MCP RAG Server

What is MCP RAG Server?

MCP RAG Server is a Model Context Protocol (MCP) server that enables Retrieval Augmented Generation (RAG) for connected LLMs by indexing project documents and providing relevant context to enhance responses. It is built with Google Genkit, ChromaDB, and Ollama, and is designed for developers who want local, privacy‑focused RAG capabilities.

How to use MCP RAG Server?

The recommended setup uses Docker Compose: clone the repository, run docker-compose up -d --build, then pull the default embedding model with docker exec ollama ollama pull nomic-embed-text. Configure environment variables (e.g., CHROMA_URL, OLLAMA_HOST, INDEX_PROJECT_ON_STARTUP) in the docker-compose.yml file. The server exposes five MCP tools (indexDocuments, queryDocuments, removeDocument, removeAllDocuments, listDocuments) for integration with any MCP client.

Key features of MCP RAG Server

• Automatic indexing of project files on startup (configurable)
• Supports .txt, .md, code files, .json, .jsonl, .csv
• Hierarchical chunking for Markdown (text and code blocks separated)
• Uses ChromaDB for persistent vector storage
• Local embeddings via Ollama (default: nomic-embed-text)
• Exposes all RAG functions as standard MCP tools

Use cases of MCP RAG Server

• Enhance LLM responses with context from local project documentation and code
• Build a fully local, privacy‑preserving document Q&A system
• Automatically index source files to improve code‑assistance tools
• Combine MCP with RAG for domain‑specific knowledge retrieval

FAQ from MCP RAG Server

What are the runtime dependencies of MCP RAG Server?

The server requires ChromaDB (vector store), Ollama (local embeddings), and Google Genkit. All are containerized when using Docker Compose.

How are documents indexed?

By default the server scans the project directory on startup and indexes all supported files. Automatic indexing can be disabled with the INDEX_PROJECT_ON_STARTUP environment variable, and manual indexing is available via the indexDocuments tool.

What file types does MCP RAG Server support?

It supports .txt, .md, code files (basic chunking), .json, .jsonl, and .csv.

How do I start MCP RAG Server?

The recommended method is Docker Compose: clone the repo, run docker-compose up -d --build, then pull the embedding model. For local development without Docker, ensure ChromaDB and Ollama are running, set the required environment variables, and run npm start.

Where are the embeddings and indexed data stored?

ChromaDB provides persistent vector storage. In the Docker Compose setup, the data is stored inside Docker volumes, ensuring it survives container restarts.