Files
nextcloud-mcp-server/nextcloud_mcp_server/server/semantic.py
T
Chris Coutinho e575c8e57b feat(vector): Support multiple embedding models with auto-generated collection names
This PR enables safe switching between embedding models and multi-server
deployments by implementing auto-generated Qdrant collection names based on
deployment ID and model name.

## Problem

Previously, all deployments used a single hardcoded collection name
"nextcloud_content", which caused two critical issues:

1. **Dimension mismatches when switching models**: Changing
   OLLAMA_EMBEDDING_MODEL (e.g., nomic-embed-text at 768D → all-minilm at
   384D) would cause runtime errors as vectors couldn't be inserted into a
   collection with incompatible dimensions.

2. **Collection collisions in multi-server setups**: Multiple MCP servers
   sharing a single Qdrant instance would overwrite each other's data,
   making horizontal scaling impossible.

## Solution

### Auto-Generated Collection Naming

Collections are now automatically named using the pattern:
\`{deployment-id}-{model-name}\`

**Deployment ID**: Uses \`OTEL_SERVICE_NAME\` if configured (and not default
value), otherwise falls back to \`hostname\` for simple Docker deployments.

**Model Name**: From \`OLLAMA_EMBEDDING_MODEL\` with path separators sanitized.

**Examples**:
- \`my-mcp-server-nomic-embed-text\` (with OTEL_SERVICE_NAME=my-mcp-server)
- \`mcp-container-all-minilm\` (simple Docker, hostname=mcp-container)

**Override**: Users can still set \`QDRANT_COLLECTION\` explicitly to bypass
auto-generation for backward compatibility.

### Dimension Validation

Added startup validation that checks collection dimensions match the
embedding service. If a mismatch is detected, the server fails fast with a
clear error message explaining:
- Expected vs actual dimensions
- Likely cause (model change)
- Solutions (delete collection, use different name, or revert model)

### Improved Sampling Error Handling

Enhanced MCP sampling rejection handling to treat user rejections as normal
behavior rather than errors:

- **User rejections** ("rejected", "denied") → INFO log, no traceback
- **Unsupported clients** → INFO log, no traceback
- **Other MCP errors** → WARNING log, no traceback
- **Unexpected errors** → ERROR log WITH traceback

This aligns with the MCP specification where clients SHOULD prompt users for
approval/denial of sampling requests.

## Changes

### Core Implementation

- **nextcloud_mcp_server/config.py**: Added \`get_collection_name()\` method
  with deployment ID detection and model name sanitization
- **nextcloud_mcp_server/vector/qdrant_client.py**: Dimension validation on
  collection open with helpful error messages
- **nextcloud_mcp_server/vector/{scanner,processor}.py**: Updated to use
  \`get_collection_name()\`
- **nextcloud_mcp_server/auth/userinfo_routes.py**: Vector sync status uses
  \`get_collection_name()\`
- **nextcloud_mcp_server/server/semantic.py**:
  - Updated semantic search tools to use \`get_collection_name()\`
  - Improved sampling rejection error handling (McpError vs Exception)

### Documentation

- **docs/semantic-search-architecture.md**: New comprehensive architecture
  document (557 lines) covering background sync, semantic search flow, RAG
  implementation, and deployment modes
- **docs/configuration.md**: Added detailed "Qdrant Collection Naming"
  section with examples and multi-server deployment guidance
- **docker-compose.yml**: Added comments explaining collection naming behavior
- **README.md**: Updated semantic search descriptions to clarify
  experimental status, Notes-only support, and infrastructure requirements

## Migration Guide

**For existing single-server deployments:**

Option 1 (Recommended): Use explicit collection name for continuity
\`\`\`bash
QDRANT_COLLECTION=nextcloud_content  # Keep existing collection
\`\`\`

Option 2: Allow auto-generation and re-embed
\`\`\`bash
# Remove QDRANT_COLLECTION override
# New collection will be created based on deployment ID + model
# Requires re-embedding all documents (may take time)
\`\`\`

**For new multi-server deployments:**

Set unique OTEL service names per server:
\`\`\`bash
# Server 1
OTEL_SERVICE_NAME=mcp-prod
OLLAMA_EMBEDDING_MODEL=nomic-embed-text
# → Collection: "mcp-prod-nomic-embed-text"

# Server 2
OTEL_SERVICE_NAME=mcp-staging
OLLAMA_EMBEDDING_MODEL=nomic-embed-text
# → Collection: "mcp-staging-nomic-embed-text"
\`\`\`

## Benefits

 **Safe model switching**: Each model gets its own collection, preventing
   dimension mismatch errors
 **Multi-server support**: Multiple MCP servers can share one Qdrant
   instance without conflicts
 **Clear ownership**: Collection names show which deployment and model owns
   the data
 **Better error messages**: Dimension validation provides actionable
   guidance
 **Backward compatible**: Existing deployments can continue using
   \`QDRANT_COLLECTION\` override

## Testing

Validated with:
- Single-server deployments (default hostname-based naming)
- Multi-server deployments (OTEL service name-based naming)
- Model switching scenarios (dimension validation)
- Collection override scenarios (backward compatibility)

Next steps: Testing various Ollama embedding models to investigate optimal
chunk sizes and performance characteristics.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-10 01:18:30 +01:00

574 lines
23 KiB
Python

"""Semantic search MCP tools using vector database."""
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,
ModelHint,
ModelPreferences,
SamplingMessage,
TextContent,
)
from nextcloud_mcp_server.auth import require_scopes
from nextcloud_mcp_server.context import get_client
from nextcloud_mcp_server.models.semantic import (
SamplingSearchResponse,
SemanticSearchResponse,
SemanticSearchResult,
VectorSyncStatusResponse,
)
logger = logging.getLogger(__name__)
def configure_semantic_tools(mcp: FastMCP):
"""Configure semantic search tools for MCP server."""
@mcp.tool()
@require_scopes("semantic:read")
async def nc_semantic_search(
query: str, ctx: Context, limit: int = 10, score_threshold: float = 0.7
) -> SemanticSearchResponse:
"""
Semantic search across all indexed Nextcloud apps using vector embeddings.
Searches documents by meaning rather than exact keywords across notes, calendar
events, deck cards, files, and contacts. Requires vector database synchronization
to be enabled (VECTOR_SYNC_ENABLED=true).
Args:
query: Natural language search query
limit: Maximum number of results to return (default: 10)
score_threshold: Minimum similarity score (0-1, default: 0.7)
Returns:
SemanticSearchResponse with matching documents and similarity scores
"""
from qdrant_client.models import FieldCondition, Filter, MatchValue
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
settings = get_settings()
# Check if vector sync is enabled
if not settings.vector_sync_enabled:
raise McpError(
ErrorData(
code=-1,
message="Semantic search is not enabled. Set VECTOR_SYNC_ENABLED=true and ensure vector database is configured.",
)
)
client = await get_client(ctx)
username = client.username
logger.info(
f"Semantic search: query='{query}', user={username}, "
f"limit={limit}, score_threshold={score_threshold}"
)
try:
# Generate embedding for query
embedding_service = get_embedding_service()
query_embedding = await embedding_service.embed(query)
logger.debug(
f"Generated embedding for query (dimension={len(query_embedding)})"
)
# Search Qdrant with user filtering
# Note: Currently only searching notes (doc_type="note")
# Future: Remove doc_type filter to search all apps
qdrant_client = await get_qdrant_client()
search_response = await qdrant_client.query_points(
collection_name=settings.get_collection_name(),
query=query_embedding,
query_filter=Filter(
must=[
FieldCondition(
key="user_id",
match=MatchValue(value=username),
),
FieldCondition(
key="doc_type",
match=MatchValue(value="note"),
),
]
),
limit=limit * 2, # Get extra for filtering
score_threshold=score_threshold,
with_payload=True,
with_vectors=False, # Don't return vectors to save bandwidth
)
logger.info(
f"Qdrant returned {len(search_response.points)} results "
f"(before deduplication and access verification)"
)
if search_response.points:
# Log top 3 scores to help with threshold tuning
top_scores = [p.score for p in search_response.points[:3]]
logger.debug(f"Top 3 similarity scores: {top_scores}")
# Deduplicate by document ID (multiple chunks per document)
seen_doc_ids = set()
results = []
for result in search_response.points:
doc_id = int(result.payload["doc_id"])
doc_type = result.payload.get("doc_type", "note")
# Skip if we've already seen this document
if doc_id in seen_doc_ids:
continue
seen_doc_ids.add(doc_id)
# Verify access via Nextcloud API (dual-phase authorization)
# Currently only supports notes, will be extended to other apps
if doc_type == "note":
try:
note = await client.notes.get_note(doc_id)
results.append(
SemanticSearchResult(
id=doc_id,
doc_type="note",
title=result.payload["title"],
category=note.get("category", ""),
excerpt=result.payload["excerpt"],
score=result.score,
chunk_index=result.payload["chunk_index"],
total_chunks=result.payload["total_chunks"],
)
)
if len(results) >= limit:
break
except HTTPStatusError as e:
if e.response.status_code == 403:
# User lost access, skip this document
logger.debug(f"Skipping note {doc_id}: access denied (403)")
continue
elif e.response.status_code == 404:
# Document was deleted but not yet removed from vector DB
logger.debug(
f"Skipping note {doc_id}: not found (404), "
f"likely deleted after indexing"
)
continue
else:
# Log other errors but continue processing
logger.warning(
f"Error verifying access to note {doc_id}: {e.response.status_code}"
)
continue
logger.info(
f"Returning {len(results)} results after deduplication and access verification"
)
if results:
result_details = [
f"note_{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 SemanticSearchResponse(
results=results,
query=query,
total_found=len(results),
search_method="semantic",
)
except ValueError as e:
if "No embedding provider configured" in str(e):
raise McpError(
ErrorData(
code=-1,
message="Embedding service not configured. Set OLLAMA_BASE_URL environment variable.",
)
)
raise McpError(ErrorData(code=-1, message=f"Configuration error: {str(e)}"))
except RequestError as e:
raise McpError(
ErrorData(code=-1, message=f"Network error during search: {str(e)}")
)
except Exception as e:
logger.error(f"Semantic search error: {e}", exc_info=True)
raise McpError(
ErrorData(code=-1, message=f"Semantic search failed: {str(e)}")
)
@mcp.tool()
@require_scopes("semantic:read")
async def nc_semantic_search_answer(
query: str,
ctx: Context,
limit: int = 5,
score_threshold: float = 0.7,
max_answer_tokens: int = 500,
) -> SamplingSearchResponse:
"""
Semantic search with LLM-generated answer using MCP sampling.
Retrieves relevant documents from indexed Nextcloud apps (notes, calendar, deck,
files, contacts) using vector similarity search, then uses MCP sampling to request
the client's LLM to generate a natural language answer based on the retrieved context.
This tool combines the power of semantic search (finding relevant content across
all your Nextcloud apps) with LLM generation (synthesizing that content into
coherent answers). The generated answer includes citations to specific documents
with their types, allowing users to verify claims and explore sources.
The LLM generation happens client-side via MCP sampling. The MCP client
controls which model is used, who pays for it, and whether to prompt the
user for approval. This keeps the server simple (no LLM API keys needed)
while giving users full control over their LLM interactions.
Args:
query: Natural language question to answer (e.g., "What are my Q1 objectives?" or "When is my next dentist appointment?")
ctx: MCP context for session access
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)
Returns:
SamplingSearchResponse containing:
- generated_answer: Natural language answer with citations
- sources: List of documents with excerpts and relevance scores
- model_used: Which model generated the answer
- stop_reason: Why generation stopped
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(
query=query,
ctx=ctx,
limit=limit,
score_threshold=score_threshold,
)
# 2. Handle no results case - don't waste a sampling call
if not search_response.results:
logger.debug(f"No documents found for query: {query}")
return SamplingSearchResponse(
query=query,
generated_answer="No relevant documents found in your Nextcloud content for this query.",
sources=[],
total_found=0,
search_method="semantic_sampling",
success=True,
)
# 3. Check if client supports sampling
from mcp.types import ClientCapabilities, SamplingCapability
client_has_sampling = ctx.session.check_client_capability(
ClientCapabilities(sampling=SamplingCapability())
)
# Log capability check result for debugging
logger.info(
f"Sampling capability check: client_has_sampling={client_has_sampling}, "
f"query='{query}'"
)
if hasattr(ctx.session, "_client_params") and ctx.session._client_params:
client_caps = ctx.session._client_params.capabilities
logger.debug(
f"Client advertised capabilities: "
f"roots={client_caps.roots is not None}, "
f"sampling={client_caps.sampling is not None}, "
f"experimental={client_caps.experimental is not None}"
)
if not client_has_sampling:
logger.info(
f"Client does not support sampling (query: '{query}'), "
f"returning {len(search_response.results)} documents"
)
return SamplingSearchResponse(
query=query,
generated_answer=(
f"[Sampling not supported by client]\n\n"
f"Your MCP client doesn't support answer generation. "
f"Found {search_response.total_found} relevant documents. "
f"Please review the sources below."
),
sources=search_response.results,
total_found=search_response.total_found,
search_method="semantic_sampling_unsupported",
success=True,
)
# 4. Construct context from retrieved documents
context_parts = []
for idx, result in enumerate(search_response.results, 1):
context_parts.append(
f"[Document {idx}]\n"
f"Type: {result.doc_type}\n"
f"Title: {result.title}\n"
f"Category: {result.category}\n"
f"Excerpt: {result.excerpt}\n"
f"Relevance Score: {result.score:.2f}\n"
)
context = "\n".join(context_parts)
# 5. Construct prompt - reuse user's query, add context and instructions
prompt = (
f"{query}\n\n"
f"Here are relevant documents from Nextcloud (notes, calendar events, deck cards, files, contacts):\n\n"
f"{context}\n\n"
f"Based on the documents above, please provide a comprehensive answer. "
f"Cite the document numbers when referencing specific information."
)
logger.info(
f"Initiating sampling request: query_length={len(query)}, "
f"documents={len(search_response.results)}, "
f"prompt_length={len(prompt)}, max_tokens={max_answer_tokens}"
)
# 6. Request LLM completion via MCP sampling with timeout
import anyio
try:
with anyio.fail_after(30):
sampling_result = await ctx.session.create_message(
messages=[
SamplingMessage(
role="user",
content=TextContent(type="text", text=prompt),
)
],
max_tokens=max_answer_tokens,
temperature=0.7,
model_preferences=ModelPreferences(
hints=[ModelHint(name="claude-3-5-sonnet")],
intelligencePriority=0.8,
speedPriority=0.5,
),
include_context="thisServer",
)
# 7. Extract answer from sampling response
if sampling_result.content.type == "text":
generated_answer = sampling_result.content.text
else:
# Handle non-text responses (shouldn't happen for text prompts)
generated_answer = f"Received non-text response of type: {sampling_result.content.type}"
logger.warning(
f"Unexpected content type from sampling: {sampling_result.content.type}"
)
logger.info(
f"Sampling successful: model={sampling_result.model}, "
f"stop_reason={sampling_result.stopReason}, "
f"answer_length={len(generated_answer)}"
)
return SamplingSearchResponse(
query=query,
generated_answer=generated_answer,
sources=search_response.results,
total_found=search_response.total_found,
search_method="semantic_sampling",
model_used=sampling_result.model,
stop_reason=sampling_result.stopReason,
success=True,
)
except TimeoutError:
logger.warning(
f"Sampling request timed out after 30 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"Found {search_response.total_found} relevant documents. "
f"Please review the sources below or try a simpler query."
),
sources=search_response.results,
total_found=search_response.total_found,
search_method="semantic_sampling_timeout",
success=True,
)
except McpError as e:
# Expected MCP protocol errors (user rejection, unsupported, etc.)
error_msg = str(e)
if "rejected" in error_msg.lower() or "denied" in error_msg.lower():
# User explicitly declined - this is normal, not an error
logger.info(f"User declined sampling request for query: '{query}'")
search_method = "semantic_sampling_user_declined"
user_message = "User declined to generate an answer"
elif "not supported" in error_msg.lower():
# Client doesn't support sampling - also normal
logger.info(f"Sampling not supported by client for query: '{query}'")
search_method = "semantic_sampling_unsupported"
user_message = "Sampling not supported by this client"
else:
# Other MCP protocol errors
logger.warning(
f"MCP error during sampling for query '{query}': {error_msg}"
)
search_method = "semantic_sampling_mcp_error"
user_message = f"Sampling unavailable: {error_msg}"
return SamplingSearchResponse(
query=query,
generated_answer=(
f"[{user_message}]\n\n"
f"Found {search_response.total_found} relevant documents. "
f"Please review the sources below."
),
sources=search_response.results,
total_found=search_response.total_found,
search_method=search_method,
success=True,
)
except Exception as e:
# Truly unexpected errors - these SHOULD have tracebacks
logger.error(
f"Unexpected error during sampling for query '{query}': "
f"{type(e).__name__}: {e}",
exc_info=True,
)
return SamplingSearchResponse(
query=query,
generated_answer=(
f"[Unexpected error during sampling]\n\n"
f"Found {search_response.total_found} relevant documents. "
f"Please review the sources below."
),
sources=search_response.results,
total_found=search_response.total_found,
search_method="semantic_sampling_error",
success=True,
)
@mcp.tool()
@require_scopes("semantic:read")
async def nc_get_vector_sync_status(ctx: Context) -> VectorSyncStatusResponse:
"""Get the current vector sync status.
Returns information about the vector sync process, including:
- Number of documents indexed in the vector database
- Number of documents pending processing
- Current sync status (idle, syncing, or disabled)
This is useful for determining when vector indexing is complete
after creating or updating content across all indexed apps.
"""
import os
# Check if vector sync is enabled
vector_sync_enabled = (
os.getenv("VECTOR_SYNC_ENABLED", "false").lower() == "true"
)
if not vector_sync_enabled:
return VectorSyncStatusResponse(
indexed_count=0,
pending_count=0,
status="disabled",
enabled=False,
)
try:
# Get document receive stream from lifespan context
lifespan_ctx = ctx.request_context.lifespan_context
document_receive_stream = getattr(
lifespan_ctx, "document_receive_stream", None
)
if document_receive_stream is None:
logger.debug(
"document_receive_stream not available in lifespan context"
)
return VectorSyncStatusResponse(
indexed_count=0,
pending_count=0,
status="unknown",
enabled=True,
)
# Get pending count from stream statistics
stream_stats = document_receive_stream.statistics()
pending_count = stream_stats.current_buffer_used
# Get Qdrant client and query indexed count
indexed_count = 0
try:
from nextcloud_mcp_server.config import get_settings
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_result = await qdrant_client.count(
collection_name=settings.get_collection_name()
)
indexed_count = count_result.count
except Exception as e:
logger.warning(f"Failed to query Qdrant for indexed count: {e}")
# Continue with indexed_count = 0
# Determine status
status = "syncing" if pending_count > 0 else "idle"
return VectorSyncStatusResponse(
indexed_count=indexed_count,
pending_count=pending_count,
status=status,
enabled=True,
)
except Exception as e:
logger.error(f"Error getting vector sync status: {e}")
raise McpError(
ErrorData(
code=-1,
message=f"Failed to retrieve vector sync status: {str(e)}",
)
)