Skip to content
FDE 101

Agent2Agent Protocol (A2A)

The Agent2Agent Protocol, or A2A, gives independently deployed agent systems a shared contract for discovery, messages, stateful tasks, updates, and task outputs.

Tiny example

Section titled “Tiny example”
airline assistant (A2A client)
          ↓ delegates policy review
passenger-care agent (A2A server)
          ↓ returns task updates and an artifact
airline assistant

The remote agent may use any framework, prompt, tools, or model internally. A2A exposes the interaction boundary without requiring either side to reveal that implementation.

Remember the boundary

Section titled “Remember the boundary”
  • Agent Card: what the remote agent claims to offer and how to connect.
  • Message: one communication turn made of typed parts.
  • Task: stateful work with a lifecycle.
  • Artifact: the concrete result of a task.

A2A does not make a discovered agent trustworthy. The client still verifies identity, authenticates, authorizes each operation, validates artifacts, and records the exchange.

Continue with the full A2A lesson or compare MCP, A2A, and A2UI.