MCP with RAG Demo

What is MCP with RAG Demo?

MCP with RAG Demo is an example implementation of a Model Context Protocol (MCP) server integrated with Retrieval-Augmented Generation (RAG) capabilities. It allows AI models to interact with a knowledge base, search for information, and add new documents. The project is intended for developers learning how to build MCP servers with RAG.

How to use MCP with RAG Demo?

Install dependencies with pip install -r requirements.txt, then start the MCP server in either SSE mode (python server.py --sse) or stdio mode (python server.py --stdio). Run the client example with python client_example.py --sse http://localhost:8000 for SSE or python client_example.py --stdio --command python --args "server.py --stdio" for stdio. An OpenAI integration example is also available via python openai_example.py after setting the OPENAI_API_KEY environment variable.

Key features of MCP with RAG Demo

• MCP server with tool and resource support
• RAG implementation with in-memory storage fallback
• Client example for interacting with the MCP server
• Support for both SSE (HTTP) and stdio communication modes
• Simple prompt templates for greetings

Use cases of MCP with RAG Demo

• Demonstrating how to build an MCP server with RAG capabilities
• Adding documents to a knowledge base and searching them with AI
• Testing MCP communication over SSE or stdio
• Integrating OpenAI’s API with an MCP server

FAQ from MCP with RAG Demo

What tools are available in MCP with RAG Demo?

The demo provides five tools: echo, add, add_document, rag_search, and list_documents. Each tool has a specific purpose, such as echoing messages or adding documents to the knowledge base.

What communication modes does the server support?

The server supports SSE (HTTP) mode using Server-Sent Events and stdio mode for standard input/output communication. Default SSE host is 0.0.0.0:8000, customizable with --host and --port.

What are the runtime requirements?

Python 3.8+ and pip are required. Dependencies are listed in requirements.txt and include the mcp-sdk package. An OpenAI API key is needed for the OpenAI example.

Where does the knowledge base data live?

The knowledge base uses in-memory storage (no persistent database). Documents are held only while the server is running. Sample data is available from the sample://data resource.

How do I troubleshoot connection issues?

Ensure the server is running in another terminal, check the URL (default http://localhost:8000), and verify the server is listening on the expected interface. Use --debug for verbose output. The client automatically appends /sse if missing.