How to Build Your First A2A Weather Agent (Fast)

⚡TL;DR

Want to build a smart agent that can be discovered and communicated with by any A2A-compliant client? This guide shows you how to create a complete Agent-to-Agent (A2A) weather system using Python, Tavily, and Gemini. You'll build an A2A-compliant weather bot server and a client that discovers it, sends weather queries, and receives polished responses. It’s the cleanest intro you’ll find to agent discovery, task handling, and structured communication with real AI output.

The Agent-to-Agent (A2A) protocol is redefining how autonomous agents collaborate. It replaces custom API spaghetti with a clean structure: agents self-discover, send tasks, and return results — all through standard JSON endpoints.

In this guide, you’ll build:

✅ A weather server agent that pulls real-time data via Tavily and formats it with Gemini

✅ A client agent that finds the server, sends a weather query, and displays a smart response

Let’s jump in.

🛠️ Project Setup (Minimal + Fast)

Let’s get your A2A weather agent project up and running in minutes — even if you’ve never built one before.

📁 Step 1: Set Up Your Project Folder

Open your terminal (Command Prompt, PowerShell, or Terminal app) and run:

bash

mkdir a2a-weather && cd a2a-weather
uv init
uv venv && source .venv/bin/activate  # (Windows: .venv\Scripts\activate)

Install your dependencies:

requirements.txt

txt

flask==3.0.0
requests==2.31.0
tavily-python>=0.3.0
python-dotenv==1.0.0
tiktoken>=0.7.0
langchain-google-genai==1.0.10

bash

uv pip install -r requirements.txt

.env

ini

TAVILY_API_KEY=your_tavily_key
GOOGLE_API_KEY=your_gemini_key

🔑 Get your keys:

🧠 Build the Weather Server Agent

File: server/weather_server.py

This agent:

Exposes /.well-known/agent.json for discovery
Accepts tasks at /tasks/send
Uses Tavily for real-time weather
Uses Gemini to summarize results

Key Endpoints:

python

@app.route("/.well-known/agent.json")
def agent_card():
    return jsonify({
        "name": "WeatherBot",
        "description": "Get real-time weather with AI summaries",
        "url": "http://127.0.0.1:5000",
        "version": "1.0",
        "capabilities": {"streaming": False, "pushNotifications": False}
    })

@app.route("/tasks/send", methods=["POST"])
def handle_task():
    task_data = request.get_json()
    query = task_data["message"]["parts"][0]["text"]

    # Tavily search
    results = tavily.search(query=f"current weather {query}", max_results=3)
    text_blobs = "\n\n".join([f"{r['title']}: {r['content']}" for r in results.get('results', [])])

    # Gemini summary
    prompt = f"""
    Provide a weather summary for {query} using:
    {text_blobs}
    Keep it under 150 words. Include temp, humidity, wind, and alerts.
    """
    summary = llm.invoke(prompt).content if text_blobs else "No data found."

    return jsonify({
        "id": task_data["id"],
        "status": {"state": "completed"},
        "messages": [
            task_data["message"],
            {"role": "agent", "parts": [{"text": summary}]}
        ]
    })

🤖 Build the Client Agent

File: client/weather_client.py

This client:

Finds the weather agent via discovery
Sends tasks in A2A format
Parses and prints the agent's response

python

client = WeatherClient()

print("Discovering agent...")
client.discover_agent()

for city in ["London", "Tokyo", "New York"]:
    print(f"\nGetting weather for {city}...")
    print(client.ask_weather(city))

🔁 A2A Workflow, Step-by-Step

Discovery
GET /.well-known/agent.json
→ Returns agent name, description, and capabilities.
Task Creation
- Client generates UUID
- Packages query in A2A format
Task Submission
POST /tasks/send
→ Server extracts message, queries Tavily, summarizes with Gemini.
Response
→ Server returns original + enhanced message
Client Output
→ Client shows the summary from the agent’s message

💻 Test It Locally

In one terminal:

bash

python server/weather_server.py

In another:

bash

python client/weather_client.py

You’ll see weather summaries like:

❝

Tokyo
26.2°C with light rain. Humidity: 89%. Winds NNW at 35.6 kph. Heat index: 30.5°C. No alerts.

❝

London
18.4°C, partly cloudy. Light winds. Humidity: 68%. Met Office warns of deep low system approaching.

🧩 Why This Matters

This simple weather app is more than a toy — it demonstrates real A2A interoperability.

✅ Clients discover agents with no config
✅ Messages follow a universal schema
✅ Any A2A agent can plug into any other

It’s Lego blocks for agents. Plug-and-play AI.

What’s Next?

In future builds:

Use Google’s Agent Development Kit (ADK)
Connect multiple A2A agents in a network
Combine A2A with Model Context Protocol (MCP)

👉 See the source code here

- Black Box Brief

Follow for weekly drops that bridge the gap between AI innovation and real-world adoption—made for founders, entrepreneurs, builders, operators, and curious outliers.

Not subscribed yet?
Subscribe