RAG-MCP Server

A general-purpose Retrieval-Augmented Generation (RAG) server using the Model Control Protocol (MCP), designed to be tested with RISC Zero’s Bonsai documentation.

Overview

This project implements a RAG server that:

• Uses MCP (Model Control Protocol) for standardized communication
• Implements RAG (Retrieval-Augmented Generation) workflow for document querying
• Can be tested with RISC Zero’s Bonsai documentation
• Supports local LLM integration through Ollama

Features

• Document ingestion and indexing
• Semantic search capabilities
• Local LLM integration
• MCP protocol compliance
• RISC Zero Bonsai documentation support

Prerequisites

• Python 3.12+
• Ollama (for local LLM support)
• Poetry (for dependency management)

Installation

1. Install Python dependencies:

poetry install

2. Install and start Ollama:

# Install Ollama
brew install ollama  # for macOS
# or
curl -fsSL https://ollama.com/install.sh | sh  # for Linux

# Start Ollama service
ollama serve

3. Pull the required model:

ollama pull llama2

Usage

1. Start the MCP server:

poetry run python mcp_server.py

2. The server will:

   • Initialize the LLM and embedding model
   • Ingest documents from the data directory
   • Process queries using the RAG workflow

3. Test with RISC Zero Bonsai docs:

   • Place RISC Zero Bonsai documentation in the data/ directory
   • Query the server about Bonsai features and implementation

Project Structure

• mcp_server.py: Main server implementation
• rag.py: RAG workflow implementation
• data/: Directory for document ingestion
• storage/: Vector store and document storage
• start_ollama.sh: Script to start Ollama service

Testing with RISC Zero Bonsai

The server is configured to work with RISC Zero’s Bonsai documentation. You can:

1. Add Bonsai documentation to the data/ directory
2. Query about Bonsai features, implementation details, and usage
3. Test the RAG workflow with Bonsai-specific questions

Made with by proofofsid