FastMC Tutorial: Connect LLMs to Your Custom Tools and Data FastMCP is a high-performance Python framework designed to build Model Context Protocol (MCP) servers with minimal boilerplate. Developed by the team at Prefect, it acts as a streamlined connector to securely bridge Large Language Models (LLMs) with private enterprise databases, local files, and custom web APIs. Instead of manually writing complex JSON-RPC transport layers, developers use simple Python decorators to instantly turn standard code into discoverable LLM capabilities. 🛠️ Prerequisites & Setup
Before building, ensure your environment has Python 3.10 or higher installed. 1. Install the Core Library Install FastMCP directly via pip: pip install –upgrade fastmcp Use code with caution. 2. Prepare Your Environment Create a clean directory for your project files: mkdir mcp-tool-server && cd mcp-tool-server touch server.py Use code with caution. 🏗️ Step 1: Create the MCP Server
Initialize the main server instance within server.py. This instance will host your custom logic and automatically generate schemas for the LLM client.
from fastmcp import FastMCP # Initialize the server with a descriptive name mcp = FastMCP(“Enterprise Tool Suite 🚀”) Use code with caution. 🔧 Step 2: Connect a Custom Tool
Tools allow an LLM to take real-world actions like running calculations, modifying databases, or hitting APIs. In FastMCP, wrapping a standard Python function with the @mcp.tool decorator automatically builds input validation and auto-documents the tool for the LLM using the function’s docstring. Add a simulated inventory check tool to server.py:
@mcp.tool def check_warehouse_stock(item_id: str, location: str = “Main”) -> str: “”” Queries the live inventory database for a specific item ID. Always use this tool when users ask if an item is available. “”” # Real-world logic goes here (e.g., hitting a PostgreSQL or MongoDB database) if item_id == “SKU-99” and location == “Main”: return “Status: Available | Quantity: 42 units” return f”Status: Out of Stock at the {location} warehouse.” Use code with caution. Key Capabilities Managed Automatically:
Schema Generation: FastMCP converts python type hints (item_id: str) into a JSON Schema for the LLM.
Input Validation: Bad formats or missing parameters are blocked by Pydantic validation before running your function. 📊 Step 3: Expose Live Data Sources (Resources)
While tools take actions, Resources act as read-only context variables. Use them to feed configuration files, system logs, or customer files directly into the LLM prompt window. Add a resource wrapper to expose enterprise documentation:
@mcp.resource(“config://return-policy”) def get_return_policy() -> str: “”“Provides the standard company policy text for orders and returns.”“” return “Customers can return any unused item within 30 days for a 100% refund.” Use code with caution. 🚀 Step 4: Run and Test Locally
To finalize your script, add a standard Python entrypoint at the bottom of server.py: if name == “main”: mcp.run() Use code with caution. Run the Server Launch your server via the command line: python server.py Use code with caution.
By default, this hosts your tools over standard input/output (stdio), which is perfect for direct local orchestration with popular MCP-compatible developer tools like Claude Desktop. Tools – FastMCP