📚 RAG MCP

This repository contains the RAG MCP (Model context protocol) server, a specialized microservice within the Multi-Agent Bot system. Its primary function is to provide tools for interacting with a vector database (ChromaDB) to perform Retrieval-Augmented Generation (RAG). This allows agents to query custom knowledge bases and retrieve relevant information to enhance their responses.

✨ Features

• Vector Database Integration: Connects with ChromaDB for storing and retrieving vectorized documents.
• Document Querying: Exposes a tool (query_docs) that allows agents to search the ChromaDB for information relevant to a given query.
• FastMCP Integration: Registers RAG functionalities as discoverable tools for the cyrene-agent (bot-api).
• Modular & Scalable: Runs as an independent microservice, allowing for easy scaling and maintenance.
• Persistent Storage: Designed to work with a Persistent Volume Claim (PVC) in Kubernetes for ChromaDB data, ensuring knowledge persists across pod restarts.

🏛️ Architecture Context

The rag-mcp interacts directly with a ChromaDB instance, which typically stores embeddings of your custom documents. The cyrene-agent discovers its query_docs tool via the fastmcp-core-server and then invokes it on the rag-mcp service when an agent needs to retrieve information from the knowledge base.

🚀 Getting Started

Prerequisites

• Python 3.12+

• ChromaDB Instance: This MCP expects to connect to a ChromaDB instance.

  • For local development, you can run ChromaDB as a Docker container or an in-memory instance (though persistence will require a file path).
  • For Kubernetes deployment, ChromaDB is typically deployed as a separate service with a PVC.

• RAG Data: You'll need documents to load into ChromaDB. The main orchestrator repository includes a rag-data-loader-job for this purpose.

Installation

Clone this repository:

git clone https://github.com/CyreneAI/rag-mcp.git
cd rag-mcp

Install Python dependencies:

pip install -r requirements.txt

Environment Variables

Create a .env file in the root of this rag-mcp directory with the following variable:

# .env in rag-mcp directory
CHROMA_DB_PATH=./chroma_data

• CHROMA_DB_PATH: The path to the ChromaDB data directory.

  • Example for local development: CHROMA_DB_PATH=./chroma_data
  • Example for Kubernetes (matching PVC mount path): CHROMA_DB_PATH=/chroma/data

Running the Application (Local Development)

1. Ensure your ChromaDB instance is running and accessible at the path specified in CHROMA_DB_PATH. You can run a simple ChromaDB server locally via Docker:

   docker run -p 8000:8000 chromadb/chroma
   # Or, for persistent data:
   # docker run -p 8000:8000 -v /path/to/your/local/chroma_data:/chroma/data chromadb/chroma

2. Load initial RAG data into your ChromaDB instance using the rag-data-loader-job (from the main orchestrator repo) or a local script.

3. Run the rag-mcp service:

   uvicorn server:app --reload --host 0.0.0.0 --port 9002

The rag-mcp server will be accessible at http://localhost:9002. It will automatically register its tools with the fastmcp-core-server if it's running and configured correctly.

🧪 Usage

Once the rag-mcp server is running and its query_docs tool is registered with fastmcp-core-server, the cyrene-agent can invoke it.

Example queries you can send to your agent (via the agent-UI or direct API chat) that would trigger the rag-mcp tool:

• “Tell me about the GAIA benchmark.”
• “What is the performance of Alita on the latest tests?”
• “Summarize the document about project X.”

Note: The quality of responses depends heavily on the data loaded into your ChromaDB.

📁 Project Structure

rag-mcp/
├── .env.example
├── .gitignore
├── README.md           # <- This file
├── Dockerfile          # Dockerfile for the rag-mcp service
├── requirements.txt    # Python dependencies for rag-mcp
└── server.py           # FastAPI application for the rag-mcp