How to beginner · 3 min read

How to define tools in MCP server

Q: How to define tools in MCP server

In an MCP server, define tools by creating Python classes that implement the Tool interface with a run method. Register these tool instances with the Server to expose them to AI agents for invocation during conversations.

Quick answer

In an MCP server, define tools by creating Python classes that implement the Tool interface with a run method. Register these tool instances with the Server to expose them to AI agents for invocation during conversations.

PREREQUISITES

Python 3.8+
pip install mcp
Basic knowledge of Python classes and async programming

Setup

Install the official mcp Python package and prepare your environment variables if needed. The mcp package provides the server and tool interfaces for defining and running tools.

bash

pip install mcp

Step by step

Define a tool by subclassing mcp.server.Tool and implementing the run method. Then create an mcp.server.Server instance, register your tool, and start the server using stdio_server for standard input/output communication.

python

import asyncio
from mcp.server import Server, Tool
from mcp.server.stdio import stdio_server

class EchoTool(Tool):
    async def run(self, input: str) -> str:
        # Simply returns the input text
        return f"Echo: {input}"

async def main():
    server = Server()
    # Register the tool with a name
    server.register_tool("echo", EchoTool())

    # Start the MCP server using stdio transport
    await stdio_server(server)

if __name__ == "__main__":
    asyncio.run(main())

Common variations

You can define multiple tools by creating additional subclasses of Tool and registering them with different names. Tools can perform synchronous or asynchronous operations. The server supports other transports besides stdio, such as SSE or TCP, depending on your deployment needs.

python

class AddTool(Tool):
    async def run(self, input: str) -> str:
        # Expect input as 'num1 num2'
        parts = input.split()
        if len(parts) != 2:
            return "Error: Provide two numbers separated by space"
        try:
            num1, num2 = map(float, parts)
            return str(num1 + num2)
        except ValueError:
            return "Error: Invalid numbers"

# Register multiple tools
server = Server()
server.register_tool("echo", EchoTool())
server.register_tool("add", AddTool())

Troubleshooting

If the server does not respond, ensure you are running it with asyncio.run(main()) and that your environment supports stdio communication.
If tools are not found, verify you registered them with the correct names before starting the server.
For debugging, add logging inside your run methods to trace input and output.

✅

Key Takeaways

Define tools by subclassing Tool and implementing the async run method.
Register tools with the Server instance before starting the MCP server.
Use stdio_server for simple local testing and communication.
Multiple tools can be registered and invoked by name from AI agents.
Add error handling and logging inside tools for robust integration.

Verified 2026-04

Verify ↗