mcp-local-rag

Local semantic search over private PDF documents works by ingesting files into a local retrieval system that combines vector search with BM25 keyword matching. This allows you to query technical terms while keeping data confined to your environment.

Context expansion in retrieval-augmented generation is a mechanism that reads neighboring semantic chunks when a search hit lacks enough surrounding explanation. It improves grounded answers by automatically expanding the retrieved text boundaries.

Yes, local retrieval-augmented search works for finding specific function definitions in your notes by using hybrid semantic and keyword matching. It retrieves relevant chunks from your configured document roots to locate exact technical terms.

You ingest local files for retrieval-augmented generation without exposing data externally by using a local MCP server workflow. This server provides tools to ingest raw content, manage indexing status, and delete files confined to configured document roots.

Yes, you can extract visual captions from PDF diagrams during local document ingestion. The system optionally uses a local VLM to extract figure and table captions, making visual content searchable alongside standard text.

The best way to manage indexed documents for local semantic search is through a local MCP server workflow that supports list, status, and delete operations. You can update your index by re-ingesting files as needed.