MCP RAG Server
An independent Model Context Protocol (MCP) server that exposes a search_documents tool. This server connects to a PostgreSQL database (pgvector) to perform semantic retrieval. It fetches relevant document snippets and returns the raw context directly to the calling Agent/LLM (e.g., OpenWebUI), so the client's LLM can perform the final synthesis.
Built with Python, FastAPI, and the official MCP SDK.
Project Structure
mcp-rag/
├── Dockerfile # Hardened Alpine-based Docker image
├── README.md
├── requirements.txt # Python dependencies
├── .env.example # Environment variables template
└── src/
├── config.py # Pydantic settings
├── rag.py # Similarity search and retrieval logic
└── server.py # FastAPI MCP server definitionSetup & Configuration
- Copy
.env.exampleto.envand fill in your PostgreSQL connection details:
cp .env.example .env- Make sure your
POSTGRES_HOSTis accessible from the container.
Running with Docker
This project provides a hardened Alpine-based Dockerfile.
- Build the image:
docker build -t mcp-rag-server .- Run the container:
docker run -d \
--name mcp-rag \
--env-file .env \
-p 8000:8000 \
mcp-rag-serverThe MCP Server will be accessible at http://localhost:8000/mcp. You can configure your MCP-compatible clients (like OpenWebUI) to connect via SSE to this endpoint.
Available MCP Tools
search_documents(question: str, top_k: int = 5) -> str
Executes a similarity search against the pgvector store and returns the raw text snippets and their citations. Does NOT perform LLM synthesis natively.
