a2a-openai-agents

A robust Python library for integrating OpenAI Agents with the A2A (Agent2Agent) Protocol, enabling seamless agent-to-agent communication and multi-agent systems.

Quick Start

Installation

pip install a2a-openai-agents

Try the Examples

# Start a weather agent
a2a-agents weather

# Start a math agent with custom skills  
a2a-agents math --port 8001

# Run multi-agent system demo
a2a-agents multi-agent

Key Features

• Drop-in Integration: Wrap existing agents.Agent instances without modification
• Auto-Skill Derivation: Agent tools automatically become A2A skills
• Multi-Agent Systems: Agents can discover and communicate with each other
• Simple & Powerful: Minimal setup for maximum functionality
• Customizable: Override auto-generated skills with custom implementations

Usage

Basic Agent

from agents import Agent, function_tool
from a2a_openai_agents import A2AWrapper, run_a2a_wrapper_server
import random

@function_tool
def get_weather(city: str) -> str:
    """Get the weather for a given city."""
    return f"The weather in {city} is {random.choice(['sunny', 'cloudy', 'rainy'])}."

# Create your OpenAI Agent
agent = Agent(
    name="WeatherBot",
    instructions="You're a helpful weather assistant.",
    model="gpt-4o-mini", 
    tools=[get_weather]
)

# Wrap with A2A capabilities  
service = A2AWrapper(
    openai_agent=agent,
    a2a_description="Weather information service"
)

# Start the A2A server
run_a2a_wrapper_server(service, port=8000)

Custom Skills

from a2a_openai_agents import A2ASkillConfig

async def calculate_sum(wrapper, params):
    numbers = params.get("numbers", [])
    return {"result": sum(numbers), "count": len(numbers)}

# Override auto-generated skills
service.a2a_skills = [
    A2ASkillConfig(
        name="calculate_sum",
        description="Add a list of numbers",
        handler=calculate_sum,
        parameters={
            "type": "object",
            "properties": {
                "numbers": {"type": "array", "items": {"type": "number"}}
            },
            "required": ["numbers"]
        }
    )
]

Multi-Agent Communication

import httpx

async def call_other_agent(agent_url: str, message: str):
    """Call another A2A agent."""
    async with httpx.AsyncClient() as client:
        response = await client.post(
            f"{agent_url}/message",
            json={"text": message}
        )
        return response.json()

# Coordinator agent that uses other agents
async def coordinate_task(wrapper, params):
    # Call weather agent
    weather = await call_other_agent(
        "http://localhost:8000", 
        f"What's the weather in {params['city']}?"
    )
    
    # Call math agent  
    calculation = await call_other_agent(
        "http://localhost:8001",
        f"Calculate the sum of {params['numbers']}"
    )
    
    return {
        "weather": weather,
        "calculation": calculation,
        "status": "completed"
    }

Examples

Built-in Examples

The library includes several ready-to-run examples:

Command	Description	Port
a2a-agents weather	Weather information agent	8000
a2a-agents math	Mathematical calculations	8001
a2a-agents multi-agent	Multi-agent coordination demo	9000-9001

Example Files

Check the examples/ directory:

• simple_weather_agent.py - Basic weather service with tool integration
• custom_skills_agent.py - Custom A2A skills without tools
• working_multi_agent.py - Multi-agent system with coordination
• multi_agent_research_team.py - Advanced 3-agent research pipeline

A2A Protocol Integration

Agent Discovery

Each agent exposes an agent card at /.well-known/agent.json:

curl http://localhost:8000/.well-known/agent.json

{
  "name": "WeatherBot",
  "description": "Weather information service", 
  "skills": [
    {
      "id": "get_weather",
      "name": "get_weather",
      "description": "Get the weather for a given city"
    }
  ],
  "url": "http://localhost:8000"
}

Message Format

Agents communicate using the A2A protocol:

{
  "id": "unique-request-id",
  "method": "message/send",
  "params": {
    "message": {
      "messageId": "unique-message-id",
      "role": "user", 
      "parts": [
        {
          "type": "text",
          "text": "Your message here"
        }
      ]
    }
  }
}

Architecture

Composition Over Inheritance

The library uses composition - your agents.Agent instances remain unchanged while A2AWrapper adds A2A capabilities around them.

# Your agent stays pure
agent = Agent(name="MyAgent", model="gpt-4o-mini", tools=[...])

# A2A wrapper adds interoperability  
service = A2AWrapper(openai_agent=agent)

Automatic Skill Derivation

Tools are automatically converted to A2A skills:

• Tool name → Skill name
• Tool description → Skill description
• Tool parameters → Skill JSON schema
• Tool function → Skill handler

Multi-Agent Patterns

The library supports several multi-agent patterns:

1. Chain: Agent A → Agent B → Agent C
2. Hub: Coordinator agent calls multiple specialists
3. Pipeline: Data flows through processing stages
4. Peer-to-Peer: Agents discover and call each other

Development

Setup

git clone https://github.com/prassanna-ravishankar/a2a-openai-agents
cd a2a-openai-agents
uv sync --all-extras

Testing

# Run tests
uv run pytest

# Run linting
uv run ruff check
uv run ruff format --check

# Test the CLI
uv run a2a-agents weather

API Reference

A2AWrapper

Main class for wrapping agents with A2A capabilities.

A2AWrapper(
    openai_agent: Agent,                           # Your Agent instance
    a2a_name: str | None = None,                   # Defaults to agent.name
    a2a_description: str | None = None,            # Service description
    a2a_version: str = "1.0.0",                   # Version string
    a2a_id: str | None = None,                    # Unique identifier
    a2a_skills: list[A2ASkillConfig] | None = None # Custom skills
)

A2ASkillConfig

Configuration for custom A2A skills.

A2ASkillConfig(
    name: str,                                     # Skill name
    description: str,                              # Human description
    handler: Callable[..., Awaitable[dict]],      # Async handler function
    parameters: dict[str, Any] | None = None      # JSON Schema
)

run_a2a_wrapper_server

Start the A2A HTTP server.

run_a2a_wrapper_server(
    wrapper_instance: A2AWrapper,
    port: int = 8000,
    host: str = "0.0.0.0"
)

Contributing

Contributions are welcome! Please check out our development guide for setup instructions and coding standards.

License

MIT License - see LICENSE for details.

Links

• PyPI Package: https://pypi.org/project/a2a-openai-agents/
• Source Code: https://github.com/prassanna-ravishankar/a2a-openai-agents
• A2A Protocol: https://a2a-documentation-url
• OpenAI Agents SDK: https://github.com/openai/agents-sdk