🌤️ Skycast

A full Model Context Protocol (MCP) project containing:

• A custom MCP Weather Server (Python, FastMCP, NWS API)
• A custom MCP Client (Python, Anthropic API, Claude tool-use)
• Fully interactive tool-calling between Claude ↔ Server
• Built following Anthropic's tutorials: Build an MCP Server + Build an MCP Client

---

✨ Features

MCP Weather Server

• Provides two MCP tools:
  • get_alerts(state) — fetch active NWS alerts
  • get_forecast(latitude, longitude) — fetch 5-period weather forecast
• Uses:
  • FastMCP
  • httpx async HTTP requests
• Safe for STDIO (no print(), uses logging)

MCP Client

• Connects to any MCP server (.py or .js)
• Sends queries to Claude using the Anthropic API
• Automatically executes tool calls requested by Claude
• Maintains full interactive chat loop

---

💬 Example Usage

Query: What's the weather in Sacramento?
→ Claude → MCP Tool → Server → Forecast → Claude → You

Query: Show me alerts in TX
→ get_alerts("TX")

---

🛠️ Technologies Used

• Python 3.10+
• MCP Python SDK
• FastMCP
• Anthropic API
• httpx
• uv
• python-dotenv

---

🧩 How It Works

Claude analyzes the query
→ chooses tools
→ client executes the tool
→ server returns data
→ client sends result back
→ Claude produces final text

---

⚠️ Notes

• Do not use print() in STDIO servers
• Use logging so output goes to stderr
• Keep API keys in .env
• Claude Desktop requires absolute paths
• NWS API only supports U.S. weather data

---

📌 Example Claude Desktop Config (Windows)

{
  "mcpServers": {
    "weather": {
      "command": "uv",
      "args": [
        "--directory",
        "C:\\ABSOLUTE\\PATH\\TO\\weather",
        "run",
        "weather.py"
      ]
    }
  }
}

---

📜 License

MIT License (modify as needed)