A2A Protocol (Agent2Agent)

A2A Protocol (Agent2Agent)

Also called: 智能体间通信协议 · Agent2Agent

An open protocol from Google (April 2025, later donated to the Linux Foundation) for agents from different vendors to discover each other and delegate tasks.

Agent A+ Agent CardAgent B+ Agent Cardtaskupdates
Schematic (simplified)

Core concepts

Each agent publishes an Agent Card — metadata describing its identity, capabilities, and endpoint — which other agents use to discover it. Collaboration is organised around a task: an initiator creates it, the two sides advance it by exchanging messages, and the task carries explicit state and artifacts. A2A deliberately does not require exposing internal implementation — interoperability is at the external-interface level only.

Division of labour with MCP

A common framing: MCP connects agents to tools/data, A2A connects agents to agents. A system can use both — an agent gains capabilities via MCP and outsources sub-tasks to a specialised agent via A2A. In 2026 A2A moved into the Agentic AI Foundation, with 150+ supporting organisations.

Common misconceptions

  • A2A is not trying to replace MCP; they address different connection relationships.
  • Adopting A2A does not mean exposing an agent's internal logic — it standardises only the external collaboration interface.

Related terms

Sources

  1. A2A Project — a2a-protocol.org
  2. Google Developers Blog — Google Cloud donates A2A to the Linux Foundation (2025)
  3. Linux Foundation — A2A one-year update (150+ organisations)

Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.