🌤️ — Weather Forecast in Israel

A learning project in MCP (Model Context Protocol) development: an LLM-based Agent that can fetch weather forecasts for cities in Israel and the US, and answer in free language in a terminal chat.

🎯 Project Goal

Build a dedicated MCP server that "teaches" the LLM to perform a task it can't do on its own — fetching a weather forecast from an Israeli site that has no API.

The agent does this in two ways:

1. Israel server (weather-Israel) — automates a real browser with Playwright: opens weather2day.co.il/forecast, searches for a city, selects it, then extracts the forecast page content and feeds it to the LLM so it can answer on its own (RAG approach).

2. US server (weather-USA) — fetches forecast and alerts via the official NWS (National Weather Service) API.

The Host connects all servers to the LLM (Gemini, via the OpenAI-compatible API), automatically collects all their Tools, and runs a chat loop in which the LLM chooses which tool to use and activates them one by one until an answer is obtained.

Related MCP server: mcp-playwright-weather-israel

🏗️ Project Structure

project-template/
├── host.py            # צ'אט בטרמינל שמחבר את ה-LLM לכל שרתי ה-MCP
├── client.py          # MCP Client גנרי (מתחבר לשרת, מריץ Tools)
├── weather_USA.py     # שרת MCP לארה"ב — דרך NWS API
├── weather_Israel.py  # שרת MCP לישראל — דרך אטמוט דפדפן (Playwright)
├── pyproject.toml     # תלויות הפרויקט
└── README.md

Tools of the Israel server

💡 The first three share the same browser via a global state (_page), because they all run in the same server process. The fourth iterates over all frames on the page, because the forecast itself loads inside an <iframe> that the main body doesn't see.

⚙️ Installation and Running

Prerequisites

• Python ≥ 3.13

• uv (environment and dependency manager)

• Gemini API key (get a free key)

Steps

# 1. התקנת התלויות והקמת סביבת העבודה
uv sync

# 2. התקנת דפדפן Chromium עבור Playwright
uv run playwright install chromium

Setting environment variables

Create a .env file in the project folder:

GEMINI_API_KEY=your_api_key_here
GEMINI_MODEL=gemini-2.0-flash

Running

uv run host.py

After launch, a Query: prompt appears — type a question (or quit to exit).

💬 Example questions the agent can answer

Forecast in Israel (Playwright)

A browser window opens showing the forecast page, and the LLM also answers in chat.

• "What's the forecast in Tel Aviv?"

• "How many degrees are expected tomorrow in Jerusalem?"

• "How's the weather in Haifa today?"

• "Open the forecast for Be'er Sheva for me"

Forecast and alerts in the US (NWS API)

• "What's the weather forecast for San Francisco?" (latitude/longitude)

• "Are there any weather alerts in California?"

• "Show me active weather alerts for NY"

🧩 How it works (flow diagram)

משתמש → "מה התחזית בתל אביב?"
   │
   ▼
 Host  ──►  LLM (Gemini) בוחר Tools להפעיל
   │
   ▼  (שרת ישראל, בזה אחר זה)
 open → enter("תל אביב") → select → get_content
   │
   ▼
 תוכן התחזית חוזר ל-LLM (RAG)
   │
   ▼
 LLM כותב תשובה בשפה חופשית בצ'אט ✨

📝 Notes

• The LLM is the one that decides which Tools to activate and in what order — we didn't write hard-coded logic.

• Two server files = two separate processes, so any shared state (like the open browser) lives in one server only.