This project demonstrates a multi-agent system using Google's Agent Development Kit (ADK) that integrates Notion for information retrieval and ElevenLabs for text-to-speech conversion.

The system demonstrates a fully decoupled Agent-to-Agent (A2A) architecture.

Please use the accompanying the YouTube Crash Course video.

Note: This repository contains the starting code to follow along with the tutorial series. For the complete implementation, please check out the lesson-complete branch.

For those looking to dive deeper into ADK development, I've created an optional Builder Pack that includes:

• Complete source code for host agent and worker agents
• Cheat sheets for ADK development and A2A patterns
• .cursor/rules for consistent prompt engineering
• Comprehensive pytest test suite
• Full Lesson Plan

You can get the Builder Pack at chongdashu.gumroad.com/l/adk-builder-pack-1.

If you find this tutorial series and codebase helpful in your AI agent development journey, consider buying me a coffee! Your support helps me create more educational content on AI and agent development.

The system consists of:

• ElevenLabsAgent (A2A Service): A worker agent for text-to-speech conversion.
• NotionAgent (A2A Service): A worker agent for information retrieval from Notion.
• HostAgent (A2A Service): A coordinator agent that orchestrates the Notion and ElevenLabs agents.
• Streamlit UIs: Two separate user interfaces are provided to demonstrate different architectural patterns:
  • Embedded Runner UI: The UI runs the HostAgent in the same process.
  • A2A Client UI: The UI communicates with the HostAgent as a separate, decoupled service.

Refer to LESSON.md for a detailed architectural breakdown and design decisions.

• Python 3.13+
• Notion account and API key
• ElevenLabs account and API key
• uv package manager (recommended) or pip

adk-a2a-mcp/
├── host_agent/             # Coordinator Agent
│   ├── agent.py            # Agent definition (with delegate_task tool)
│   ├── agent_executor.py   # Handles session management (context_id -> session_id)
│   ├── prompt.py           # Orchestration instructions
│   ├── remote_connections.py # A2A client logic
│   ├── tools.py            # ADK Tools for A2A delegation
│   ├── test_client.py      # Standalone test client
│   └── __main__.py         # A2A service entry point
├── notion_agent/           # Worker Agent
│   ├── agent.py
│   ├── agent_executor.py
│   ├── prompt.py
│   ├── test_client.py
│   └── __main__.py
├── elevenlabs_agent/       # Worker Agent
│   ├── agent.py
│   ├── agent_executor.py
│   ├── prompt.py
│   ├── test_client.py
│   └── __main__.py
├── ui/                     # Streamlit UIs
│   ├── app.py              # Embedded Runner UI
│   └── a2a_app.py          # Decoupled A2A Client UI
├── scripts/
│   └── start_agents.py     # Convenience script to launch all agent services
├── tests/                  # Pytest integration tests
│   ├── test_host_agent.py
│   ├── test_notion_agent.py
│   └── test_elevenlabs_agent.py
├── logs/                   # Agent logs are stored here
├── .env.example            # Example environment variables
├── pyproject.toml          # Project config and dependencies (single source of truth)
└── README.md               # This file

1. Clone the repository:

   git clone <repository-url>
   cd adk-a2a-mcp

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies from the single source of truth, pyproject.toml:

   uv pip install -e ".[dev]"

4. Create a .env file from the example and add your API keys:

   cp .env.example .env
   # Now edit .env and add your keys

   The .env file should contain:

   NOTION_API_KEY="your_notion_api_key_here"
   ELEVENLABS_API_KEY="your_elevenlabs_api_key_here"
   ANTHROPIC_API_KEY="your_anthropic_api_key_here" # Required for some agents
   
   # A2A Service URLs (defaults)
   HOST_AGENT_A2A_URL="http://localhost:8001"
   NOTION_AGENT_A2A_URL="http://localhost:8002"
   ELEVENLABS_AGENT_A2A_URL="http://localhost:8003"

This is the recommended way to run the entire multi-agent system.

1. Start all agent services: A convenience script is provided to launch the Host, Notion, and ElevenLabs agents in the background. Their logs will be saved to the logs/ directory.

   python scripts/start_agents.py

   You can check the status of the agents with ps aux | grep _agent.

2. Run a UI Application: Choose one of the two UIs to interact with the system.

   • For the Decoupled A2A UI (Recommended for testing the full architecture):

     streamlit run ui/a2a_app.py --server.port 8080

   • For the Embedded Runner UI:

     streamlit run ui/app.py --server.port 8080

3. Stopping the services: You can stop the background agent processes with:

   pkill -f "_agent"

For development and debugging, it's crucial to test each agent independently. Each agent has a test_client.py that communicates directly with its A2A service.

1. Start the target agent's service. For example, to test the Notion agent:

   python -m notion_agent --port 8002

2. In a separate terminal, run its test client:

   python notion_agent/test_client.py

   Follow the same pattern for the other agents:

   • python -m elevenlabs_agent --port 8003 -> python elevenlabs_agent/test_client.py
   • python -m host_agent --port 8001 -> python host_agent/test_client.py

The tests/ directory contains integration tests that automate the testing process using pytest. Each test file (e.g., tests/test_notion_agent.py) is responsible for:

1. Automatically starting the required agent's A2A service in a subprocess.
2. Executing the logic from the corresponding test_client.py against the running service.
3. Asserting that the communication was successful and the agent behaved as expected.

To run all integration tests:

pytest -v

To run tests for a specific agent:

pytest -v tests/test_notion_agent.py

• Code Formatting: black . && isort .
• Type Checking: mypy .
• Testing: pytest

This project is licensed under the MIT License - see the LICENSE file for details.

This software is provided "as is", without warranty of any kind, express or implied.