MINIRAG-MCP

An MCP Server wrapper around MiniRAG that leverages LLM sampling for client-managed inference.

Core function

MiniRag is an MCP project geared towards efficient and reliable retrieval by small language models). Minirag-MCP makes MiniRAG easier to plug into on-device agent flows.

Additional bells and whistles

• UV management of dependencies that avoid complications in minirag setup
• LLM sampling (inference calls managed by MCP client) with built-in constrained decoding for enhanced reliability

Feature roadmap

• [x] RAG retrieval
  • naive retrieval algorithm, implementing a basic query of the vector database
  • minirag algorithm, which operates on graph
• [ ] basic insertion (and reindexing)
• [ ] advanced insertion of content from the web

Quick Start

uv run server.py