MCP Server RAG

RAG (Retrieval-Augmented Generation) server for MCP (Model Control Protocol). This server provides vector-based document retrieval functionality to enhance LLM interactions with contextual information.

Overview

The MCP Server RAG implementation provides a bridge between your LLM applications and document collections, offering:

• Vector-based document search using ChromaDB and Sentence Transformers
• Multiple specialized collections (liv, ken, ufa, sap commerce)
• Strongly typed API responses
• Integration with the MCP server infrastructure

Installation

pip install mcp-server-rag-MiniLM

Usage

After installation, you can run the server with:

mcp-server-rag-MiniLM

Or using Python module syntax:

python -m mcp-server-rag-MiniLM

Features

• Vector-based search for contextual document retrieval
• Persistent storage of document embeddings via ChromaDB
• Multiple collection support for domain-specific searches
• Configurable via environment variables
• Fully typed response objects with detailed metadata
• Seamless integration with the MCP protocol

Available Collections

The server provides access to four specialized collections:

• liv-rag: LIV document collection (retrieve_liv_context tool)
• ken-rag: Kennametal document collection (retrieve_ken_context tool)
• ufa-rag: UFA document collection (retrieve_ufa_context tool)
• sap-comm-rag: SAP Commerce document collection (retrieve_sap_comm_context tool)

Configuration

The following environment variables can be used to configure the server:

Dependencies

• click
• mcp (version 1.2.0 or higher)
• chromadb
• sentence-transformers

Development

Building and Publishing

The repo includes a publish.sh script that helps with building and publishing the package.

./publish.sh

This script will clean previous build artifacts, install build dependencies, build the package, and check it before providing instructions for publishing to PyPI.

License

MIT