⛅ WeatherWise — AI Agent with MCP + Google ADK

An AI agent that uses Model Context Protocol (MCP) to connect to a live weather data source, retrieves structured meteorological data, and uses Google ADK (Agent Development Kit) + Gemini 2.0 Flash to generate human-friendly weather reports.

---

🏗️ Architecture

User Request
    │
    ▼
FastAPI Server (main.py)
    │
    ▼
Google ADK LlmAgent  ←──── Gemini 2.0 Flash (LLM)
    │
    │  MCP stdio transport
    ▼
Weather MCP Server (mcp_server/weather_mcp_server.py)
    │
    │  HTTP (httpx)
    ▼
Open-Meteo API  ──→  Geocoding API
                ──→  Forecast API (current + 5-day)

Components

Component	Role
Google ADK	Orchestrates the agent loop, manages tool calls
MCP Server	Exposes get_weather tool via stdio transport
Open-Meteo	Free, no-API-key weather data (global coverage)
FastAPI	REST API + web UI server
Cloud Run	Serverless Google Cloud deployment

---

📁 Project Structure

weather-mcp-agent/
├── main.py                        # FastAPI server + API endpoints
├── agent/
│   └── weather_agent.py           # Google ADK agent definition
├── mcp_server/
│   └── weather_mcp_server.py      # MCP server exposing get_weather tool
├── static/
│   └── index.html                 # Frontend web UI
├── requirements.txt
├── Dockerfile
├── deploy.sh                      # One-command Cloud Run deploy
├── .env.example                   # Environment variable template
└── .gitignore

---

🖥️ Running Locally in VS Code

Prerequisites

• Python 3.11 or 3.12
• VS Code with Python extension
• A Google API Key (for Gemini / ADK)

Step 1 — Clone / open the project

Open the weather-mcp-agent/ folder in VS Code:

File → Open Folder → weather-mcp-agent/

Step 2 — Create a virtual environment

Open the VS Code terminal (Ctrl+`` ``) and run:

python -m venv .venv

Activate it:

• Windows: .venv\Scripts\activate
• Mac/Linux: source .venv/bin/activate

VS Code should auto-detect the venv. If prompted, click "Yes" to use it as the workspace interpreter.

Step 3 — Install dependencies

pip install -r requirements.txt

Step 4 — Configure environment

cp .env.example .env

Edit .env and set your key:

GOOGLE_API_KEY=AIza...your-key-here

Get a key at: https://aistudio.google.com/app/apikey (free tier works)

Step 5 — Run the server

python main.py

Or use the VS Code Run button if you have a launch config. The server starts on http://localhost:8080

Step 6 — Open the UI

Visit http://localhost:8080 in your browser — you'll see the WeatherWise UI.

Or test via curl:

curl -X POST http://localhost:8080/api/weather \
  -H "Content-Type: application/json" \
  -d '{"city": "Bengaluru"}'

VS Code launch.json (optional)

Create .vscode/launch.json:

{
  "version": "0.2.0",
  "configurations": [
    {
      "name": "Run WeatherWise",
      "type": "python",
      "request": "launch",
      "program": "${workspaceFolder}/main.py",
      "envFile": "${workspaceFolder}/.env",
      "console": "integratedTerminal"
    }
  ]
}

---

🧪 Testing the MCP Server Directly

Run the agent from the CLI (bypasses FastAPI):

python agent/weather_agent.py "Tokyo"
python agent/weather_agent.py "London"

Test raw weather data (no LLM required):

curl http://localhost:8080/api/raw-weather/Mumbai

---

☁️ Deploying to Google Cloud Run

Prerequisites

• Google Cloud account with a project
• gcloud CLI installed: https://cloud.google.com/sdk/docs/install
• Docker installed (for local builds) or Cloud Build enabled

Step 1 — Set environment variables

export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_API_KEY=your-google-api-key
export GOOGLE_CLOUD_REGION=us-central1   # or asia-south1 for India

Step 2 — Authenticate

gcloud auth login
gcloud config set project $GOOGLE_CLOUD_PROJECT

Step 3 — Deploy (one command)

chmod +x deploy.sh
./deploy.sh

This script automatically:

1. Enables required Google Cloud APIs
2. Builds the Docker image via Cloud Build
3. Pushes to Google Container Registry
4. Deploys to Cloud Run with your API key
5. Prints the live Cloud Run URL

Step 4 — Test your deployment

# Replace with your actual Cloud Run URL
curl -X POST https://weatherwise-agent-xxxx-uc.a.run.app/api/weather \
  -H "Content-Type: application/json" \
  -d '{"city": "Bengaluru"}'

Manual Cloud Run deploy (alternative)

# Build image
gcloud builds submit --tag gcr.io/$GOOGLE_CLOUD_PROJECT/weatherwise-agent .

# Deploy
gcloud run deploy weatherwise-agent \
  --image gcr.io/$GOOGLE_CLOUD_PROJECT/weatherwise-agent \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --memory 512Mi \
  --set-env-vars "GOOGLE_API_KEY=$GOOGLE_API_KEY"

---

🔌 MCP Tool Reference

The MCP server exposes one tool:

get_weather

Parameter	Type	Required	Description
city	string	✅	City name (e.g. "Bengaluru", "Tokyo")

Returns:

{
  "location": { "city": "Bengaluru", "country": "India", "latitude": 12.97, "longitude": 77.59 },
  "current": {
    "temperature_c": 24.5,
    "feels_like_c": 25.1,
    "humidity_pct": 68,
    "wind_speed_kmh": 12.4,
    "wind_direction_deg": 270,
    "precipitation_mm": 0.0,
    "pressure_hpa": 1013.2,
    "uv_index": 6,
    "condition": "Partly cloudy"
  },
  "forecast_5_days": [...]
}

---

📡 API Endpoints

Method	Endpoint	Description
GET	/	Web UI
GET	/health	Health check
POST	/api/weather	AI-generated weather report
GET	/api/raw-weather/{city}	Raw weather data (no LLM)

---

🛠️ How MCP Integration Works

1. Agent startup: MCPToolset launches the MCP server as a subprocess via stdio
2. Tool discovery: ADK queries the MCP server for available tools (list_tools)
3. LLM decides: Gemini sees the tool schema and decides to call get_weather
4. Tool call: ADK sends the call via MCP protocol → server fetches from Open-Meteo
5. Response: Structured JSON weather data flows back to the agent
6. Report: Gemini uses the data to craft a detailed, friendly weather report

---

📝 Submission Checklist

• Implemented using Google ADK
• Uses MCP to connect to one tool (Weather via Open-Meteo)
• Retrieves structured data (current conditions + 5-day forecast)
• Uses retrieved data to generate the final response
• Deployable to Google Cloud Run
• Runnable locally in VS Code

---

🔑 Getting a Free Google API Key

1. Go to https://aistudio.google.com/app/apikey
2. Click "Create API Key"
3. Copy the key and paste it into .env

The free tier includes generous Gemini 2.0 Flash usage.

---

📄 License

MIT — feel free to use and extend.