nonatofabio/local-faiss-mcp

Overview

A Python server that wraps a local FAISS vector store into the Model Context Protocol. It provides drop-in semantic search and memory for agents like Claude and Copilot without external cloud services.

Best for

Best for
Developers who need a lightweight, self-hosted semantic memory for AI agents running on their own hardware.

Use cases

• Give Claude persistent memory by storing and retrieving conversation context
• Perform local semantic search on documents for Copilot agents
• Embed queries and return similar items for RAG workflows

How to use

Install

pip install local-faiss-mcp

Tools exposed

• all-MiniLM-L6-v2
• all-mpnet-base-v2
• paraphrase-multilingual-MiniLM-L12-v2
• all-MiniLM-L12-v2

Tested with

Claude Desktop, Claude Code

Notes

A Python server that wraps a local FAISS vector store into the Model Context Protocol. It provides drop-in semantic search and memory for agents like Claude and Copilot without external cloud services.

30 stars on GitHub. Last updated 2026-04-24. Licensed MIT.

Use cases

• Give Claude persistent memory by storing and retrieving conversation context
• Perform local semantic search on documents for Copilot agents
• Embed queries and return similar items for RAG workflows

Pros

• Fully local and private, no cloud dependency
• Works with any MCP-compatible agent out of the box
• Simple Python setup with minimal dependencies

Cons

• Small user base (30 stars) limits community support and examples
• Requires self-hosting and manual maintenance
• FAISS index is in-memory, not designed for large or persistent datasets

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.