Skip to content

guide-weather-mcp.md

Latest commit

 

History

History
150 lines (113 loc) · 4.23 KB

guide-weather-mcp.md

File metadata and controls

150 lines (113 loc) · 4.23 KB

Guide: Building a Weather MCP Server

This guide builds upon the Calculator example. Now, we will give Claude the power to reach out to the internet and fetch real-world data.

We will build a Weather Server that fetches the forecast from Open-Meteo (a free weather API).

Prerequisites

Step 1: Set Up Your Project

  1. Create a new directory:

    mkdir weather-mcp
cd weather-mcp

  2. Initialize the project:

    uv init

  3. Add dependencies: We need mcp (for the server) and requests (to make simple web requests).

    uv add "mcp[cli]" requests

Step 2: Write the Server Code

Create a file named weather.py.

Notice how this code looks just like a normal Python function. No async or await is needed here!

# weather.py
from mcp.server.fastmcp import FastMCP
import requests

# 1. Initialize the Server
mcp = FastMCP("Weather")

@mcp.tool()
def get_weather(latitude: float, longitude: float) -> str:
    """
    Get the current weather for a specific location.
    
    Args:
        latitude: The latitude of the location.
        longitude: The longitude of the location.
    """
    # Using Open-Meteo: A free weather API that doesn't need a key
    url = f"https://api.open-meteo.com/v1/forecast?latitude={latitude}&longitude={longitude}&current_weather=true"
    
    # Make the request and get the data
    response = requests.get(url)
    data = response.json()
    
    # Extract the current temperature
    temp = data["current_weather"]["temperature"]
    return f"The current temperature at {latitude}, {longitude} is {temp}°C."

@mcp.tool()
def get_coordinates(city_name: str) -> str:
    """
    Find the latitude and longitude for a city name.
    
    Args:
        city_name: The name of the city (e.g., "Paris", "San Francisco").
    """
    # Use Open-Meteo's Geocoding API to find the location
    url = f"https://geocoding-api.open-meteo.com/v1/search?name={city_name}&count=1&language=en&format=json"
    
    response = requests.get(url)
    data = response.json()
    
    if "results" in data:
        result = data["results"][0]
        lat = result["latitude"]
        lon = result["longitude"]
        return f"{lat},{lon}"
    else:
        return "City not found."

if __name__ == "__main__":
    mcp.run(transport="stdio")

Step 3: Configure Claude Desktop

  1. Open your config:

    code ~/Library/Application\ Support/Claude/claude_desktop_config.json

  2. Add the "weather" server: Add a new entry to your mcpServers list. (Replace /Users/YOUR_NAME/... with your actual path)

    {
  "mcpServers": {
    "my_calculator": { ... }, 
    "weather": {
      "command": "uv",
      "args": [
        "--directory",
        "/Users/YOUR_NAME/path/to/weather-mcp",
        "run",
        "weather.py"
      ]
    }
  }
}

  3. Restart Claude.

Step 4: Test It Out!

  1. Check the 🔌 icon. You should see both Calculator and Weather.

  2. Ask Claude:

    "What is the weather in San Francisco?"

  3. Watch the Magic (Tool Chaining):

    • Claude sees it has a get_weather tool, but that tool needs coordinates.
    • It sees it has a get_coordinates tool that takes a city name.
    • Step 1: Claude calls get_coordinates("San Francisco") -> gets "37.77,-122.41".
    • Step 2: Claude takes that result and calls get_weather(37.77, -122.41).
    • Step 3: It tells you the answer.

    You didn't have to program this logic. Claude figured out the workflow on its own!

Step 5: Key Concepts Learned

  1. Tool Chaining: You saw how Claude automatically used one tool (get_coordinates) to get the data it needed for another tool (get_weather).
  2. External APIs: This MCP server acts as a bridge. Claude asks the server, the server calls Open-Meteo, and the answer flows back.
  3. Multiple Servers: You can have as many MCP servers running as you want (Calculator + Weather + Files, etc.).