local_faiss_mcp

Overview

Local FAISS MCP Server implements the Model Context Protocol to provide local vector database functionality using FAISS for Retrieval-Augmented Generation (RAG) applications. It allows users to ingest documents in multiple formats (PDF, TXT, MD natively, plus 40+ formats with pandoc), chunk them, embed them, and then perform semantic search queries through MCP tools. The server supports custom embedding models and re-ranking for improved search relevance, making it suitable for both developers and end users looking to add local document search capabilities to AI agents.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• ingest_documentdocument: string, source?: string

  Ingest a document into the vector store

  Try:Ingest the document at ./path/to/document.pdfTry:Add this text to the vector store: FAISS is a library for efficient similarity search...Try:Ingest the research paper with source 'my_paper'
• query_rag_storequery: string, top_k?: number

  Query the vector store for relevant document chunks

  Try:Search for information about FAISS similarity searchTry:Find documents related to machine learning techniquesTry:What does this document say about vector databases?
• extract-answerquery: string, chunks: array

  Extract the most relevant answer from retrieved document chunks with citations

  Try:Extract an answer from these document chunks about FAISSTry:Based on the retrieved documents, provide a detailed answer with citationsTry:Summarize the key points from these document chunks
• summarize-documentstopic: string, chunks: array, max_length?: number

  Create a focused summary from multiple document chunks

  Try:Summarize these documents about vector databasesTry:Provide a brief summary of the research papers on machine learningTry:Create a summary of the following text

Comparable tools

Installation

pip install local-faiss-mcp

git clone https://github.com/nonatofabio/local_faiss_mcp.git
cd local_faiss_mcp
pip install -e .

{
  "mcpServers": {
    "local-faiss-mcp": {
      "command": "local-faiss-mcp",
      "args": ["--index-dir", "/path/to/index/directory"]
    }
  }
}

FAQ

What document formats are supported?

Native support for TXT, MD, and PDF. With pandoc installed, it supports DOCX, HTML, EPUB, and 40+ additional formats.

Can I use custom embedding models?

Yes, you can specify any Hugging Face sentence-transformers model using the --embed flag when starting the server.

Compare local_faiss_mcp with