An open-source gateway to make AI agents interoperable using Google's A2A protocol – like Hugging Face, but for agents.
A2A Gateway is an open-source infrastructure that enables AI agents (Claude, GPT, local LLMs, etc.) to interoperate using the A2A protocol by Google.
Just as Hugging Face unified access to AI models, A2A Gateway aims to become the standard for collaboration among agents – public, federated, autonomous, and open.
But it is more than just a protocol implementation.
A2A Gateway is also a shared space — an open agora — for everyone interested in agentic AI interoperability to:
- 💬 Discuss standards and best practices
- 🤝 Cooperate on toolkits, adapters, and orchestrators
- 🚀 Contribute agents, demos, ideas, and feedback
- 🌱 Grow a federated ecosystem that is transparent, ethical, and developer-first
If you're building, experimenting, or just curious — you're welcome in the conversation. Join us and shape the interoperable future of AI agents.
- 🛰️ Export Claude, GPT, or custom agents as A2A endpoints
- 📄 Create and share Agent Card JSONs
- 📦 Use the CLI to manage tasks between agents
- 🌍 Publish and discover agents via a public registry
- 🧪 Monitor activities and tasks via a dashboard (beta)
A2A Gateway is built to be developer-first, lightweight, and interoperable by design.
Whether you're building your first agent, federating across services, or integrating LangChain — we've got step-by-step guides to get you there fast.
Explore the core documentation to get started:
-
🛰️ How to Create Your First A2A Agent
Build a working agent with FastAPI and expose it via A2A in 10 minutes. -
🤝 How to Federate with Other Agents
Discover, import, and route tasks across trusted agent networks. -
🔌 How to Publish a LangChain Adapter
Wrap LangChain LLMs or chains and connect them to the A2A ecosystem.
→ 📖 View full documentation index
pip install .python demo_agent.pya2a-gateway publish --file agent_card.jsona2a-gateway task --to demo-gpt --input "Translate: Hello world"agent_card.json– A2A-compliant agent carddemo_agent.py– FastAPI test agentexamples/– A2A requests viacurl, CLI and JSON
{
"id": "demo-gpt",
"name": "Demo GPT Agent",
"description": "A simple GPT agent that translates text into English",
"type": "llm",
"entrypoint": "http://localhost:5001/execute",
"capabilities": ["translation", "chat"],
"language": "python",
"version": "0.1.0",
"author": "Innoturismo"
}python demo_agent.pyThe agent responds at http://localhost:5001/execute and simulates a translation.
Perfect for testing a2a-gateway publish and a2a-gateway task.
The examples/ folder includes:
task_via_curl.sh: send an A2A task viacurltask_via_cli.sh: send a task using the CLItask.json: sample payload for testing or automation
The project includes basic tests under tests/:
test_cli.py: tests CLI (--help, commands)test_mocked.py: unit tests using mocks
To run tests:
pip install pytest
pytestTo contribute or work locally with tests, linting and builds:
pip install -r requirements-dev.txtOr use:
make install
make test
make lintTo run the demo agent locally:
make agentTo build and publish the PyPI package:
make build
make publishWe're just getting started – your contribution can make a real difference 💡
- ⭐ Star the project
- 🛠️ Submit a custom agent, guide or bugfix
- 🧠 Join the Discord community
- ✍️ Write articles or tutorials – we’ll amplify them!
If your agent or project is compatible with the A2A Gateway protocol, you can proudly show it using one of our official badges!
Add any of these snippets to your project's README.md:
[](https://github.com/therealpan/a2a-gateway)Replace the image or color to best match your brand, or ask us on Discord for a custom badge!
Have an A2A-compatible agent? Open a pull request to be featured in the registry 🛰️
Project initiated by Pan and the A2A enthusiast crew, with support from Harmonya.
With support from the AI community and inspired by Google’s A2A protocol.
Open-source, developer-first, federated design.
Licensed under MIT.