288,6 +342,9 @@ def configure_semantic_tools(mcp: FastMCP):
             ctx=ctx,
             limit=limit,
             score_threshold=score_threshold,
+            fusion=fusion,
+            include_context=include_context,
+            context_chars=context_chars,
         )
 
         # 2. Handle no results case - don't waste a sampling call
@@ -442,9 +499,11 @@ def configure_semantic_tools(mcp: FastMCP):
         )
 
         # 6. Request LLM completion via MCP sampling with timeout
+        # Note: 5 minute timeout to accommodate slower local LLMs (e.g., Ollama)
+        sampling_timeout_seconds = 300
 
         try:
-            with anyio.fail_after(30):
+            with anyio.fail_after(sampling_timeout_seconds):
                 sampling_result = await ctx.session.create_message(
                     messages=[
                         SamplingMessage(
@@ -491,14 +550,14 @@ def configure_semantic_tools(mcp: FastMCP):
 
         except TimeoutError:
             logger.warning(
-                f"Sampling request timed out after 30 seconds for query: '{query}', "
+                f"Sampling request timed out after {sampling_timeout_seconds} seconds for query: '{query}', "
                 f"returning search results only"
             )
             return SamplingSearchResponse(
                 query=query,
                 generated_answer=(
                     f"[Sampling request timed out]\n\n"
-                    f"The answer generation took too long (>30s). "
+                    f"The answer generation took too long (>{sampling_timeout_seconds}s). "
                     f"Found {len(accessible_results)} relevant documents. "
                     f"Please review the sources below or try a simpler query."
                 ),
@@ -618,15 +677,22 @@ def configure_semantic_tools(mcp: FastMCP):
             # Get Qdrant client and query indexed count
             indexed_count = 0
             try:
+                from qdrant_client.models import Filter
+
                 from nextcloud_mcp_server.config import get_settings
+                from nextcloud_mcp_server.vector.placeholder import (
+                    get_placeholder_filter,
+                )
                 from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
                 settings = get_settings()
                 qdrant_client = await get_qdrant_client()
 
-                # Count documents in collection
+                # Count documents in collection, excluding placeholders
+                # Placeholders are zero-vector points used to track processing state
                 count_result = await qdrant_client.count(
-                    collection_name=settings.get_collection_name()
+                    collection_name=settings.get_collection_name(),
+                    count_filter=Filter(must=[get_placeholder_filter()]),
                 )
                 indexed_count = count_result.count
 

nextcloud_mcp_server/server/webdav.py

@@ -64,20 +64,6 @@ def configure_webdav_tools(mcp: FastMCP):
             - Text files are decoded to UTF-8
             - Documents (PDF, DOCX, etc.) are parsed and text is extracted
             - Other binary files are base64 encoded
-
-        Examples:
-            # Read a text file
-            result = await nc_webdav_read_file("Documents/readme.txt")
-            logger.info(result['content'])  # Decoded text content
-
-            # Read a PDF document (automatically parsed)
-            result = await nc_webdav_read_file("Documents/report.pdf")
-            logger.info(result['content'])  # Extracted text from PDF
-            logger.info(result['parsing_metadata'])  # Document parsing info
-
-            # Read a binary file
-            result = await nc_webdav_read_file("Images/photo.jpg")
-            logger.info(result['encoding'])  # 'base64'
         """
         client = await get_client(ctx)
         content, content_type = await client.webdav.read_file(path)

nextcloud_mcp_server/smithery_main.py

@@ -0,0 +1,60 @@
+"""Smithery-specific entrypoint for stateless deployment.
+
+ADR-016: This entrypoint is used when deploying on Smithery's hosting platform.
+It configures the server for stateless operation with per-session authentication.
+
+Features disabled in Smithery mode:
+- Vector sync / semantic search (no persistent storage)
+- Admin UI at /app (no webhooks, no vector viz)
+- OAuth provisioning tools (no token storage)
+
+Features enabled:
+- Core Nextcloud tools (notes, calendar, contacts, files, deck, tables, cookbook)
+- Per-session app password authentication via Smithery configSchema
+- Health check endpoints (/health/live, /health/ready)
+"""
+
+import logging
+import os
+
+import uvicorn
+
+from nextcloud_mcp_server.config import setup_logging
+
+logger = logging.getLogger(__name__)
+
+
+def main():
+    """Start the MCP server in Smithery stateless mode."""
+    # Setup logging first
+    setup_logging()
+
+    # Force stateless mode environment variables
+    os.environ["SMITHERY_DEPLOYMENT"] = "true"
+    os.environ["VECTOR_SYNC_ENABLED"] = "false"
+
+    logger.info("Starting Nextcloud MCP Server in Smithery stateless mode")
+
+    # Import app after setting environment variables
+    from nextcloud_mcp_server.app import get_app
+
+    # Create the app with streamable-http transport (required for Smithery)
+    app = get_app(transport="streamable-http")
+
+    # Smithery sets PORT environment variable
+    port = int(os.environ.get("PORT", 8081))
+
+    logger.info(f"Listening on port {port}")
+
+    uvicorn.run(
+        app,
+        host="0.0.0.0",
+        port=port,
+        log_level="info",
+        # Disable access log for cleaner output
+        access_log=False,
+    )
+
+
+if __name__ == "__main__":
+    main()

nextcloud_mcp_server/vector/document_chunker.py

@@ -1,51 +1,97 @@
-"""Document chunking for large texts."""
+"""Document chunking for large texts using LangChain text splitters."""
 
 import logging
+from dataclasses import dataclass
+
+from langchain_text_splitters import RecursiveCharacterTextSplitter
 
 logger = logging.getLogger(__name__)
 
 
-class DocumentChunker:
-    """Chunk large documents for optimal embedding."""
+@dataclass
+class ChunkWithPosition:
+    """A text chunk with its character position in the original document."""
 
-    def __init__(self, chunk_size: int = 512, overlap: int = 50):
+    text: str
+    start_offset: int  # Character position where chunk starts
+    end_offset: int  # Character position where chunk ends (exclusive)
+    page_number: int | None = None  # Page number for PDF chunks (optional)
+    metadata: dict | None = None  # Additional processor-specific metadata (optional)
+
+
+class DocumentChunker:
+    """Chunk large documents for optimal embedding using LangChain text splitters.
+
+    Uses RecursiveCharacterTextSplitter which preserves semantic boundaries
+    by splitting on sentence and paragraph boundaries before resorting to
+    character-level splitting.
+    """
+
+    def __init__(self, chunk_size: int = 2048, overlap: int = 200):
         """
         Initialize document chunker.
 
         Args:
-            chunk_size: Number of words per chunk (default: 512)
-            overlap: Number of overlapping words between chunks (default: 50)
+            chunk_size: Number of characters per chunk (default: 2048)
+            overlap: Number of overlapping characters between chunks (default: 200)
         """
         self.chunk_size = chunk_size
         self.overlap = overlap
 
-    def chunk_text(self, content: str) -> list[str]:
-        """
-        Split text into overlapping chunks.
+        # Initialize LangChain RecursiveCharacterTextSplitter
+        # Uses hierarchical splitting to preserve semantic boundaries:
+        # - Paragraphs (\n\n)
+        # - Sentences (. ! ?)
+        # - Words (spaces)
+        # - Characters (last resort)
+        # This prevents mid-sentence splitting while maintaining semantic coherence
+        self.splitter = RecursiveCharacterTextSplitter(
+            chunk_size=chunk_size,
+            chunk_overlap=overlap,
+            add_start_index=True,  # Enable position tracking
+            strip_whitespace=True,
+        )
 
-        Uses simple word-based chunking with configurable overlap to preserve
-        context across chunk boundaries.
+    async def chunk_text(self, content: str) -> list[ChunkWithPosition]:
+        """
+        Split text into overlapping chunks with position tracking.
+
+        Uses LangChain's RecursiveCharacterTextSplitter to create chunks that
+        preserve semantic boundaries by splitting at paragraphs and sentences
+        before resorting to word or character-level splitting. This ensures
+        sentences are kept intact. Preserves character positions for each chunk
+        to enable precise document retrieval.
 
         Args:
             content: Text content to chunk
 
         Returns:
-            List of text chunks (may be single item if content is small)
+            List of chunks with their character positions in the original content
         """
-        # Simple word-based chunking
-        words = content.split()
+        import anyio
 
-        if len(words) <= self.chunk_size:
-            return [content]
+        # Handle empty content - return single empty chunk for backward compatibility
+        if not content:
+            return [ChunkWithPosition(text="", start_offset=0, end_offset=0)]
 
-        chunks = []
-        start = 0
+        # Run CPU-bound text splitting in thread pool to avoid blocking event loop
+        docs = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+            self.splitter.create_documents,
+            [content],
+        )
 
-        while start < len(words):
-            end = start + self.chunk_size
-            chunk_words = words[start:end]
-            chunks.append(" ".join(chunk_words))
-            start = end - self.overlap
+        # Convert LangChain Documents to ChunkWithPosition objects
+        chunks = [
+            ChunkWithPosition(
+                text=doc.page_content,
+                start_offset=doc.metadata.get("start_index", 0),
+                end_offset=doc.metadata.get("start_index", 0) + len(doc.page_content),
+            )
+            for doc in docs
+        ]
 
-        logger.debug(f"Chunked document into {len(chunks)} chunks ({len(words)} words)")
+        logger.debug(
+            f"Chunked document into {len(chunks)} chunks "
+            f"(chunk_size={self.chunk_size}, overlap={self.overlap})"
+        )
         return chunks

nextcloud_mcp_server/vector/html_processor.py

@@ -0,0 +1,49 @@
+"""HTML to Markdown conversion utilities for vector sync."""
+
+import logging
+
+from markdownify import markdownify as md
+
+logger = logging.getLogger(__name__)
+
+
+def html_to_markdown(html_content: str | None) -> str:
+    """Convert HTML content to Markdown, preserving semantic structure.
+
+    This function converts HTML (typically from RSS/Atom feed items) to Markdown
+    for better text embedding. Markdown preserves:
+    - Heading hierarchy (important for document structure)
+    - Lists (bullet and numbered)
+    - Links (as [text](url))
+    - Bold/italic emphasis
+    - Paragraphs and line breaks
+
+    Args:
+        html_content: HTML string to convert (may be None or empty)
+
+    Returns:
+        Markdown string, or empty string if input is None/empty
+
+    Example:
+        >>> html_to_markdown("<h1>Title</h1><p>Content with <b>bold</b>.</p>")
+        '# Title\\n\\nContent with **bold**.\\n\\n'
+    """
+    if not html_content:
+        return ""
+
+    try:
+        markdown = md(
+            html_content,
+            heading_style="ATX",  # Use # style headings
+            strip=["script", "style", "iframe", "noscript"],  # Remove unsafe elements
+            bullets="-",  # Use - for unordered lists
+            code_language="",  # Don't add language hints to code blocks
+        )
+        return markdown.strip()
+    except Exception as e:
+        logger.warning(f"Failed to convert HTML to Markdown: {e}")
+        # Fallback: strip all HTML tags as a last resort
+        import re
+
+        text = re.sub(r"<[^>]+>", " ", html_content)
+        return " ".join(text.split())  # Normalize whitespace

nextcloud_mcp_server/vector/placeholder.py

@@ -0,0 +1,306 @@
+"""Placeholder point management for Qdrant state tracking.
+
+Placeholders are zero-vector points stored in Qdrant to track document processing
+state. They prevent duplicate work by marking documents as "in-flight" during the
+gap between scanner queuing and processor completion.
+
+Architecture:
+- Scanner writes placeholders when queuing documents for processing
+- Processor deletes placeholders and writes real vectors after processing
+- All user-facing queries filter out placeholders (is_placeholder: False)
+
+Placeholders contain:
+- Zero vectors (dimension from embedding service)
+- is_placeholder: True flag (for filtering)
+- status: "pending", "processing", "completed", "failed"
+- modified_at, etag from source document
+- queued_at timestamp
+"""
+
+import logging
+import time
+import uuid
+
+from qdrant_client.models import FieldCondition, Filter, MatchValue, PointStruct
+
+from nextcloud_mcp_server.config import get_settings
+from nextcloud_mcp_server.embedding import get_embedding_service
+from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
+
+logger = logging.getLogger(__name__)
+
+
+def _generate_placeholder_id(doc_type: str, doc_id: str | int) -> str:
+    """Generate deterministic UUID for placeholder point.
+
+    Args:
+        doc_type: Document type (note, file, etc.)
+        doc_id: Document ID
+
+    Returns:
+        UUID string for point ID
+    """
+    point_name = f"{doc_type}:{doc_id}:placeholder"
+    return str(uuid.uuid5(uuid.NAMESPACE_DNS, point_name))
+
+
+async def write_placeholder_point(
+    doc_id: str | int,
+    doc_type: str,
+    user_id: str,
+    modified_at: int,
+    etag: str = "",
+    file_path: str | None = None,
+) -> None:
+    """Write a placeholder point to Qdrant to mark document as queued.
+
+    This should be called by the scanner BEFORE queuing a document for processing.
+    The placeholder prevents duplicate work if the scanner runs again before
+    processing completes.
+
+    Args:
+        doc_id: Document ID (int for notes/files)
+        doc_type: Document type (note, file, etc.)
+        user_id: User ID who owns the document
+        modified_at: Document modification timestamp
+        etag: Document ETag (if available)
+        file_path: File path (for files only)
+
+    Raises:
+        Exception: If Qdrant write fails
+    """
+    try:
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+        embedding_service = get_embedding_service()
+
+        # Get dimension dynamically (never hardcode)
+        dimension = embedding_service.get_dimension()
+
+        # Create zero vectors
+        zero_dense = [0.0] * dimension
+
+        # Create empty sparse vector for placeholders
+        # Use models.SparseVector with empty indices/values
+        from qdrant_client import models
+
+        empty_sparse = models.SparseVector(indices=[], values=[])
+
+        # Generate deterministic point ID
+        point_id = _generate_placeholder_id(doc_type, doc_id)
+
+        # Build payload
+        payload = {
+            "user_id": user_id,
+            "doc_id": doc_id,
+            "doc_type": doc_type,
+            "is_placeholder": True,
+            "status": "pending",
+            "modified_at": modified_at,
+            "etag": etag,
+            "queued_at": int(time.time()),
+        }
+
+        # Add file_path for files
+        if doc_type == "file" and file_path:
+            payload["file_path"] = file_path
+
+        # Create placeholder point
+        point = PointStruct(
+            id=point_id,
+            vector={
+                "dense": zero_dense,
+                "sparse": empty_sparse,  # Empty sparse vector for placeholders
+            },
+            payload=payload,
+        )
+
+        # Upsert to Qdrant
+        await qdrant_client.upsert(
+            collection_name=settings.get_collection_name(),
+            points=[point],
+            wait=True,
+        )
+
+        logger.debug(
+            f"Wrote placeholder for {doc_type}_{doc_id} (user={user_id}, "
+            f"modified_at={modified_at})"
+        )
+
+    except Exception as e:
+        logger.error(
+            f"Failed to write placeholder for {doc_type}_{doc_id}: {e}",
+            exc_info=True,
+        )
+        raise
+
+
+async def query_document_metadata(
+    doc_id: str | int,
+    doc_type: str,
+    user_id: str,
+) -> dict | None:
+    """Query Qdrant for existing document entry (placeholder or real).
+
+    Returns the payload of the first matching point, which could be:
+    - A placeholder (is_placeholder: True)
+    - A real indexed document (is_placeholder: False or missing)
+    - None if document not in Qdrant
+
+    Args:
+        doc_id: Document ID
+        doc_type: Document type
+        user_id: User ID
+
+    Returns:
+        Payload dict if found, None otherwise
+    """
+    try:
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Query for any entry matching doc_id, doc_type, user_id
+        scroll_result = await qdrant_client.scroll(
+            collection_name=settings.get_collection_name(),
+            scroll_filter=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=doc_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value=doc_type)),
+                ]
+            ),
+            limit=1,
+            with_payload=True,
+            with_vectors=False,
+        )
+
+        if scroll_result[0]:
+            point = scroll_result[0][0]
+            return dict(point.payload)
+
+        return None
+
+    except Exception as e:
+        logger.warning(f"Error querying document metadata for {doc_type}_{doc_id}: {e}")
+        return None
+
+
+async def delete_placeholder_point(
+    doc_id: str | int,
+    doc_type: str,
+    user_id: str,
+) -> None:
+    """Delete a placeholder point from Qdrant.
+
+    This should be called by the processor BEFORE writing real vectors.
+    We delete the placeholder to avoid duplicates, then write the real chunks.
+
+    Args:
+        doc_id: Document ID
+        doc_type: Document type
+        user_id: User ID
+
+    Raises:
+        Exception: If Qdrant delete fails
+    """
+    try:
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Delete by filter (in case there are multiple chunks from old indexing)
+        await qdrant_client.delete(
+            collection_name=settings.get_collection_name(),
+            points_selector=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=doc_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value=doc_type)),
+                    FieldCondition(key="is_placeholder", match=MatchValue(value=True)),
+                ]
+            ),
+        )
+
+        logger.debug(f"Deleted placeholder for {doc_type}_{doc_id} (user={user_id})")
+
+    except Exception as e:
+        logger.error(
+            f"Failed to delete placeholder for {doc_type}_{doc_id}: {e}",
+            exc_info=True,
+        )
+        raise
+
+
+async def update_placeholder_status(
+    doc_id: str | int,
+    doc_type: str,
+    user_id: str,
+    status: str,
+) -> None:
+    """Update the status field of a placeholder point.
+
+    Status values:
+    - "pending": Queued for processing
+    - "processing": Currently being processed
+    - "completed": Processing completed successfully
+    - "failed": Processing failed
+
+    Args:
+        doc_id: Document ID
+        doc_type: Document type
+        user_id: User ID
+        status: New status value
+
+    Raises:
+        Exception: If Qdrant update fails
+    """
+    try:
+        qdrant_client = await get_qdrant_client()
+        settings = get_settings()
+
+        # Update payload using set_payload
+        await qdrant_client.set_payload(
+            collection_name=settings.get_collection_name(),
+            payload={"status": status},
+            points=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_id", match=MatchValue(value=doc_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value=doc_type)),
+                    FieldCondition(key="is_placeholder", match=MatchValue(value=True)),
+                ]
+            ),
+        )
+
+        logger.debug(
+            f"Updated placeholder status for {doc_type}_{doc_id} to '{status}' "
+            f"(user={user_id})"
+        )
+
+    except Exception as e:
+        logger.warning(
+            f"Failed to update placeholder status for {doc_type}_{doc_id}: {e}"
+        )
+        # Don't raise - status updates are non-critical
+
+
+def get_placeholder_filter() -> FieldCondition:
+    """Get a filter condition to exclude placeholders from queries.
+
+    Add this to all user-facing search/visualization queries to ensure
+    placeholders are never returned to users.
+
+    Returns:
+        FieldCondition that filters out is_placeholder: True
+
+    Example:
+        Filter(
+            must=[
+                get_placeholder_filter(),  # Exclude placeholders
+                FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+            ]
+        )
+    """
+    return FieldCondition(
+        key="is_placeholder",
+        match=MatchValue(value=False),
+    )

nextcloud_mcp_server/vector/processor.py

@@ -15,7 +15,7 @@ from qdrant_client.models import FieldCondition, Filter, MatchValue, PointStruct
 
 from nextcloud_mcp_server.client import NextcloudClient
 from nextcloud_mcp_server.config import get_settings
-from nextcloud_mcp_server.embedding import get_embedding_service
+from nextcloud_mcp_server.embedding import get_bm25_service, get_embedding_service
 from nextcloud_mcp_server.observability.metrics import (
     record_qdrant_operation,
     record_vector_sync_processing,
@@ -23,12 +23,50 @@ from nextcloud_mcp_server.observability.metrics import (
 )
 from nextcloud_mcp_server.observability.tracing import trace_operation
 from nextcloud_mcp_server.vector.document_chunker import DocumentChunker
+from nextcloud_mcp_server.vector.placeholder import delete_placeholder_point
 from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 from nextcloud_mcp_server.vector.scanner import DocumentTask
 
 logger = logging.getLogger(__name__)
 
 
+def assign_page_numbers(chunks, page_boundaries):
+    """Assign page numbers to chunks based on page boundaries.
+
+    Each chunk gets the page number where most of its content appears.
+    For chunks spanning multiple pages, assigns the page containing the
+    majority of the chunk's characters.
+
+    Args:
+        chunks: List of ChunkWithPosition objects
+        page_boundaries: List of dicts with {page, start_offset, end_offset}
+
+    Returns:
+        None (modifies chunks in place)
+    """
+    if not page_boundaries:
+        return
+
+    for chunk in chunks:
+        # Find which page(s) this chunk overlaps with
+        max_overlap = 0
+        assigned_page = None
+
+        for boundary in page_boundaries:
+            # Calculate overlap between chunk and page
+            overlap_start = max(chunk.start_offset, boundary["start_offset"])
+            overlap_end = min(chunk.end_offset, boundary["end_offset"])
+            overlap = max(0, overlap_end - overlap_start)
+
+            # Assign to page with maximum overlap
+            if overlap > max_overlap:
+                max_overlap = overlap
+                assigned_page = boundary["page"]
+
+        if assigned_page is not None:
+            chunk.page_number = assigned_page
+
+
 async def processor_task(
     worker_id: int,
     receive_stream: MemoryObjectReceiveStream[DocumentTask],
@@ -218,30 +256,313 @@ async def _index_document(
     settings = get_settings()
 
     # Fetch document content
-    if doc_task.doc_type == "note":
-        document = await nc_client.notes.get_note(int(doc_task.doc_id))
-        content = f"{document['title']}\n\n{document['content']}"
-        title = document["title"]
-        etag = document.get("etag", "")
-    else:
-        raise ValueError(f"Unsupported doc_type: {doc_task.doc_type}")
+    with trace_operation(
+        "vector_sync.fetch_content",
+        attributes={
+            "vector_sync.doc_type": doc_task.doc_type,
+            "vector_sync.doc_id": doc_task.doc_id,
+        },
+    ):
+        if doc_task.doc_type == "note":
+            document = await nc_client.notes.get_note(int(doc_task.doc_id))
+            content = f"{document['title']}\n\n{document['content']}"
+            title = document["title"]
+            etag = document.get("etag", "")
+            file_metadata = {}  # No file-specific metadata for notes
+            file_path = None  # Notes don't have file paths
+            content_bytes = None  # Notes don't have binary content
+            content_type = None
+        elif doc_task.doc_type == "news_item":
+            from nextcloud_mcp_server.vector.html_processor import html_to_markdown
+
+            item = await nc_client.news.get_item(int(doc_task.doc_id))
+            # Convert HTML body to Markdown for better embedding
+            body_markdown = html_to_markdown(item.get("body", ""))
+            # Build content: title + URL + body
+            item_title = item.get("title", "")
+            item_url = item.get("url", "")
+            feed_title = item.get("feedTitle", "")
+
+            # Structure content for embedding
+            content_parts = [item_title]
+            if feed_title:
+                content_parts.append(f"Source: {feed_title}")
+            if item_url:
+                content_parts.append(f"URL: {item_url}")
+            content_parts.append("")  # Blank line
+            content_parts.append(body_markdown)
+            content = "\n".join(content_parts)
+
+            title = item_title
+            etag = item.get("guidHash", "")
+            # Store news-specific metadata for later use in payload
+            file_metadata = {
+                "feed_id": item.get("feedId"),
+                "feed_title": feed_title,
+                "author": item.get("author"),
+                "pub_date": item.get("pubDate"),
+                "starred": item.get("starred", False),
+                "unread": item.get("unread", True),
+                "url": item_url,
+                "guid_hash": item.get("guidHash"),
+                "enclosure_link": item.get("enclosureLink"),
+                "enclosure_mime": item.get("enclosureMime"),
+            }
+            file_path = None
+            content_bytes = None
+            content_type = None
+        elif doc_task.doc_type == "file":
+            # For files, doc_id is now the numeric file ID, file_path comes from DocumentTask
+            if not doc_task.file_path:
+                raise ValueError(
+                    f"File path required for file indexing but not provided (file_id={doc_task.doc_id})"
+                )
+            file_path = doc_task.file_path
+
+            # Read file content via WebDAV
+            content_bytes, content_type = await nc_client.webdav.read_file(file_path)
+        else:
+            raise ValueError(f"Unsupported doc_type: {doc_task.doc_type}")
+
+    # Process file content (text extraction)
+    if doc_task.doc_type == "file":
+        # Type narrowing: content_bytes and content_type are set for files
+        assert content_bytes is not None
+        assert content_type is not None
+        assert file_path is not None
+
+        with trace_operation(
+            "vector_sync.document_process",
+            attributes={
+                "vector_sync.content_type": content_type,
+                "vector_sync.file_size": len(content_bytes),
+            },
+        ):
+            # Use document processor registry to extract text
+            from nextcloud_mcp_server.document_processors import get_registry
+
+            registry = get_registry()
+
+            try:
+                result = await registry.process(
+                    content=content_bytes,
+                    content_type=content_type,
+                    filename=file_path,
+                )
+                content = result.text
+                file_metadata = result.metadata
+                title = file_metadata.get("title") or file_path.split("/")[-1]
+                etag = ""  # WebDAV read_file doesn't return etag
+
+                # Diagnostic: Log page boundary information if available
+                if "page_boundaries" in file_metadata:
+                    page_boundaries = file_metadata["page_boundaries"]
+                    logger.info(
+                        f"Page boundaries for {file_path}: "
+                        f"{len(page_boundaries)} pages, text length: {len(content)}"
+                    )
+                    # Log first 3 page boundaries for debugging
+                    for boundary in page_boundaries[:3]:
+                        logger.debug(
+                            f"  Page {boundary['page']}: "
+                            f"offsets [{boundary['start_offset']}:{boundary['end_offset']}]"
+                        )
+                    # Verify last boundary matches text length
+                    if page_boundaries:
+                        last_boundary = page_boundaries[-1]
+                        if last_boundary["end_offset"] != len(content):
+                            logger.warning(
+                                f"Text length mismatch: content={len(content)}, "
+                                f"last_boundary_end={last_boundary['end_offset']}"
+                            )
+                else:
+                    logger.debug(f"No page_boundaries in metadata for {file_path}")
+            except Exception as e:
+                logger.error(f"Failed to process file {file_path}: {e}")
+                raise
 
     # Tokenize and chunk (using configured chunk size and overlap)
-    chunker = DocumentChunker(
-        chunk_size=settings.document_chunk_size,
-        overlap=settings.document_chunk_overlap,
-    )
-    chunks = chunker.chunk_text(content)
+    with trace_operation(
+        "vector_sync.chunk_text",
+        attributes={
+            "vector_sync.input_chars": len(content),
+            "vector_sync.chunk_size": settings.document_chunk_size,
+            "vector_sync.overlap": settings.document_chunk_overlap,
+        },
+    ):
+        chunker = DocumentChunker(
+            chunk_size=settings.document_chunk_size,
+            overlap=settings.document_chunk_overlap,
+        )
+        chunks = await chunker.chunk_text(content)
 
-    # Generate embeddings (I/O bound - external API call)
-    embedding_service = get_embedding_service()
-    embeddings = await embedding_service.embed_batch(chunks)
+    # Assign page numbers to chunks if page boundaries are available (PDFs)
+    page_boundaries = file_metadata.get("page_boundaries")
+    if doc_task.doc_type == "file" and page_boundaries is not None:
+        with trace_operation(
+            "vector_sync.assign_page_numbers",
+            attributes={
+                "vector_sync.chunk_count": len(chunks),
+                "vector_sync.page_count": len(page_boundaries),
+            },
+        ):
+            assign_page_numbers(chunks, page_boundaries)
+
+            # Diagnostic: Verify page number assignment
+            assigned_count = sum(1 for c in chunks if c.page_number is not None)
+            logger.info(
+                f"Assigned page numbers to {assigned_count}/{len(chunks)} chunks "
+                f"for {file_path}"
+            )
+
+            # Log first 3 chunks to see their page assignments
+            for i, chunk in enumerate(chunks[:3]):
+                logger.debug(
+                    f"  Chunk {i}: page={chunk.page_number}, "
+                    f"offsets=[{chunk.start_offset}:{chunk.end_offset}]"
+                )
+
+            # Warning if NO page numbers were assigned
+            if assigned_count == 0:
+                logger.warning(
+                    f"NO page numbers assigned! "
+                    f"Text length: {len(content)}, "
+                    f"Chunks: {len(chunks)}, "
+                    f"Chunk offset range: [{chunks[0].start_offset}:{chunks[-1].end_offset}], "
+                    f"Page boundaries: {len(page_boundaries)} pages, "
+                    f"First boundary: {page_boundaries[0] if page_boundaries else 'None'}"
+                )
+
+    # Extract chunk texts for embedding
+    chunk_texts = [chunk.text for chunk in chunks]
+
+    # Initialize results containers
+    dense_embeddings: list = []
+    sparse_embeddings: list = []
+    chunk_images: dict[int, dict] = {}
+
+    # Determine if we need PDF highlighting
+    is_pdf = doc_task.doc_type == "file" and content_type == "application/pdf"
+
+    # Define async tasks for parallel execution
+    async def generate_dense_embeddings():
+        """Generate dense embeddings (I/O bound - external API call)."""
+        nonlocal dense_embeddings
+        with trace_operation(
+            "vector_sync.embed_dense",
+            attributes={
+                "vector_sync.chunk_count": len(chunk_texts),
+                "vector_sync.total_chars": sum(len(t) for t in chunk_texts),
+            },
+        ):
+            embedding_service = get_embedding_service()
+            dense_embeddings = await embedding_service.embed_batch(chunk_texts)
+
+    async def generate_sparse_embeddings():
+        """Generate sparse embeddings (BM25 for keyword matching)."""
+        nonlocal sparse_embeddings
+        with trace_operation(
+            "vector_sync.embed_sparse",
+            attributes={
+                "vector_sync.chunk_count": len(chunk_texts),
+            },
+        ):
+            bm25_service = get_bm25_service()
+            sparse_embeddings = await bm25_service.encode_batch(chunk_texts)
+
+    async def generate_highlights():
+        """Generate highlighted page images for PDF chunks (CPU-bound)."""
+        nonlocal chunk_images
+        if not is_pdf:
+            return
+
+        # Type narrowing: content_bytes is set for PDF files
+        assert content_bytes is not None
+
+        with trace_operation(
+            "vector_sync.generate_highlights",
+            attributes={
+                "vector_sync.chunk_count": len(chunks),
+                "vector_sync.pdf_size": len(content_bytes),
+            },
+        ):
+            import base64
+
+            from nextcloud_mcp_server.search.pdf_highlighter import PDFHighlighter
+
+            # Build chunk data for batch processing
+            # Format: (chunk_index, start_offset, end_offset, page_number, chunk_text)
+            chunk_data: list[tuple[int, int, int, int | None, str]] = [
+                (i, chunk.start_offset, chunk.end_offset, chunk.page_number, chunk.text)
+                for i, chunk in enumerate(chunks)
+                if chunk.page_number is not None
+            ]
+
+            # Get pre-computed page boundaries from document processor
+            page_boundaries = file_metadata.get("page_boundaries")
+            if not page_boundaries:
+                logger.warning("No page boundaries available, skipping highlighting")
+                return
+
+            logger.info(
+                f"Batch generating highlighted page images for {len(chunk_data)} PDF chunks"
+            )
+
+            # Run CPU-bound highlighting in thread pool
+            # Pass pre-computed page boundaries and full text to avoid re-processing the PDF
+            batch_results = await anyio.to_thread.run_sync(  # type: ignore[attr-defined]
+                lambda: PDFHighlighter.highlight_chunks_batch(
+                    pdf_bytes=content_bytes,
+                    chunks=chunk_data,
+                    page_boundaries=page_boundaries,
+                    full_text=content,
+                    color="yellow",
+                    zoom=2.0,
+                )
+            )
+
+            # Convert results to storage format
+            for chunk_index, (
+                png_bytes,
+                actual_page_num,
+                highlight_count,
+            ) in batch_results.items():
+                image_base64 = base64.b64encode(png_bytes).decode("utf-8")
+                chunk_images[chunk_index] = {
+                    "image": image_base64,
+                    "page": actual_page_num,
+                    "highlights": highlight_count,
+                    "size": len(png_bytes),
+                }
+
+            logger.info(
+                f"Generated {len(chunk_images)}/{len(chunks)} highlighted page images "
+                f"(avg {sum(img['size'] for img in chunk_images.values()) // max(len(chunk_images), 1):,} bytes)"
+            )
+
+    # Run all embedding/highlighting operations in parallel
+    # - Dense embeddings: I/O bound (API call)
+    # - Sparse embeddings: CPU bound (local BM25)
+    # - Highlighting: CPU bound (PyMuPDF rendering, runs in thread pool)
+    with trace_operation(
+        "vector_sync.parallel_processing",
+        attributes={
+            "vector_sync.is_pdf": is_pdf,
+            "vector_sync.chunk_count": len(chunks),
+        },
+    ):
+        async with anyio.create_task_group() as tg:
+            tg.start_soon(generate_dense_embeddings)
+            tg.start_soon(generate_sparse_embeddings)
+            tg.start_soon(generate_highlights)
 
     # Prepare Qdrant points
     indexed_at = int(time.time())
     points = []
 
-    for i, (chunk, embedding) in enumerate(zip(chunks, embeddings)):
+    for i, (chunk, dense_emb, sparse_emb) in enumerate(
+        zip(chunks, dense_embeddings, sparse_embeddings)
+    ):
         # Generate deterministic UUID for point ID
         # Using uuid5 with DNS namespace and combining doc info
         point_name = f"{doc_task.doc_type}:{doc_task.doc_id}:chunk:{i}"
@@ -250,28 +571,110 @@ async def _index_document(
         points.append(
             PointStruct(
                 id=point_id,
-                vector=embedding,
+                vector={
+                    "dense": dense_emb,
+                    "sparse": sparse_emb,
+                },
                 payload={
                     "user_id": doc_task.user_id,
                     "doc_id": doc_task.doc_id,
                     "doc_type": doc_task.doc_type,
+                    "is_placeholder": False,  # Real indexed document (not placeholder)
                     "title": title,
-                    "excerpt": chunk[:200],
+                    "excerpt": chunk.text,  # Full chunk text (up to chunk_size, default 2048 chars)
                     "indexed_at": indexed_at,
                     "modified_at": doc_task.modified_at,
                     "etag": etag,
                     "chunk_index": i,
                     "total_chunks": len(chunks),
+                    "chunk_start_offset": chunk.start_offset,
+                    "chunk_end_offset": chunk.end_offset,
+                    "metadata_version": 2,  # v2 includes position metadata
+                    # File-specific metadata (PDF, etc.)
+                    **(
+                        {
+                            "file_path": file_path,  # Store file path for retrieval
+                            "mime_type": content_type,  # From WebDAV response
+                            "file_size": file_metadata.get("file_size"),
+                            "page_number": chunk.page_number,
+                            "page_count": file_metadata.get("page_count"),
+                            "author": file_metadata.get("author"),
+                            "creation_date": file_metadata.get("creation_date"),
+                            "has_images": file_metadata.get("has_images", False),
+                            "image_count": file_metadata.get("image_count", 0),
+                        }
+                        if doc_task.doc_type == "file"
+                        else {}
+                    ),
+                    # News item-specific metadata
+                    **(
+                        {
+                            "feed_id": file_metadata.get("feed_id"),
+                            "feed_title": file_metadata.get("feed_title"),
+                            "author": file_metadata.get("author"),
+                            "pub_date": file_metadata.get("pub_date"),
+                            "starred": file_metadata.get("starred"),
+                            "unread": file_metadata.get("unread"),
+                            "url": file_metadata.get("url"),
+                            "guid_hash": file_metadata.get("guid_hash"),
+                            "enclosure_link": file_metadata.get("enclosure_link"),
+                            "enclosure_mime": file_metadata.get("enclosure_mime"),
+                        }
+                        if doc_task.doc_type == "news_item"
+                        else {}
+                    ),
+                    # Highlighted page image (PDF only)
+                    **(
+                        {
+                            "highlighted_page_image": chunk_images[i]["image"],
+                            "highlighted_page_number": chunk_images[i]["page"],
+                            "highlight_count": chunk_images[i]["highlights"],
+                        }
+                        if i in chunk_images
+                        else {}
+                    ),
                 },
             )
         )
 
-    # Upsert to Qdrant
-    await qdrant_client.upsert(
-        collection_name=settings.get_collection_name(),
-        points=points,
-        wait=True,
-    )
+    # Delete placeholder before writing real vectors
+    # This prevents duplicates and cleans up the placeholder state
+    try:
+        await delete_placeholder_point(
+            doc_id=doc_task.doc_id,
+            doc_type=doc_task.doc_type,
+            user_id=doc_task.user_id,
+        )
+    except Exception as e:
+        # Log but don't fail indexing if placeholder deletion fails
+        logger.warning(
+            f"Failed to delete placeholder for {doc_task.doc_type}_{doc_task.doc_id}: {e}"
+        )
+
+    # Upsert to Qdrant in batches to avoid timeout with large payloads
+    # Each batch is limited to avoid WriteTimeout when sending large image payloads
+    BATCH_SIZE = 10  # ~2MB per batch with images
+    with trace_operation(
+        "vector_sync.qdrant_upsert",
+        attributes={
+            "vector_sync.point_count": len(points),
+            "vector_sync.collection": settings.get_collection_name(),
+            "vector_sync.images_count": len(chunk_images),
+            "vector_sync.batch_size": BATCH_SIZE,
+        },
+    ):
+        for batch_start in range(0, len(points), BATCH_SIZE):
+            batch_end = min(batch_start + BATCH_SIZE, len(points))
+            batch = points[batch_start:batch_end]
+            await qdrant_client.upsert(
+                collection_name=settings.get_collection_name(),
+                points=batch,
+                wait=True,
+            )
+            if batch_end < len(points):
+                logger.debug(
+                    f"Upserted batch {batch_start // BATCH_SIZE + 1}/{(len(points) + BATCH_SIZE - 1) // BATCH_SIZE}"
+                )
 
     logger.info(
         f"Indexed {doc_task.doc_type}_{doc_task.doc_id} for {doc_task.user_id} "

nextcloud_mcp_server/vector/qdrant_client.py

@@ -2,7 +2,7 @@
 
 import logging
 
-from qdrant_client import AsyncQdrantClient
+from qdrant_client import AsyncQdrantClient, models
 from qdrant_client.models import Distance, VectorParams
 
 from nextcloud_mcp_server.config import get_settings
@@ -84,45 +84,62 @@ async def get_qdrant_client() -> AsyncQdrantClient:
                 f"Collection '{collection_name}' found, validating dimensions..."
             )
             collection_info = await _qdrant_client.get_collection(collection_name)
-            actual_dimension = collection_info.config.params.vectors.size
+            # Handle both named vectors (dict) and legacy single vector
+            vectors = collection_info.config.params.vectors
+            if isinstance(vectors, dict):
+                actual_dimension = vectors["dense"].size
+            else:
+                actual_dimension = vectors.size
 
             # Validate dimension matches
             if actual_dimension != expected_dimension:
+                embedding_model = settings.get_embedding_model_name()
                 raise ValueError(
                     f"Dimension mismatch for collection '{collection_name}':\n"
-                    f"  Expected: {expected_dimension} (from embedding model '{settings.ollama_embedding_model}')\n"
+                    f"  Expected: {expected_dimension} (from embedding model '{embedding_model}')\n"
                     f"  Found: {actual_dimension}\n"
                     f"This usually means you changed the embedding model.\n"
                     f"Solutions:\n"
                     f"  1. Delete the old collection: Collection will be recreated with new dimensions\n"
                     f"  2. Set QDRANT_COLLECTION to use a different collection name\n"
-                    f"  3. Revert OLLAMA_EMBEDDING_MODEL to the original model"
+                    f"  3. Revert to the original embedding model"
                 )
 
             logger.info(
                 f"Using existing Qdrant collection: {collection_name} "
-                f"(dimension={actual_dimension}, model={settings.ollama_embedding_model})"
+                f"(dimension={actual_dimension}, model={settings.get_embedding_model_name()})"
             )
 
         else:
             # Collection doesn't exist - create it
+            embedding_model = settings.get_embedding_model_name()
             logger.info(
                 f"Collection '{collection_name}' not found, creating with "
-                f"dimension={expected_dimension}, model={settings.ollama_embedding_model}..."
+                f"dimension={expected_dimension}, model={embedding_model}..."
             )
             await _qdrant_client.create_collection(
                 collection_name=collection_name,
-                vectors_config=VectorParams(
-                    size=expected_dimension,
-                    distance=Distance.COSINE,
-                ),
+                vectors_config={
+                    "dense": VectorParams(
+                        size=expected_dimension,
+                        distance=Distance.COSINE,
+                    ),
+                },
+                sparse_vectors_config={
+                    "sparse": models.SparseVectorParams(
+                        index=models.SparseIndexParams(
+                            on_disk=False,
+                        )
+                    ),
+                },
             )
             logger.info(
                 f"Created Qdrant collection: {collection_name}\n"
-                f"  Dimension: {expected_dimension}\n"
-                f"  Model: {settings.ollama_embedding_model}\n"
+                f"  Dense vector dimension: {expected_dimension}\n"
+                f"  Dense embedding model: {embedding_model}\n"
+                f"  Sparse vectors: BM25 (for hybrid search)\n"
                 f"  Distance: COSINE\n"
-                f"Background sync will index all documents with this embedding model."
+                f"Background sync will index all documents with dense + sparse vectors."
             )
 
     return _qdrant_client

nextcloud_mcp_server/vector/scanner.py

@@ -4,6 +4,7 @@ Periodically scans enabled users' content and queues changed documents for proce
 """
 
 import logging
+import os
 import time
 from dataclasses import dataclass
 
@@ -16,6 +17,10 @@ from nextcloud_mcp_server.client import NextcloudClient
 from nextcloud_mcp_server.config import get_settings
 from nextcloud_mcp_server.observability.metrics import record_vector_sync_scan
 from nextcloud_mcp_server.observability.tracing import trace_operation
+from nextcloud_mcp_server.vector.placeholder import (
+    query_document_metadata,
+    write_placeholder_point,
+)
 from nextcloud_mcp_server.vector.qdrant_client import get_qdrant_client
 
 logger = logging.getLogger(__name__)
@@ -26,10 +31,11 @@ class DocumentTask:
     """Document task for processing queue."""
 
     user_id: str
-    doc_id: str
+    doc_id: int | str  # int for files/notes, str for legacy
     doc_type: str  # "note", "file", "calendar"
     operation: str  # "index" or "delete"
     modified_at: int
+    file_path: str | None = None  # File path for files (when doc_id is file_id)
 
 
 # Track documents potentially deleted (grace period before actual deletion)
@@ -182,8 +188,9 @@ async def scan_user_documents(
                 f"[SCAN-{scan_id}] Using pruneBefore={prune_before} to optimize data transfer"
             )
 
-        # Get indexed state from Qdrant first (for incremental sync)
-        indexed_docs = {}
+        # For deletion tracking, get all doc_ids in Qdrant (for incremental sync)
+        # Note: We no longer bulk-query indexed_at, instead check per-document
+        indexed_doc_ids = set()
         if not initial_sync:
             qdrant_client = await get_qdrant_client()
             scroll_result = await qdrant_client.scroll(
@@ -194,17 +201,18 @@ async def scan_user_documents(
                         FieldCondition(key="doc_type", match=MatchValue(value="note")),
                     ]
                 ),
-                with_payload=["doc_id", "indexed_at"],
+                with_payload=["doc_id"],
                 with_vectors=False,
                 limit=10000,
             )
 
-            indexed_docs = {
-                point.payload["doc_id"]: point.payload["indexed_at"]
-                for point in scroll_result[0]
+            indexed_doc_ids = {
+                point.payload["doc_id"]
+                for point in (scroll_result[0] or [])
+                if point.payload is not None
             }
 
-            logger.debug(f"Found {len(indexed_docs)} indexed documents in Qdrant")
+            logger.debug(f"Found {len(indexed_doc_ids)} indexed documents in Qdrant")
 
         # Stream notes from Nextcloud and process immediately
         note_count = 0
@@ -218,7 +226,14 @@ async def scan_user_documents(
             modified_at = note.get("modified", 0)
 
             if initial_sync:
-                # Send everything on first sync
+                # Send everything on first sync - write placeholder first
+                await write_placeholder_point(
+                    doc_id=doc_id,
+                    doc_type="note",
+                    user_id=user_id,
+                    modified_at=modified_at,
+                    etag=note.get("etag", ""),
+                )
                 await send_stream.send(
                     DocumentTask(
                         user_id=user_id,
@@ -230,9 +245,7 @@ async def scan_user_documents(
                 )
                 queued += 1
             else:
-                # Incremental sync: compare with indexed state
-                indexed_at = indexed_docs.get(doc_id)
-
+                # Incremental sync: check if document exists and compare modified_at
                 # If document reappeared, remove from potentially_deleted
                 doc_key = (user_id, doc_id)
                 if doc_key in _potentially_deleted:
@@ -241,8 +254,48 @@ async def scan_user_documents(
                     )
                     del _potentially_deleted[doc_key]
 
+                # Query Qdrant for existing entry (placeholder or real)
+                existing_metadata = await query_document_metadata(
+                    doc_id=doc_id, doc_type="note", user_id=user_id
+                )
+
                 # Send if never indexed or modified since last index
-                if indexed_at is None or modified_at > indexed_at:
+                # Compare against stored modified_at (not indexed_at!)
+                needs_indexing = False
+                if existing_metadata is None:
+                    # Never seen before
+                    needs_indexing = True
+                elif existing_metadata.get("modified_at", 0) < modified_at:
+                    # Document modified since last indexing
+                    needs_indexing = True
+                elif existing_metadata.get("is_placeholder", False):
+                    # Placeholder exists - check if it's stale (processing may have failed)
+                    # Only requeue if placeholder is older than 5x scan interval
+                    # (Large PDFs can take 3-4 minutes to process)
+                    queued_at = existing_metadata.get("queued_at", 0)
+                    placeholder_age = time.time() - queued_at
+                    stale_threshold = get_settings().vector_sync_scan_interval * 5
+                    if placeholder_age > stale_threshold:
+                        logger.debug(
+                            f"Found stale placeholder for note {doc_id} "
+                            f"(age={placeholder_age:.1f}s), requeuing"
+                        )
+                        needs_indexing = True
+                    else:
+                        logger.debug(
+                            f"Skipping note {doc_id} with recent placeholder "
+                            f"(age={placeholder_age:.1f}s < {stale_threshold:.1f}s)"
+                        )
+
+                if needs_indexing:
+                    # Write placeholder before queuing
+                    await write_placeholder_point(
+                        doc_id=doc_id,
+                        doc_type="note",
+                        user_id=user_id,
+                        modified_at=modified_at,
+                        etag=note.get("etag", ""),
+                    )
                     await send_stream.send(
                         DocumentTask(
                             user_id=user_id,
@@ -270,7 +323,7 @@ async def scan_user_documents(
         )  # Allow 1.5 scan intervals
         current_time = time.time()
 
-        for doc_id in indexed_docs:
+        for doc_id in indexed_doc_ids:
             if doc_id not in nextcloud_doc_ids:
                 doc_key = (user_id, doc_id)
 
@@ -309,7 +362,394 @@ async def scan_user_documents(
                     )
                     _potentially_deleted[doc_key] = current_time
 
+        # Scan tagged PDF files (after notes)
+        # Get indexed file IDs from Qdrant (for deletion tracking)
+        indexed_file_ids = set()
+        if not initial_sync:
+            file_scroll_result = await qdrant_client.scroll(
+                collection_name=settings.get_collection_name(),
+                scroll_filter=Filter(
+                    must=[
+                        FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                        FieldCondition(key="doc_type", match=MatchValue(value="file")),
+                    ]
+                ),
+                limit=10000,  # Reasonable limit for file count
+                with_payload=["doc_id"],
+                with_vectors=False,
+            )
+
+            indexed_file_ids = {
+                point.payload["doc_id"]
+                for point in (file_scroll_result[0] or [])
+                if point.payload is not None
+            }
+
+            logger.debug(f"Found {len(indexed_file_ids)} indexed files in Qdrant")
+
+        # Scan for tagged PDF files
+        file_count = 0
+        file_queued = 0
+        nextcloud_file_ids = set()
+
+        try:
+            # Find files with vector-index tag using OCS Tags API
+            settings = get_settings()
+            tag_name = os.getenv("VECTOR_SYNC_PDF_TAG", "vector-index")
+            # Use NextcloudClient.find_files_by_tag() which uses proper OCS API
+            # and filters by PDF MIME type
+            tagged_files = await nc_client.find_files_by_tag(
+                tag_name, mime_type_filter="application/pdf"
+            )
+
+            for file_info in tagged_files:
+                # Files are already filtered by MIME type in find_files_by_tag()
+                file_count += 1
+                file_id = file_info["id"]  # Use numeric file ID, not path
+                file_path = file_info["path"]  # Keep path for logging
+                nextcloud_file_ids.add(file_id)
+
+                # Use last_modified timestamp if available, otherwise use current time
+                modified_at = file_info.get("last_modified_timestamp", int(time.time()))
+                if isinstance(file_info.get("last_modified"), str):
+                    # Parse RFC 2822 date format if needed
+                    from email.utils import parsedate_to_datetime
+
+                    try:
+                        dt = parsedate_to_datetime(file_info["last_modified"])
+                        modified_at = int(dt.timestamp())
+                    except (ValueError, KeyError):
+                        pass
+
+                if initial_sync:
+                    # Send everything on first sync - write placeholder first
+                    await write_placeholder_point(
+                        doc_id=file_id,
+                        doc_type="file",
+                        user_id=user_id,
+                        modified_at=modified_at,
+                        file_path=file_path,
+                    )
+                    await send_stream.send(
+                        DocumentTask(
+                            user_id=user_id,
+                            doc_id=file_id,  # Use numeric file ID
+                            doc_type="file",
+                            operation="index",
+                            modified_at=modified_at,
+                            file_path=file_path,  # Pass file path for content retrieval
+                        )
+                    )
+                    file_queued += 1
+                else:
+                    # Incremental sync: check if file exists and compare modified_at
+                    # If file reappeared, remove from potentially_deleted
+                    file_key = (user_id, file_id)
+                    if file_key in _potentially_deleted:
+                        logger.debug(
+                            f"File {file_path} (ID: {file_id}) reappeared, removing from deletion grace period"
+                        )
+                        del _potentially_deleted[file_key]
+
+                    # Query Qdrant for existing entry (placeholder or real)
+                    existing_metadata = await query_document_metadata(
+                        doc_id=file_id, doc_type="file", user_id=user_id
+                    )
+
+                    # Send if never indexed or modified since last index
+                    # Compare against stored modified_at (not indexed_at!)
+                    needs_indexing = False
+                    if existing_metadata is None:
+                        # Never seen before
+                        needs_indexing = True
+                    elif existing_metadata.get("modified_at", 0) < modified_at:
+                        # File modified since last indexing
+                        needs_indexing = True
+                    elif existing_metadata.get("is_placeholder", False):
+                        # Placeholder exists - check if it's stale (processing may have failed)
+                        # Only requeue if placeholder is older than 5x scan interval
+                        # (Large PDFs can take 3-4 minutes to process)
+                        queued_at = existing_metadata.get("queued_at", 0)
+                        placeholder_age = time.time() - queued_at
+                        stale_threshold = get_settings().vector_sync_scan_interval * 5
+                        if placeholder_age > stale_threshold:
+                            logger.debug(
+                                f"Found stale placeholder for file {file_path} (ID: {file_id}) "
+                                f"(age={placeholder_age:.1f}s), requeuing"
+                            )
+                            needs_indexing = True
+                        else:
+                            logger.debug(
+                                f"Skipping file {file_path} (ID: {file_id}) with recent placeholder "
+                                f"(age={placeholder_age:.1f}s < {stale_threshold:.1f}s)"
+                            )
+
+                    if needs_indexing:
+                        # Write placeholder before queuing
+                        await write_placeholder_point(
+                            doc_id=file_id,
+                            doc_type="file",
+                            user_id=user_id,
+                            modified_at=modified_at,
+                            file_path=file_path,
+                        )
+                        await send_stream.send(
+                            DocumentTask(
+                                user_id=user_id,
+                                doc_id=file_id,  # Use numeric file ID
+                                doc_type="file",
+                                operation="index",
+                                modified_at=modified_at,
+                                file_path=file_path,  # Pass file path for content retrieval
+                            )
+                        )
+                        file_queued += 1
+
+            logger.info(
+                f"[SCAN-{scan_id}] Found {file_count} tagged PDFs for {user_id}"
+            )
+            record_vector_sync_scan(file_count)
+
+            # Check for deleted files (not initial sync)
+            if not initial_sync:
+                for file_id in indexed_file_ids:
+                    if file_id not in nextcloud_file_ids:
+                        file_key = (user_id, file_id)
+
+                        if file_key in _potentially_deleted:
+                            # Check if grace period elapsed
+                            first_missing_time = _potentially_deleted[file_key]
+                            time_missing = current_time - first_missing_time
+
+                            if time_missing >= grace_period:
+                                # Grace period elapsed, send for deletion
+                                logger.info(
+                                    f"File ID {file_id} missing for {time_missing:.1f}s "
+                                    f"(>{grace_period:.1f}s grace period), sending deletion"
+                                )
+                                await send_stream.send(
+                                    DocumentTask(
+                                        user_id=user_id,
+                                        doc_id=file_id,  # Use numeric file ID
+                                        doc_type="file",
+                                        operation="delete",
+                                        modified_at=0,
+                                    )
+                                )
+                                file_queued += 1
+                                del _potentially_deleted[file_key]
+                        else:
+                            # First time missing, add to grace period tracking
+                            logger.debug(
+                                f"File ID {file_id} missing for first time, starting grace period"
+                            )
+                            _potentially_deleted[file_key] = current_time
+
+        except Exception as e:
+            logger.warning(f"Failed to scan tagged files for {user_id}: {e}")
+
+        queued += file_queued
+
+        # Scan News items (starred + unread)
+        news_queued = 0
+        try:
+            news_queued = await scan_news_items(
+                user_id=user_id,
+                send_stream=send_stream,
+                nc_client=nc_client,
+                initial_sync=initial_sync,
+                scan_id=scan_id,
+            )
+            queued += news_queued
+        except Exception as e:
+            logger.warning(f"Failed to scan news items for {user_id}: {e}")
+
         if queued > 0:
-            logger.info(f"Sent {queued} documents for incremental sync: {user_id}")
+            logger.info(
+                f"Sent {queued} documents ({file_queued} files, {news_queued} news items) for incremental sync: {user_id}"
+            )
         else:
             logger.debug(f"No changes detected for {user_id}")
+
+
+async def scan_news_items(
+    user_id: str,
+    send_stream: MemoryObjectSendStream[DocumentTask],
+    nc_client: NextcloudClient,
+    initial_sync: bool,
+    scan_id: int,
+) -> int:
+    """
+    Scan user's News items and queue changed items for indexing.
+
+    Indexes all items from the user's feeds. The News app's auto-purge
+    feature (default: 200 items per feed) naturally limits the total
+    number of items, making explicit filtering unnecessary.
+
+    Args:
+        user_id: User to scan
+        send_stream: Stream to send changed documents to processors
+        nc_client: Authenticated Nextcloud client
+        initial_sync: If True, send all documents (first-time sync)
+        scan_id: Scan identifier for logging
+
+    Returns:
+        Number of items queued for processing
+    """
+    from nextcloud_mcp_server.client.news import NewsItemType
+
+    settings = get_settings()
+    queued = 0
+
+    # Get indexed news item IDs from Qdrant (for deletion tracking)
+    indexed_item_ids: set[str] = set()
+    if not initial_sync:
+        qdrant_client = await get_qdrant_client()
+        scroll_result = await qdrant_client.scroll(
+            collection_name=settings.get_collection_name(),
+            scroll_filter=Filter(
+                must=[
+                    FieldCondition(key="user_id", match=MatchValue(value=user_id)),
+                    FieldCondition(key="doc_type", match=MatchValue(value="news_item")),
+                ]
+            ),
+            with_payload=["doc_id"],
+            with_vectors=False,
+            limit=10000,
+        )
+        indexed_item_ids = {
+            point.payload["doc_id"]
+            for point in (scroll_result[0] or [])
+            if point.payload is not None
+        }
+        logger.debug(f"Found {len(indexed_item_ids)} indexed news items in Qdrant")
+
+    # Fetch all items (News app caps at ~200 per feed via auto-purge)
+    all_items = await nc_client.news.get_items(
+        batch_size=-1,
+        type_=NewsItemType.ALL,
+        get_read=True,
+    )
+    logger.debug(f"[SCAN-{scan_id}] Found {len(all_items)} news items")
+
+    item_count = len(all_items)
+    nextcloud_item_ids: set[str] = set()
+
+    for item in all_items:
+        doc_id = str(item["id"])
+        nextcloud_item_ids.add(doc_id)
+
+        # Use lastModified timestamp (microseconds in News API)
+        modified_at = item.get("lastModified", 0)
+        # Convert to seconds if needed (News API uses microseconds)
+        if modified_at > 10000000000:  # > year 2286 in seconds
+            modified_at = modified_at // 1000000
+
+        if initial_sync:
+            # Send everything on first sync - write placeholder first
+            await write_placeholder_point(
+                doc_id=doc_id,
+                doc_type="news_item",
+                user_id=user_id,
+                modified_at=modified_at,
+            )
+            await send_stream.send(
+                DocumentTask(
+                    user_id=user_id,
+                    doc_id=doc_id,
+                    doc_type="news_item",
+                    operation="index",
+                    modified_at=modified_at,
+                )
+            )
+            queued += 1
+        else:
+            # Incremental sync: check if item exists and compare modified_at
+            doc_key = (user_id, doc_id)
+            if doc_key in _potentially_deleted:
+                logger.debug(
+                    f"News item {doc_id} reappeared, removing from deletion grace period"
+                )
+                del _potentially_deleted[doc_key]
+
+            # Query Qdrant for existing entry
+            existing_metadata = await query_document_metadata(
+                doc_id=doc_id, doc_type="news_item", user_id=user_id
+            )
+
+            needs_indexing = False
+            if existing_metadata is None:
+                needs_indexing = True
+            elif existing_metadata.get("modified_at", 0) < modified_at:
+                needs_indexing = True
+            elif existing_metadata.get("is_placeholder", False):
+                queued_at = existing_metadata.get("queued_at", 0)
+                placeholder_age = time.time() - queued_at
+                stale_threshold = settings.vector_sync_scan_interval * 5
+                if placeholder_age > stale_threshold:
+                    logger.debug(
+                        f"Found stale placeholder for news item {doc_id} "
+                        f"(age={placeholder_age:.1f}s), requeuing"
+                    )
+                    needs_indexing = True
+
+            if needs_indexing:
+                await write_placeholder_point(
+                    doc_id=doc_id,
+                    doc_type="news_item",
+                    user_id=user_id,
+                    modified_at=modified_at,
+                )
+                await send_stream.send(
+                    DocumentTask(
+                        user_id=user_id,
+                        doc_id=doc_id,
+                        doc_type="news_item",
+                        operation="index",
+                        modified_at=modified_at,
+                    )
+                )
+                queued += 1
+
+    logger.info(
+        f"[SCAN-{scan_id}] Found {item_count} news items (starred+unread) for {user_id}"
+    )
+    record_vector_sync_scan(item_count)
+
+    # Check for deleted items (not initial sync)
+    # Items become "deleted" when they are no longer starred AND become read
+    if not initial_sync:
+        grace_period = settings.vector_sync_scan_interval * 1.5
+        current_time = time.time()
+
+        for doc_id in indexed_item_ids:
+            if doc_id not in nextcloud_item_ids:
+                doc_key = (user_id, doc_id)
+
+                if doc_key in _potentially_deleted:
+                    first_missing_time = _potentially_deleted[doc_key]
+                    time_missing = current_time - first_missing_time
+
+                    if time_missing >= grace_period:
+                        logger.info(
+                            f"News item {doc_id} missing for {time_missing:.1f}s "
+                            f"(>{grace_period:.1f}s grace period), sending deletion"
+                        )
+                        await send_stream.send(
+                            DocumentTask(
+                                user_id=user_id,
+                                doc_id=doc_id,
+                                doc_type="news_item",
+                                operation="delete",
+                                modified_at=0,
+                            )
+                        )
+                        queued += 1
+                        del _potentially_deleted[doc_key]
+                else:
+                    logger.debug(
+                        f"News item {doc_id} missing for first time, starting grace period"
+                    )
+                    _potentially_deleted[doc_key] = current_time
+
+    return queued

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nextcloud-mcp-server"
-version = "0.35.0"
+version = "0.49.0"
 description = "Model Context Protocol (MCP) server for Nextcloud integration - enables AI assistants to interact with Nextcloud data"
 authors = [
     {name = "Chris Coutinho", email = "chris@coutinho.io"}
@@ -10,9 +10,9 @@ license = {text = "AGPL-3.0-only"}
 requires-python = ">=3.11"
 keywords = ["nextcloud", "mcp", "model-context-protocol", "llm", "ai", "claude", "webdav", "caldav", "carddav"]
 dependencies = [
-    "mcp[cli] (>=1.21,<1.22)",
+    "mcp[cli] (>=1.23,<1.24)",
     "httpx (>=0.28.1,<0.29.0)",
-    "pillow (>=12.0.0,<12.1.0)",
+    "pillow (>=10.3.0,<12.0.0)", # Compatible with fastembed
     "icalendar (>=6.0.0,<7.0.0)",
     "pythonvcard4>=0.2.0",
     "pydantic>=2.11.4",
@@ -22,6 +22,9 @@ dependencies = [
     "aiosqlite>=0.20.0", # Async SQLite for refresh token storage
     "authlib>=1.6.5",
     "qdrant-client>=1.7.0",
+    "fastembed>=0.7.3", # BM25 sparse vector embeddings for hybrid search
+    "anthropic>=0.42.0", # For RAG evaluation with Anthropic LLMs
+    "boto3>=1.35.0", # For Amazon Bedrock provider (optional)
     # Observability dependencies
     "prometheus-client>=0.21.0", # Prometheus metrics
     "opentelemetry-api>=1.28.2", # OpenTelemetry API
@@ -31,6 +34,13 @@ dependencies = [
     "opentelemetry-instrumentation-logging>=0.49b2", # Logging integration
     "opentelemetry-exporter-otlp-proto-grpc>=1.28.2", # OTLP gRPC exporter
     "python-json-logger>=3.2.0", # Structured JSON logging
+    "jinja2>=3.1.6",
+    "langchain-text-splitters>=1.0.0",
+    "markdownify>=0.14.1",  # HTML to Markdown conversion for News items
+    "pymupdf>=1.26.6",
+    "pymupdf4llm>=0.2.2",
+    "pymupdf-layout>=1.26.6",
+    "openai>=2.8.1",
 ]
 classifiers = [
     "Development Status :: 4 - Beta",
@@ -102,9 +112,8 @@ module-root = ""
 
 [dependency-groups]
 dev = [
-    "anthropic>=0.42.0",  # For RAG evaluation with Anthropic LLMs
     "commitizen>=4.8.2",
-    "datasets>=3.3.0",  # For BeIR nfcorpus dataset loading
+    "datasets>=3.3.0", # For BeIR nfcorpus dataset loading
     "ipython>=9.2.0",
     "playwright>=1.49.1",
     "pytest>=8.3.5",
@@ -119,6 +128,7 @@ dev = [
 
 [project.scripts]
 nextcloud-mcp-server = "nextcloud_mcp_server.cli:run"
+smithery-main = "nextcloud_mcp_server.smithery_main:main"
 
 [[tool.uv.index]]
 name = "testpypi"

renovate.json

@@ -4,5 +4,11 @@
     "config:best-practices",
     "mergeConfidence:all-badges"
   ],
-  "dependencyDashboard": true
+  "dependencyDashboard": true,
+  "packageRules": [
+    {
+      "matchPackageNames": ["pillow"],
+      "allowedVersions": "<12.0.0"
+    }
+  ]
 }

smithery.yaml

@@ -0,0 +1,38 @@
+# Smithery configuration for Nextcloud MCP Server
+# See: https://smithery.ai/docs/build/configuration
+# ADR-016: Stateless deployment mode for multi-user public Nextcloud instances
+
+runtime: "container"
+
+build:
+  dockerfile: "Dockerfile.smithery"
+  dockerBuildPath: "."
+
+startCommand:
+  type: "http"
+  configSchema:
+    type: "object"
+    required:
+      - "nextcloud_url"
+      - "username"
+      - "app_password"
+    properties:
+      nextcloud_url:
+        type: "string"
+        title: "Nextcloud URL"
+        description: "Your Nextcloud instance URL (e.g., https://cloud.example.com). Must be publicly accessible."
+        pattern: "^https?://.+"
+      username:
+        type: "string"
+        title: "Username"
+        description: "Your Nextcloud username"
+        minLength: 1
+      app_password:
+        type: "string"
+        title: "App Password"
+        description: "Nextcloud app password. Generate at Settings > Security > App passwords. Do NOT use your main password."
+        minLength: 1
+  exampleConfig:
+    nextcloud_url: "https://cloud.example.com"
+    username: "alice"
+    app_password: "xxxxx-xxxxx-xxxxx-xxxxx-xxxxx"

tests/client/conftest.py

@@ -480,3 +480,222 @@ def create_mock_table_row_ocs_response(
     ocs_response = {"ocs": {"meta": {"status": "ok"}, "data": row_data}}
 
     return create_mock_response(status_code=200, json_data=ocs_response)
+
+
+# ============================================================================
+# News Mock Response Helpers
+# ============================================================================
+
+
+def create_mock_news_folders_response(
+    folders: list[dict] | None = None,
+) -> httpx.Response:
+    """Create a mock response for News folders list.
+
+    Args:
+        folders: List of folder dictionaries. If None, returns empty list.
+
+    Returns:
+        Mock httpx.Response with folders data
+    """
+    if folders is None:
+        folders = []
+
+    return create_mock_response(status_code=200, json_data={"folders": folders})
+
+
+def create_mock_news_folder_response(
+    folder_id: int = 1,
+    name: str = "Test Folder",
+    **kwargs,
+) -> httpx.Response:
+    """Create a mock response for a News folder.
+
+    Args:
+        folder_id: Folder ID
+        name: Folder name
+        **kwargs: Additional folder fields
+
+    Returns:
+        Mock httpx.Response with folder data
+    """
+    folder_data = {
+        "id": folder_id,
+        "name": name,
+        **kwargs,
+    }
+
+    return create_mock_response(status_code=200, json_data={"folders": [folder_data]})
+
+
+def create_mock_news_feeds_response(
+    feeds: list[dict] | None = None,
+    starred_count: int = 0,
+    newest_item_id: int | None = None,
+) -> httpx.Response:
+    """Create a mock response for News feeds list.
+
+    Args:
+        feeds: List of feed dictionaries. If None, returns empty list.
+        starred_count: Number of starred items
+        newest_item_id: ID of newest item
+
+    Returns:
+        Mock httpx.Response with feeds data
+    """
+    if feeds is None:
+        feeds = []
+
+    data = {
+        "feeds": feeds,
+        "starredCount": starred_count,
+    }
+    if newest_item_id is not None:
+        data["newestItemId"] = newest_item_id
+
+    return create_mock_response(status_code=200, json_data=data)
+
+
+def create_mock_news_feed_response(
+    feed_id: int = 1,
+    url: str = "https://example.com/feed",
+    title: str = "Test Feed",
+    favicon_link: str | None = None,
+    folder_id: int | None = None,
+    unread_count: int = 0,
+    **kwargs,
+) -> httpx.Response:
+    """Create a mock response for a News feed.
+
+    Args:
+        feed_id: Feed ID
+        url: Feed URL
+        title: Feed title
+        favicon_link: Favicon URL
+        folder_id: Parent folder ID
+        unread_count: Number of unread items
+        **kwargs: Additional feed fields
+
+    Returns:
+        Mock httpx.Response with feed data
+    """
+    feed_data = {
+        "id": feed_id,
+        "url": url,
+        "title": title,
+        "faviconLink": favicon_link,
+        "folderId": folder_id,
+        "unreadCount": unread_count,
+        "link": kwargs.get("link", "https://example.com"),
+        "added": kwargs.get("added", 1700000000),
+        "updateErrorCount": kwargs.get("updateErrorCount", 0),
+        "lastUpdateError": kwargs.get("lastUpdateError"),
+        **{
+            k: v
+            for k, v in kwargs.items()
+            if k not in ["link", "added", "updateErrorCount", "lastUpdateError"]
+        },
+    }
+
+    return create_mock_response(status_code=200, json_data={"feeds": [feed_data]})
+
+
+def create_mock_news_items_response(
+    items: list[dict] | None = None,
+) -> httpx.Response:
+    """Create a mock response for News items list.
+
+    Args:
+        items: List of item dictionaries. If None, returns empty list.
+
+    Returns:
+        Mock httpx.Response with items data
+    """
+    if items is None:
+        items = []
+
+    return create_mock_response(status_code=200, json_data={"items": items})
+
+
+def create_mock_news_item(
+    item_id: int = 1,
+    feed_id: int = 1,
+    title: str = "Test Article",
+    body: str = "<p>Test content</p>",
+    url: str = "https://example.com/article",
+    author: str | None = "Test Author",
+    pub_date: int = 1700000000,
+    unread: bool = True,
+    starred: bool = False,
+    **kwargs,
+) -> dict:
+    """Create a mock News item dictionary.
+
+    Args:
+        item_id: Item ID
+        feed_id: Parent feed ID
+        title: Article title
+        body: Article body (HTML)
+        url: Article URL
+        author: Article author
+        pub_date: Publication timestamp (Unix)
+        unread: Whether item is unread
+        starred: Whether item is starred
+        **kwargs: Additional item fields
+
+    Returns:
+        Item dictionary
+    """
+    return {
+        "id": item_id,
+        "feedId": feed_id,
+        "title": title,
+        "body": body,
+        "url": url,
+        "author": author,
+        "pubDate": pub_date,
+        "unread": unread,
+        "starred": starred,
+        "guid": kwargs.get("guid", f"guid-{item_id}"),
+        "guidHash": kwargs.get("guidHash", f"hash-{item_id}"),
+        "lastModified": kwargs.get("lastModified", pub_date * 1000000),
+        "enclosureLink": kwargs.get("enclosureLink"),
+        "enclosureMime": kwargs.get("enclosureMime"),
+        "fingerprint": kwargs.get("fingerprint", f"fp-{item_id}"),
+        "contentHash": kwargs.get("contentHash", f"ch-{item_id}"),
+        **{
+            k: v
+            for k, v in kwargs.items()
+            if k
+            not in [
+                "guid",
+                "guidHash",
+                "lastModified",
+                "enclosureLink",
+                "enclosureMime",
+                "fingerprint",
+                "contentHash",
+            ]
+        },
+    }
+
+
+def create_mock_news_status_response(
+    version: str = "25.0.0",
+    warnings: dict | None = None,
+) -> httpx.Response:
+    """Create a mock response for News status.
+
+    Args:
+        version: News app version
+        warnings: Warning messages
+
+    Returns:
+        Mock httpx.Response with status data
+    """
+    data = {
+        "version": version,
+        "warnings": warnings or {},
+    }
+
+    return create_mock_response(status_code=200, json_data=data)

tests/client/news/test_news_api.py

@@ -0,0 +1,561 @@
+"""Unit tests for NewsClient API methods."""
+
+import logging
+
+import httpx
+import pytest
+
+from nextcloud_mcp_server.client.news import NewsClient, NewsItemType
+from tests.client.conftest import (
+    create_mock_error_response,
+    create_mock_news_feed_response,
+    create_mock_news_feeds_response,
+    create_mock_news_folder_response,
+    create_mock_news_folders_response,
+    create_mock_news_item,
+    create_mock_news_items_response,
+    create_mock_news_status_response,
+    create_mock_response,
+)
+
+logger = logging.getLogger(__name__)
+
+# Mark all tests in this module as unit tests
+pytestmark = pytest.mark.unit
+
+
+# ============================================================================
+# Folder Tests
+# ============================================================================
+
+
+async def test_news_api_get_folders(mocker):
+    """Test that get_folders correctly parses the API response."""
+    mock_response = create_mock_news_folders_response(
+        folders=[
+            {"id": 1, "name": "Tech"},
+            {"id": 2, "name": "News"},
+        ]
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    folders = await client.get_folders()
+
+    assert len(folders) == 2
+    assert folders[0]["id"] == 1
+    assert folders[0]["name"] == "Tech"
+    assert folders[1]["name"] == "News"
+
+    mock_make_request.assert_called_once_with("GET", "/apps/news/api/v1-3/folders")
+
+
+async def test_news_api_create_folder(mocker):
+    """Test that create_folder correctly creates a folder."""
+    mock_response = create_mock_news_folder_response(folder_id=3, name="New Folder")
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    folder = await client.create_folder(name="New Folder")
+
+    assert folder["id"] == 3
+    assert folder["name"] == "New Folder"
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/folders", json={"name": "New Folder"}
+    )
+
+
+async def test_news_api_rename_folder(mocker):
+    """Test that rename_folder makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.rename_folder(folder_id=1, name="Renamed")
+
+    mock_make_request.assert_called_once_with(
+        "PUT", "/apps/news/api/v1-3/folders/1", json={"name": "Renamed"}
+    )
+
+
+async def test_news_api_delete_folder(mocker):
+    """Test that delete_folder makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.delete_folder(folder_id=1)
+
+    mock_make_request.assert_called_once_with("DELETE", "/apps/news/api/v1-3/folders/1")
+
+
+# ============================================================================
+# Feed Tests
+# ============================================================================
+
+
+async def test_news_api_get_feeds(mocker):
+    """Test that get_feeds correctly parses the API response."""
+    mock_response = create_mock_news_feeds_response(
+        feeds=[
+            {"id": 1, "url": "https://example.com/feed1", "title": "Feed 1"},
+            {"id": 2, "url": "https://example.com/feed2", "title": "Feed 2"},
+        ],
+        starred_count=5,
+        newest_item_id=100,
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_feeds()
+
+    assert len(result["feeds"]) == 2
+    assert result["starredCount"] == 5
+    assert result["newestItemId"] == 100
+
+    mock_make_request.assert_called_once_with("GET", "/apps/news/api/v1-3/feeds")
+
+
+async def test_news_api_create_feed(mocker):
+    """Test that create_feed correctly creates a feed."""
+    mock_response = create_mock_news_feed_response(
+        feed_id=10, url="https://example.com/new-feed", title="New Feed"
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    feed = await client.create_feed(url="https://example.com/new-feed")
+
+    assert feed["id"] == 10
+    assert feed["url"] == "https://example.com/new-feed"
+
+    mock_make_request.assert_called_once_with(
+        "POST",
+        "/apps/news/api/v1-3/feeds",
+        json={"url": "https://example.com/new-feed"},
+    )
+
+
+async def test_news_api_create_feed_with_folder(mocker):
+    """Test that create_feed correctly creates a feed in a folder."""
+    mock_response = create_mock_news_feed_response(
+        feed_id=10, url="https://example.com/feed", folder_id=5
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    feed = await client.create_feed(url="https://example.com/feed", folder_id=5)
+
+    assert feed["folderId"] == 5
+
+    mock_make_request.assert_called_once_with(
+        "POST",
+        "/apps/news/api/v1-3/feeds",
+        json={"url": "https://example.com/feed", "folderId": 5},
+    )
+
+
+async def test_news_api_delete_feed(mocker):
+    """Test that delete_feed makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.delete_feed(feed_id=10)
+
+    mock_make_request.assert_called_once_with("DELETE", "/apps/news/api/v1-3/feeds/10")
+
+
+async def test_news_api_move_feed(mocker):
+    """Test that move_feed makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.move_feed(feed_id=10, folder_id=5)
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/feeds/10/move", json={"folderId": 5}
+    )
+
+
+async def test_news_api_rename_feed(mocker):
+    """Test that rename_feed makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.rename_feed(feed_id=10, title="Renamed Feed")
+
+    mock_make_request.assert_called_once_with(
+        "POST",
+        "/apps/news/api/v1-3/feeds/10/rename",
+        json={"feedTitle": "Renamed Feed"},
+    )
+
+
+# ============================================================================
+# Item Tests
+# ============================================================================
+
+
+async def test_news_api_get_items(mocker):
+    """Test that get_items correctly parses the API response."""
+    items = [
+        create_mock_news_item(item_id=1, title="Article 1"),
+        create_mock_news_item(item_id=2, title="Article 2"),
+    ]
+    mock_response = create_mock_news_items_response(items=items)
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_items()
+
+    assert len(result) == 2
+    assert result[0]["title"] == "Article 1"
+    assert result[1]["title"] == "Article 2"
+
+    # Verify default parameters
+    call_args = mock_make_request.call_args
+    assert call_args[0] == ("GET", "/apps/news/api/v1-3/items")
+    params = call_args[1]["params"]
+    assert params["batchSize"] == 50
+    assert params["type"] == NewsItemType.ALL
+
+
+async def test_news_api_get_items_starred(mocker):
+    """Test that get_items with STARRED type filters correctly."""
+    items = [create_mock_news_item(item_id=1, starred=True)]
+    mock_response = create_mock_news_items_response(items=items)
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_items(type_=NewsItemType.STARRED)
+
+    assert len(result) == 1
+    assert result[0]["starred"] is True
+
+    call_args = mock_make_request.call_args
+    params = call_args[1]["params"]
+    assert params["type"] == NewsItemType.STARRED
+
+
+async def test_news_api_get_items_unread_only(mocker):
+    """Test that get_items with get_read=False filters correctly."""
+    items = [create_mock_news_item(item_id=1, unread=True)]
+    mock_response = create_mock_news_items_response(items=items)
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_items(get_read=False)
+
+    assert len(result) == 1
+
+    call_args = mock_make_request.call_args
+    params = call_args[1]["params"]
+    assert params["getRead"] == "false"
+
+
+async def test_news_api_get_item(mocker):
+    """Test that get_item fetches a single item by ID."""
+    item = create_mock_news_item(item_id=123, title="Single Item")
+    mock_response = create_mock_response(status_code=200, json_data=item)
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_item(item_id=123)
+
+    assert result["id"] == 123
+    assert result["title"] == "Single Item"
+
+    mock_make_request.assert_called_once_with("GET", "/apps/news/api/v1-3/items/123")
+
+
+async def test_news_api_get_updated_items(mocker):
+    """Test that get_updated_items correctly calls the updated endpoint."""
+    items = [create_mock_news_item(item_id=1)]
+    mock_response = create_mock_news_items_response(items=items)
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    result = await client.get_updated_items(last_modified=1700000000)
+
+    assert len(result) == 1
+
+    call_args = mock_make_request.call_args
+    assert call_args[0] == ("GET", "/apps/news/api/v1-3/items/updated")
+    params = call_args[1]["params"]
+    assert params["lastModified"] == 1700000000
+
+
+async def test_news_api_mark_item_read(mocker):
+    """Test that mark_item_read makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.mark_item_read(item_id=123)
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/123/read"
+    )
+
+
+async def test_news_api_mark_item_unread(mocker):
+    """Test that mark_item_unread makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.mark_item_unread(item_id=123)
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/123/unread"
+    )
+
+
+async def test_news_api_star_item(mocker):
+    """Test that star_item makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.star_item(item_id=123)
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/123/star"
+    )
+
+
+async def test_news_api_unstar_item(mocker):
+    """Test that unstar_item makes the correct API call."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.unstar_item(item_id=123)
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/123/unstar"
+    )
+
+
+async def test_news_api_mark_items_read_multiple(mocker):
+    """Test that mark_items_read makes the correct API call for multiple items."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.mark_items_read(item_ids=[1, 2, 3])
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/read/multiple", json={"itemIds": [1, 2, 3]}
+    )
+
+
+async def test_news_api_star_items_multiple(mocker):
+    """Test that star_items makes the correct API call for multiple items."""
+    mock_response = create_mock_response(status_code=200, json_data={})
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    await client.star_items(item_ids=[1, 2, 3])
+
+    mock_make_request.assert_called_once_with(
+        "POST", "/apps/news/api/v1-3/items/star/multiple", json={"itemIds": [1, 2, 3]}
+    )
+
+
+# ============================================================================
+# Status Tests
+# ============================================================================
+
+
+async def test_news_api_get_status(mocker):
+    """Test that get_status correctly parses the API response."""
+    mock_response = create_mock_news_status_response(
+        version="25.0.0",
+        warnings={"improperlyConfiguredCron": False},
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    status = await client.get_status()
+
+    assert status["version"] == "25.0.0"
+    assert "warnings" in status
+
+    mock_make_request.assert_called_once_with("GET", "/apps/news/api/v1-3/status")
+
+
+async def test_news_api_get_version(mocker):
+    """Test that get_version correctly parses the API response."""
+    mock_response = create_mock_response(
+        status_code=200, json_data={"version": "25.0.0"}
+    )
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(
+        NewsClient, "_make_request", return_value=mock_response
+    )
+
+    client = NewsClient(mock_client, "testuser")
+    version = await client.get_version()
+
+    assert version == "25.0.0"
+
+    mock_make_request.assert_called_once_with("GET", "/apps/news/api/v1-3/version")
+
+
+# ============================================================================
+# Error Handling Tests
+# ============================================================================
+
+
+async def test_news_api_create_folder_conflict(mocker):
+    """Test that create_folder raises HTTPStatusError on 409 conflict."""
+    error_response = create_mock_error_response(409, "Folder name already exists")
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(NewsClient, "_make_request")
+    mock_make_request.side_effect = httpx.HTTPStatusError(
+        "409 Conflict",
+        request=httpx.Request("POST", "http://test.local"),
+        response=error_response,
+    )
+
+    client = NewsClient(mock_client, "testuser")
+
+    with pytest.raises(httpx.HTTPStatusError) as excinfo:
+        await client.create_folder(name="Existing Folder")
+
+    assert excinfo.value.response.status_code == 409
+
+
+async def test_news_api_delete_feed_not_found(mocker):
+    """Test that delete_feed raises HTTPStatusError on 404."""
+    error_response = create_mock_error_response(404, "Feed not found")
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(NewsClient, "_make_request")
+    mock_make_request.side_effect = httpx.HTTPStatusError(
+        "404 Not Found",
+        request=httpx.Request("DELETE", "http://test.local"),
+        response=error_response,
+    )
+
+    client = NewsClient(mock_client, "testuser")
+
+    with pytest.raises(httpx.HTTPStatusError) as excinfo:
+        await client.delete_feed(feed_id=999999)
+
+    assert excinfo.value.response.status_code == 404
+
+
+async def test_news_api_create_feed_invalid_url(mocker):
+    """Test that create_feed raises HTTPStatusError on 422 for invalid URL."""
+    error_response = create_mock_error_response(422, "Invalid feed URL")
+
+    mock_client = mocker.AsyncMock(spec=httpx.AsyncClient)
+    mock_make_request = mocker.patch.object(NewsClient, "_make_request")
+    mock_make_request.side_effect = httpx.HTTPStatusError(
+        "422 Unprocessable Entity",
+        request=httpx.Request("POST", "http://test.local"),
+        response=error_response,
+    )
+
+    client = NewsClient(mock_client, "testuser")
+
+    with pytest.raises(httpx.HTTPStatusError) as excinfo:
+        await client.create_feed(url="not-a-valid-url")
+
+    assert excinfo.value.response.status_code == 422

tests/conftest.py

@@ -9,7 +9,6 @@ import pytest
 from httpx import HTTPStatusError
 from mcp import ClientSession
 from mcp.client.session import RequestContext
-from mcp.client.sse import sse_client
 from mcp.client.streamable_http import streamablehttp_client
 from mcp.types import ElicitRequestParams, ElicitResult, ErrorData
 
@@ -114,6 +113,7 @@ async def create_mcp_client_session(
     token: str | None = None,
     client_name: str = "MCP",
     elicitation_callback: Any = None,
+    sampling_callback: Any = None,
 ) -> AsyncGenerator[ClientSession, Any]:
     """
     Factory function to create an MCP client session with proper lifecycle management.
@@ -133,6 +133,8 @@ async def create_mcp_client_session(
         client_name: Client name for logging (e.g., "OAuth MCP (Playwright)")
         elicitation_callback: Optional callback for handling elicitation requests.
             Should match signature: async def callback(context: RequestContext, params: ElicitRequestParams) -> ElicitResult | ErrorData
+        sampling_callback: Optional callback for handling sampling (LLM generation) requests.
+            Should match signature: async def callback(context: RequestContext, params: CreateMessageRequestParams) -> CreateMessageResult | ErrorData
 
     Yields:
         Initialized MCP ClientSession
@@ -156,52 +158,10 @@ async def create_mcp_client_session(
         _,
     ):
         async with ClientSession(
-            read_stream, write_stream, elicitation_callback=elicitation_callback
-        ) as session:
-            await session.initialize()
-            logger.info(f"{client_name} client session initialized successfully")
-            yield session
-
-    # Cleanup happens automatically in LIFO order - no exception suppression needed
-    logger.debug(f"{client_name} client session cleaned up successfully")
-
-
-async def create_mcp_client_session_sse(
-    url: str,
-    token: str | None = None,
-    client_name: str = "MCP",
-    elicitation_callback: Any = None,
-) -> AsyncGenerator[ClientSession, Any]:
-    """
-    Factory function to create an MCP client session using SSE transport.
-
-    Similar to create_mcp_client_session but uses SSE transport instead of streamable-http.
-    Uses native async context managers to ensure correct LIFO cleanup order.
-
-    Args:
-        url: MCP server URL (e.g., "http://localhost:8000/sse")
-        token: Optional OAuth access token for Bearer authentication
-        client_name: Client name for logging (e.g., "Basic MCP (SSE)")
-        elicitation_callback: Optional callback for handling elicitation requests
-
-    Yields:
-        Initialized MCP ClientSession
-
-    Note:
-        SSE transport is being deprecated in favor of streamable-http.
-        This function exists for compatibility testing only.
-    """
-    logger.info(f"Creating SSE client for {client_name}")
-
-    # Prepare headers with OAuth token if provided
-    headers = {"Authorization": f"Bearer {token}"} if token else None
-
-    # Use native async with - Python ensures LIFO cleanup
-    # Cleanup order will be: ClientSession.__aexit__ -> sse_client.__aexit__
-    # Note: sse_client yields only (read_stream, write_stream), not 3 values like streamablehttp_client
-    async with sse_client(url, headers=headers) as (read_stream, write_stream):
-        async with ClientSession(
-            read_stream, write_stream, elicitation_callback=elicitation_callback
+            read_stream,
+            write_stream,
+            elicitation_callback=elicitation_callback,
+            sampling_callback=sampling_callback,
         ) as session:
             await session.initialize()
             logger.info(f"{client_name} client session initialized successfully")
@@ -249,18 +209,10 @@ async def nc_client(anyio_backend) -> AsyncGenerator[NextcloudClient, Any]:
 @pytest.fixture(scope="session")
 async def nc_mcp_client(anyio_backend) -> AsyncGenerator[ClientSession, Any]:
     """
-    Fixture to create an MCP client session for integration tests using SSE transport.
+    Fixture to create an MCP client session for integration tests using streamable-http.
 
     Uses anyio pytest plugin for proper async fixture handling.
-
-    Note: SSE transport is being deprecated. This fixture uses SSE for compatibility testing.
     """
-
-    # async for session in create_mcp_client_session_sse(
-    # url="http://localhost:8000/sse", client_name="Basic MCP (SSE)"
-    # ):
-    # yield session
-
     async for session in create_mcp_client_session(
         url="http://localhost:8000/mcp",
         client_name="Basic MCP (HTTP)",

tests/integration/conftest.py

@@ -0,0 +1,26 @@
+"""Pytest configuration for integration tests.
+
+This conftest.py provides hooks and fixtures specific to integration tests,
+including the --provider flag for RAG tests.
+"""
+
+# Valid provider names
+VALID_PROVIDERS = ["openai", "ollama", "anthropic", "bedrock"]
+
+
+def pytest_addoption(parser):
+    """Add --provider command line option for RAG tests."""
+    parser.addoption(
+        "--provider",
+        action="store",
+        default=None,
+        choices=VALID_PROVIDERS,
+        help="LLM provider for RAG tests: openai, ollama, anthropic, bedrock",
+    )
+
+
+def pytest_configure(config):
+    """Configure custom markers."""
+    config.addinivalue_line(
+        "markers", "rag: mark test as RAG integration test (requires --provider flag)"
+    )