Google A2A — Agent-to-Agent Communication Protocol

What it is

Google A2A (Agent-to-Agent) is an open protocol that enables AI agents to discover, authenticate, and communicate with each other regardless of the framework or provider that built them. Each agent publishes an Agent Card describing its capabilities, and other agents use that card to initiate structured communication.

A2A targets teams building multi-agent systems where agents from different vendors or frameworks need to collaborate on tasks, share context, or delegate subtasks.

How it saves time or tokens

Without A2A, connecting agents from different frameworks requires custom integration code for each pair. A2A provides a standard protocol so any compliant agent can talk to any other compliant agent. The discovery mechanism (Agent Cards) eliminates manual configuration of agent endpoints. Structured message formats reduce the token overhead of free-form natural language communication between agents. Estimated token usage is around 2,600 tokens per multi-agent interaction.

How to use

1. Install the A2A SDK:

pip install a2a-sdk

1. Create an A2A server with an Agent Card:

from a2a import A2AServer, AgentCard

card = AgentCard(
    name='code-reviewer',
    description='Reviews code for bugs and style issues',
    capabilities=['code-review', 'security-audit'],
    endpoint='https://my-agent.example.com/a2a'
)

server = A2AServer(card=card)

@server.on_task
async def handle_task(task):
    code = task.input['code']
    return {'review': 'Looks good, no issues found.'}

server.run(port=8000)

1. Discover and call another agent:

from a2a import A2AClient

client = A2AClient()
agent = await client.discover('https://other-agent.example.com/a2a')
result = await agent.send_task({
    'type': 'code-review',
    'input': {'code': 'def hello(): print("hi")'}
})

Example

Agent Card JSON structure:

{
  "name": "code-reviewer",
  "description": "Reviews code for bugs and style issues",
  "capabilities": ["code-review", "security-audit"],
  "endpoint": "https://my-agent.example.com/a2a",
  "authentication": {
    "type": "bearer",
    "token_url": "https://auth.example.com/token"
  }
}

Related on TokRepo

• AI agent tools — frameworks for building and orchestrating AI agents
• Multi-agent frameworks — compare multi-agent orchestration approaches

Common pitfalls

• A2A is a communication protocol, not an agent framework. You still need LangGraph, CrewAI, or similar to build the agent logic.
• Agent Card discovery requires a well-known URL or registry. Without proper DNS or service discovery, agents cannot find each other.
• Authentication between agents must be configured explicitly. Do not expose agent endpoints without proper token-based auth in production.