Model Context Protocol (MCP): Give Your AI Real Tools

A model that can only talk is a novelty. A model that can call tools — search, query a database, read files — is an employee. MCP is an open standard that gives models a uniform way to discover and use your tools.

The mental model

MCP is a client-server protocol. An MCP server exposes tools and resources; an MCP client (Claude, an app, your own code) lets a model call them. Think of it as USB-C for AI integrations — one standard, many devices.

A tiny MCP server

# server.py  (using the official SDK)
from mcp.server import Server
from mcp.server.stdio import stdio_server

server = Server("demo")

@server.tool()
async def add(a: float, b: float) -> str:
    "Add two numbers."
    return f"{a} + {b} = {a + b}"

@server.tool()
async def weather(city: str) -> str:
    "Return the current temperature for a city (demo)."
    return f"It's 24°C and clear in {city}."

async def main():
    async with stdio_server() as (read, write):
        await server.run(read, write)

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

What the model sees

The client reads the server's tool list — names, descriptions, and parameter types — and the model uses those to decide which tool to call and with what arguments. Your description is the model's only guide, so write it carefully.

{
  "tools": [
    { "name": "add", "description": "Add two numbers.",
      "parameters": { "a": "float", "b": "float" } },
    { "name": "weather", "description": "Current temperature for a city.",
      "parameters": { "city": "string" } }
  ]
}

Real tools to expose

• A company knowledge-base search.
• Database queries with read-only access.
• File and document operations.
• Sending emails or creating tickets (with confirmation).
• Calling internal APIs.

Treat tools like a public API. Validate all inputs and gate destructive actions behind explicit user confirmation.

MCP is still young, but it's the direction the industry is moving. Building your tools against a standard now means they'll work with any model that comes next.