SQLite Database MCP Server

Objective

The Model Context Protocol (MCP) is revolutionizing how AI Agents connect to data. In the past, if you wanted an LLM to read your database, you had to write custom glue code, custom prompts, and a custom API.

With MCP, you write a standardized "Server". Once running, any MCP-compatible client (like Claude Desktop, an OpenAI agent, or cursor) can instantly connect to it, discover its tools, and query your database without you writing a single prompt!

Related MCP server: SQLite MCP Server

Instructions

1. Open server.py.

2. Notice we are using the official mcp.server.fastmcp module, which makes building an MCP server as easy as building a FastAPI app.

3. Implement list_tables(). Use Python's built-in sqlite3 to connect to orders.db, query the sqlite_master table, and return a string containing all the table names.

4. Implement read_query(sql). Connect to the database, execute the raw SQL string passed by the LLM, fetch the results, and return them as a formatted string.

5. Notice that mcp.run() is called at the bottom. This starts the standard IO (stdio) server loop!

6. Run tests using pytest. The test suite will literally spin up an official MCP Client, connect to your server over stdio, and ask it to list its tools and run a query!

Setup

pip install -r requirements.txt

Running Tests

pytest test_server.py