brandon/nextcloud-mcp-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: Centralize PDF processing and generate separate images per chunk

Browse Source 
Previously, pymupdf4llm.to_markdown() was called twice - once in
PyMuPDFProcessor during indexing and again in PDFHighlighter during
visualization. Different image path lengths caused different character
offsets, leading to highlighted pages not matching their chunks.

Also fixed issue where all chunks on the same page showed all highlights
instead of just their own highlight. Now restores original page contents
between chunks using xref stream caching.

Changes:
- Add PDFHighlighter class requiring pre-computed page_boundaries and
  full_text from document processor (no fallback extraction)
- Pass pre-computed data from processor to highlighter
- Extract page-relative portion of chunk text for cross-page chunks
- Add bounding box highlighting using text anchor search
- Run highlight generation in parallel with embedding/BM25
- Cache and restore page contents to isolate highlights per chunk

Results: Highlighting success rate improved from 51% to 95% (121/128).

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

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
2025-11-22 02:42:28 +01:00
parent a62a007c87
commit fffe483c02
8 changed files with 1211 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nextcloud_mcp_server/auth/static/vector-viz.css
+27
View File
@@ -190,3 +190,30 @@
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/templates/vector_viz.html
+14
View File
@@ -147,6 +147,20 @@
</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>
nextcloud_mcp_server/auth/viz_routes.py
+65 -10
View File
@@ -564,17 +564,72 @@ async def chunk_context_endpoint(request: Request) -> JSONResponse:
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
return JSONResponse(
{
"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,
}
)
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}")