MCP RAG Server

What is MCP RAG Server?

MCP RAG Server is a Python server that provides Retrieval-Augmented Generation (RAG) functionality compliant with the Model Context Protocol (MCP). It indexes documents in formats such as Markdown, text, PowerPoint, Word, and PDF using the multilingual-e5-large embedding model, and retrieves relevant information via vector search with PostgreSQL and pgvector. It is designed for developers building MCP-based RAG applications.

How to use MCP RAG Server?

Install dependencies with uv sync, set up PostgreSQL 14+ with the pgvector extension, and configure environment variables in .env. Start the MCP server with uv run python -m src.main or use CLI commands like python -m src.cli index to index documents. MCP hosts (e.g., Claude Desktop, Cline, Cursor) can integrate the server via a JSON configuration entry specifying command and args.

Key features of MCP RAG Server

• Supports multiple document formats (Markdown, text, PowerPoint, Word, PDF)
• Embedding model configurable (default: multilingual-e5-large)
• Vector search with PostgreSQL and pgvector
• Differential indexing (processes only new or modified files)
• Context chunk and full document retrieval
• CLI tools for index management and document count

Use cases of MCP RAG Server

• Searching document repositories with natural language queries
• Building MCP-compatible RAG applications
• Indexing and querying slides, PDFs, and text documents
• Incrementally updating a document index without reprocessing unchanged files

FAQ from MCP RAG Server

What are the system requirements?

Python 3.10 or higher and PostgreSQL 14 or higher with the pgvector extension are required.

How do I change the embedding model?

Set the EMBEDDING_MODEL, EMBEDDING_DIM, and prefix environment variables in .env, then clear and reindex using python -m src.cli clear and python -m src.cli index.

How can I backup and restore the indexed data?

Backup the PostgreSQL database with pg_dump and optionally the processed document directory. To restore, set up a new PostgreSQL instance with pgvector, create the database, restore the dump, and copy processed files if needed.

What transport does the server use?

The server uses JSON-RPC over stdio.

How is authentication handled?

Authentication is not mentioned in the README.