MCP Knowledge Base

A simple MCP client-server

Requirements

• Python 3.9 or higher
• Poetry for dependency management
• OpenAI API key

Setup

1. Install dependencies using Poetry:

poetry install

2Create a .env file in the project root or parent directory with your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

Project Structure

• server.py: MCP server implementation with tools
• client-sse.py: MCP client implementation with LLM capabilities
• data/kb.json: Knowledge base data with MCP-related Q&A
• pyproject.toml: Poetry configuration file

Running the Application

1. Start the server:

poetry run python server.py

2. In a separate terminal, run the client:

poetry run python client-sse.py

Using the Client

The client has two modes:

1. Direct tool calls:

• Uncomment the asyncio.run(test_direct_tool_calls()) line in client-sse.py
• This directly calls the tools without using an LLM

2. LLM-powered interactions (default):

• Uses OpenAI to interpret queries and call appropriate tools
• Ask questions like "What is MCP?" or "What is the difference between stdio and SSE transports?"

Customizing

• Add new tools to server.py by creating additional functions with the @mcp.tool() decorator
• Modify the knowledge base by updating data/kb.json
• Change the OpenAI model by modifying the model parameter in the MCPClient class