Israel Weather Forecast MCP Server

A standalone MCP server that fetches an Israeli weather forecast by driving a real browser (Playwright), rather than by calling a conventional API. The idea: the model (the Host, for example, Claude Desktop / Claude Code) calls the tools one after another, exactly like a human user browsing the site: opens the page, types a city name, selects it from the autocomplete list, and finally extracts the page content to answer the user's question.

Project Structure

israel-weather-mcp/
├── server.py           # השרת עצמו - כל ה-tools מוגדרים כאן
├── requirements.txt
└── README.md

Related MCP server: Israel Weather MCP Agent

Implemented Tools

Stage A - Navigation and City Search

Stage B - Content Extraction (RAG)

The typical flow (the call order expected by the LLM):

open_weather_forecast_israel
        ↓
enter_weather_forecast_city_israel("תל אביב")
        ↓
select_weather_forecast_city_israel("תל אביב")
        ↓
get_weather_forecast_content()   → הטקסט חוזר למודל, שמנסח תשובה למשתמש

Installation

cd israel-weather-mcp
python -m venv .venv
source .venv/bin/activate        # ב-Windows: .venv\Scripts\activate
pip install -r requirements.txt
playwright install chromium      # מוריד את הדפדפן שה-Playwright מפעיל

Connecting to a Host (for example, Claude Desktop)

In the MCP configuration file (for example, claude_desktop_config.json), add:

{
  "mcpServers": {
    "israel-weather-forecast": {
      "command": "python",
      "args": ["/נתיב-מלא/israel-weather-mcp/server.py"]
    }
  }
}

After saving, restart the Host. The tools will appear in the list of available tools, and you'll be able to ask, for example: "What's today's forecast in Haifa?" and the Host will run the tools one after another to answer.

Important Note About the Selectors

The exact HTML structure of the ims.gov.il site (the class/id names of the search box, the autocomplete items and the forecast area) can change with site updates, so all the selectors are centralized in the SELECTORS dictionary at the top of server.py, with several alternatives (fallback) for each element. It is recommended:

1. Run with HEADLESS = False in server.py (a line near the general settings) to see the browser in action.

2. Open the forecast site in a regular browser, right-click → "Inspect" on the search box / autocomplete list / forecast area, and update the lists in the SELECTORS dictionary accordingly.

This way you can adapt the server even if you decide to point it at a different Israeli forecast site (for example, a private company's site) instead of ims.gov.il - you just need to update IMS_FORECAST_URL and SELECTORS.

Uploading to GitHub

The project is already set up as a local git repo (it has an initial commit). To upload it to GitHub and get a link to send to your lecturer/teacher:

1. Go to github.com and click New repository (without checking "Add README" / ".gitignore" / "license" - we already have those).

2. Give it a name, for example israel-weather-mcp, and click Create repository.

3. In the terminal, inside the project folder (israel-weather-mcp/), run:

   git remote add origin https://github.com/<שם-המשתמש-שלכם>/israel-weather-mcp.git
   git branch -M main
   git push -u origin main

4. After the push, the link you'll send to your lecturer is the repository URL: https://github.com/<שם-המשתמש-שלכם>/israel-weather-mcp

If you haven't set up git for the first time on your computer yet, you may need to run once:

git config --global user.name "השם שלכם"
git config --global user.email "האימייל שלכם"

Direct Run for Testing (Without a Host)

You can also run a quick test with the MCP CLI:

mcp dev server.py

This opens a local test interface (MCP Inspector) where you can call the tools directly and see the output, without needing a connection to Claude Desktop.