Gives Claude a local FAISS vector database for RAG workflows without external dependencies. Exposes two core tools: ingest_document for chunking and embedding files (PDF, TXT, MD natively, plus DOCX and HTML with pandoc), and query_rag_store for semantic search that returns ranked document chunks. Includes built in prompts for answer extraction with citations and document summarization. Supports custom Hugging Face embedding models and optional two stage reranking with cross encoders for better relevance. Index and metadata persist to disk between sessions. Best when you need document search capabilities in your Claude workflow but want to avoid hosted vector database services or API costs.
claude mcp add --transport stdio nonatofabio-local_faiss_mcp uvx local-faiss-mcp