add server for a2a

[atharvacoolkni]

This pull request adds support for running an A2A-protocol-compliant agent server in the CLI, allowing users to expose their agentflow graphs as A2A-compatible services. It introduces a new CLI command (a2a), provides configuration and customization options via both CLI flags and agentflow.json, and includes a working example for a currency conversion agent with a custom executor. The changes also add an optional dependency group for A2A support.

A2A Server Feature Integration:

• Added a new A2ACommand class in a2a.py to implement the A2A server command, including configuration loading, environment setup, agent card creation, executor resolution, and server startup using Uvicorn. The command supports overrides via CLI flags and integration with agentflow-specific executors and skills.
• Registered the new a2a command in the CLI (main.py), with options for config, host, port, agent name, description, streaming, and logging verbosity. [1] [2]
• Added A2A-related default constants (DEFAULT_A2A_HOST, DEFAULT_A2A_PORT, etc.) to constants.py for consistent configuration.

Example: Currency Agent with Custom Executor:

• Added a complete example in examples/currency_agent:
  • agentflow.json demonstrates A2A configuration, including a custom executor and skills.
  • executor.py provides a custom executor that emits INPUT_REQUIRED when the LLM requests missing info.
  • graph.py defines the agent's graph, including a tool node for currency conversion using the Frankfurter API and an LLM node for conversational logic.

Dependency Management:

• Added an a2a optional dependency group in pyproject.toml to install required packages for A2A server support.