mcp-weather

How the LLM Uses These Tools

Example 1: Weather Alerts

You: "Are there any weather alerts in California?"

Claude's process:

1. Recognizes it needs weather alert info
2. Calls get_alerts(state="CA")
3. Gets the formatted response
4. Presents it to you in natural language

Example 2: Weather Forecast

You: "What's the weather forecast for San Francisco?"

Claude's process:

1. Knows SF coordinates (or looks them up)
2. Calls get_forecast(latitude=37.7749, longitude=-122.4194)
3. Gets 5-period forecast
4. Summarizes it for you

Run the mcp server

uv run weather.py

Update the claude config file claude_desktop_config.json to below content

{
"mcpServers": {
"weather": {
"command": "/Users/santhosh.sharma/.local/bin/uv",
"args": [
"--directory",
"/Users/santhosh.sharma/Repositories/mcp-weather",
"run",
"weather.py"
]
}
}
}

Reference : https://modelcontextprotocol.io/docs/develop/build-server#python

Analyze logs in ~/Library/Logs/Claude/mcp.log

When you ask Claude (with this MCP server connected):

Docstring best practises:

1. First line = One-line summary (imperative mood: "Get", "Format", "Calculate")
2. Use present tense ("Returns the sum" not "Will return")
3. Be specific about parameter types and expected values
4. Include examples for complex functions
5. Keep it updated when code changes