Good practice for A2A client definition?

[Waymond-ai]

Questions Regarding a2a-python Integration

Context

After using and reading through the a2a-python-SDK, I have the following questions.

For context, I am using LlamaIndex FunctionAgent as the core of the orchestration agent. This agent is responsible for rerouting context to other specialized agents in a supervisor architecture using the A2A protocol.

My initial plan for the system flow is as follows:

• Upon application startup, clients for the sub-agent endpoints (Agent 1 and Agent 2) will be initialized.
• These clients will then be defined as callable functions and passed as tools to the orchestration agent.

File Structure

├── agent_2
│   ├── README.md
│   ├── __pycache__
│   │   └── mcp.cpython-313.pyc
│   ├── admin.py
│   ├── apps.py
│   ├── mcp.py
│   ├── migrations/
│   ├── models.py
│   ├── tests.py
│   └── views.py
├── agent_1
│   ├── admin.py
│   ├── apps.py
│   ├── migrations/
│   ├── models.py
│   ├── tests.py
│   └── views.py
├── agent_orchestrator
│   ├── admin.py
│   ├── agent/
│   │   └── agent_orchestrator.py
│   ├── apps.py
│   ├── migrations/
│   ├── models.py
│   ├── tests.py
│   └── views.py

The A2A servers will be implemented in Agent 1 and Agent 2.
The A2A clients will be defined in a:

shared_folder/

These client definitions will be initialized at app startup and used by the orchestration agent as tools.

---

Questions

1. Is this the correct way to handle A2A connections?
2. If not, what are some good practices to keep the codebase low in dependencies and maintainable?