RAG MCP Server

This project is a standalone FastMCP server for public-facing RAG chatbots. It exposes a small set of read-only tools over HTTP/SSE so a chatbot can search a knowledge base, search the web, inspect public GitHub repositories, and answer time-based questions without exposing write-capable actions.

Features

• FastMCP server with 5 read-only tools
• Knowledge base search using OpenAI embeddings and Pinecone
• Optional full-document hydration from AWS S3
• Web search powered by Tavily
• Public GitHub README exploration
• GitHub code search
• Current date and time lookup
• Heroku-friendly Python project layout

Tools

• search_knowledge_base
• web_search
• get_current_datetime
• explore_public_repo_readmes
• search_github_code

Environment Variables

Copy .env.example to .env and provide values for:

OPENAI_API_KEY
PINECONE_API_KEY
PINECONE_INDEX
S3_BUCKET
AWS_REGION
AWS_ACCESS_KEY_ID
AWS_SECRET_ACCESS_KEY
TAVILY_API_KEY
GITHUB_TOKEN
PORT

Run Locally

pip install -r requirements.txt
python server.py

The server listens on PORT, defaulting to 8000.