🌤️ Multi-Agent Weather App

A multi-agent AI app built on Databricks Apps that fetches real-time weather using the Model Context Protocol (MCP) and Open-Meteo API

Architecture

User Query
    ↓
SupervisorAgent            — coordinates the pipeline
    ├── LocationResolverAgent   — extracts city, activity, date from query
    ├── WeatherDataAgent        — decides what to fetch; calls MCP tool layer
    │       ↓
    │   MCPWeatherClient → mcp/weather_mcp_server.py → Open-Meteo API
    ├── SafetyEvaluatorAgent    — assesses risk for the activity
    ├── ActivityAdvisorAgent    — produces go/delay/avoid recommendation
    └── ResponseComposerAgent   — writes the final natural language answer
    ↓
Gradio UI on Databricks Apps

Files

File	Purpose	Notes
app.py	Gradio UI only	Main entrypoint
agents/	Six-agent pipeline coordinated by SupervisorAgent	LLM reasoning and response composition
agents/base_agent.py	Shared Databricks Foundation Model API helper	Uses WorkspaceClient()
schemas/	Typed dataclass contracts between agents	Location, weather, safety, activity
mcp/weather_mcp_client.py	MCP stdio client used by WeatherDataAgent	Spawns local MCP server
mcp/weather_mcp_server.py	MCP server wrapping Open-Meteo API	Spawned as subprocess
app.yaml	Databricks Apps entrypoint config	Points to app.py
requirements.txt	Python dependencies	gradio, mcp, httpx, nest_asyncio, databricks-sdk
MCP_INTEGRATION_GUIDE.md	MCP architecture detail	Transport: stdio
README.md	This file	—

Supported Cities

Any city worldwide is supported via Open-Meteo Geocoding — not limited to a hardcoded list.

Example queries:

What's the weather in Mumbai?
Temperature in Berlin in celsius
Is it raining in São Paulo?

Local Setup

Step 1:

pip install -r requirements.txt

Step 2:

python app.py

Step 3: Open http://localhost:7860 in your browser.

No API key or environment variable needed — Open-Meteo is free and open.

Deploying to Databricks Apps

1. Upload all files to a Databricks Workspace folder or Repo
2. Go to Compute → Apps → Create App
3. Point the source to the folder containing app.yaml
4. Click Deploy — Databricks reads requirements.txt and installs dependencies automatically
5. Once status shows Running, click the assigned URL

Requires Databricks Premium plan or above with Apps enabled.

MCP Transport

The MCP server (mcp/weather_mcp_server.py) is spawned as a local subprocess by mcp/weather_mcp_client.py using stdio transport. Both processes run inside the same Databricks Apps container, so no separate server deployment is needed. Do not change the transport to SSE or HTTP.

Data Source

• Geocoding: https://geocoding-api.open-meteo.com/v1/search — converts city name to lat/lon
• Forecast: https://api.open-meteo.com/v1/forecast — returns current temperature, wind speed, weather code, and humidity

Both endpoints are free, require no API key, and have no rate limit for reasonable usage.