MCP for Developers: Build Your First MCP Server from Scratch

If you already understand MCP concepts (if not, read the "What Is MCP" article first), this article takes you to the next step: building your own MCP Server to connect your internal tools or custom systems to Claude.

When Building Your Own Server Makes Sense

The official and community MCP Server directory already has over 1,000 ready-made Servers covering most major tools. You'd typically build your own when: you have internal custom tools with no existing Server; existing Servers don't meet your requirements (specific auth methods, specific data formats); you want to expose your service to Claude.

MCP Server Basic Architecture

An MCP Server implements the MCP protocol, doing three things:

1. Declare Tools: tell Claude "what I can do"
2. Receive Requests: receive Claude's tool calls
3. Execute and Return Results: perform the actual operation, return results to Claude

Build a Minimal Working MCP Server in Python

Here's a minimal weather-querying MCP Server using Anthropic's official Python SDK:

from mcp.server import Server
from mcp.server.stdio import stdio_server
from mcp import types
import httpx

app = Server("weather-server")

@app.list_tools()
async def list_tools() -> list[types.Tool]:
    return [
        types.Tool(
            name="get_weather",
            description="Get current weather for a city",
            inputSchema={
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "The city name"}
                },
                "required": ["city"]
            }
        )
    ]

@app.call_tool()
async def call_tool(name: str, arguments: dict) -> list[types.TextContent]:
    if name == "get_weather":
        city = arguments["city"]
        async with httpx.AsyncClient() as client:
            geo = await client.get(
                f"https://geocoding-api.open-meteo.com/v1/search?name={city}&count=1"
            )
            results = geo.json().get("results", [])
            if not results:
                return [types.TextContent(type="text", text=f"City not found: {city}")]
            lat, lon = results[0]["latitude"], results[0]["longitude"]
            weather = await client.get(
                f"https://api.open-meteo.com/v1/forecast?"
                f"latitude={lat}&longitude={lon}&current=temperature_2m"
            )
            temp = weather.json()["current"]["temperature_2m"]
            return [types.TextContent(type="text", text=f"{city}: {temp}°C")]
    return [types.TextContent(type="text", text="Unknown tool")]

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

Install dependencies: pip install mcp httpx

Add to Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "weather": {
      "command": "python",
      "args": ["/path/to/weather_server.py"]
    }
  }
}

Restart Claude Desktop and ask: "What's the Temperature in Taipei?" — Claude will call your Server.

Key Design Principles

Minimum privilege: only expose operations that are actually needed. Read-only where possible.

Clear tool descriptions: the description field is critical — Claude decides when to call the tool based on this. Clearer description = higher calling accuracy.

Error handling: provide explicit error responses for all failure modes, not empty results that leave Claude guessing.

Consider reversibility: for state-changing operations (delete, update, send), consider adding confirmation mechanisms or operation logs at the Server layer.