Hands-on Model Context Protocol (MCP)

This project demonstrates how to use the Model Context Protocol (MCP) to create a natural language interface to SQLite databases. It's a practical introduction to building AI systems that can interact with external tools.

What is MCP?

MCP is designed as a universal protocol that connects AI systems with data sources and tools using a standardized communication layer. As described by Anthropic:

"The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools." anthropic.com It functions somewhat like an "Internet for Agents," providing a standardized way for AI systems to access external tools without needing custom integrations for each one.

This demo shows how an LLM (Groq's Llama3-70B) can execute SQL queries through MCP, without directly accessing the database.

Getting Started

1. Make sure you have Python 3.12+ installed
2. Set your Groq API key in the .env file:

   GROQ_API_KEY=your_api_key_here
   

3. Install dependencies:

   uv venv .venv --python=3.12
   .venv\Scripts\activate
   uv pip install -e .
   

4. Run the demo:

   run_sql_demo.bat
   

   This will open two windows - an MCP server and an MCP client.

Hands-on with MCP

In the client window, try natural language queries like:

• "all users" - The LLM converts this to a SQL query using MCP's tool interface
• "users older than 30" - Watch how complex queries get properly translated
• "average age of users" - See aggregation functions at work
• "add a new user named Alice Smith with email [email protected] and age 35" - Try data manipulation

Each query demonstrates how MCP enables an LLM to use external tools (SQL queries) safely and effectively.

How MCP Works in This Project

1. MCP Server (mcp_server.py):

   • Defines tools (like query_data) with schemas
   • Handles execution of SQL queries against a database
   • Returns results in a structured format

2. MCP Client (mcp_client.py):

   • Takes natural language input from users
   • Uses Groq's LLM to generate tool calls via MCP
   • Handles the communication protocol with the server

The magic happens through MCP's standardized communication, allowing LLMs to use tools without accessing the underlying code.

Project Structure

• mcp_server.py: MCP server with SQL execution capabilities
• mcp_client.py: MCP client for LLM-driven tool calling
• database.db: Sample SQLite database with user data
• run_sql_demo.bat: Script to run both MCP components
• pyproject.toml: Python project configuration
• .env: Environment variables (API keys)