Install uv

If you do not already have it:

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
 
# Windows (PowerShell)
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

uv is Astral's fast Python package manager. It replaces pip + venv + pip-tools in one binary. Standard pip works fine too — every command below has a pip install equivalent.

Create the Project

# Create the project directory
mkdir agent-team
cd agent-team
 
# Initialize a uv project
uv init --python 3.12
 
# Add ADK with the A2A extra
uv add 'google-adk[a2a]' python-dotenv

The [a2a] extra pulls in the A2A SDK, uvicorn for the HTTP server, and the agent card publishing utilities. Without it, you only get the in-process agent runtime.

Get Your Gemini API Key

  1. Go to Google AI Studio
  2. Click Create API Key
  3. Copy the key

Create a .env file in your project root:

# .env
GOOGLE_API_KEY=your-gemini-api-key-here

Heads up: Gemini's free tier is generous for development. gemini-flash-latest is the model we will use across all three agents — it is fast and cheap enough that a multi-agent conversation still costs cents.

Verify ADK Is Installed

uv run adk --version

You should see 1.x.x. If you get "command not found", make sure you ran uv add and not just uv pip install.

Verify A2A Is Wired In

A quick import check that confirms the A2A extras are present:

uv run python -c "from google.adk.a2a.utils.agent_to_a2a import to_a2a; from google.adk.agents.remote_a2a_agent import RemoteA2aAgent; print('A2A ready')"

If you see A2A ready, you are good. If you get ModuleNotFoundError, your install missed the [a2a] extra — rerun uv add 'google-adk[a2a]'.

Create the Agent Packages

Each of the three agents lives in its own Python package — that is what ADK's CLI and the A2A wrappers expect:

# Three packages, one per agent
mkdir researcher writer orchestrator
 
# ADK package marker files
touch researcher/__init__.py researcher/agent.py
touch writer/__init__.py writer/agent.py
touch orchestrator/__init__.py orchestrator/agent.py

Each __init__.py should re-export the agent module so ADK can discover it:

# researcher/__init__.py
# writer/__init__.py
# orchestrator/__init__.py
# (Same one-liner in all three)
 
from . import agent

Verify Your API Key

# test_setup.py
# ==========================================
# Smoke test — make sure your key reaches Gemini
# ==========================================
 
from dotenv import load_dotenv
import google.genai as genai
 
load_dotenv()
 
client = genai.Client()
response = client.models.generate_content(
    model="gemini-flash-latest",
    contents="Say hello in exactly five words.",
)
 
print(response.text)
print("\nSetup verified.")

Run it:

uv run python test_setup.py

A five-word greeting means everything is wired correctly. An authentication error means double-check .env.

Your Project So Far

agent-team/
├── researcher/
│   ├── __init__.py       ✅
│   └── agent.py          📝 (next step)
├── writer/
│   ├── __init__.py       ✅
│   └── agent.py          📝
├── orchestrator/
│   ├── __init__.py       ✅
│   └── agent.py          📝
├── .env                  ✅
├── pyproject.toml        ✅ (managed by uv)
└── test_setup.py         ✅

Three empty agent packages, ready to be filled in. We do the Researcher next.

Reference: uv install docs · ADK Python quickstart · Google AI Studio