Summary
The Agent-to-Agent (A2A) protocol is an open standard that lets independent AI agents discover one another, exchange messages, and delegate tasks across vendor and framework boundaries. Originated by Google and donated to the Linux Foundation, it is the sibling protocol to MCP: where MCP connects an agent to tools, A2A connects agents to each other.
What is the A2A protocol?
In A2A, each agent publishes an Agent Card that declares its identity, capabilities, and supported authentication schemes (such as OAuth 2.0, OpenID Connect, API keys, or mTLS). Cards can be cryptographically signed so a receiving agent can verify their origin and integrity. Communication runs over JSON-RPC 2.0 on HTTP(S) and supports several interaction modes: synchronous request/response, streaming for long-running tasks, and asynchronous push notifications for detached work.
A design goal is opacity preservation: agents can collaborate without exposing their internal state, memory, or tools—relevant for protecting intellectual property in multi-vendor settings. A2A sits as a protocol layer above existing agent frameworks rather than replacing them, so frameworks like Google ADK or LangGraph can act as A2A servers. Like MCP, it answers who an agent is, but the question of what an agent is ultimately permitted to do remains an open design area.
Why is the A2A protocol relevant?
- Cross-vendor interoperability: Lets agents from different vendors and frameworks collaborate over a common standard
- Verifiable identity: Signed Agent Cards let agents confirm who they are talking to
- Standard transport: JSON-RPC over HTTP with sync, streaming, and async modes fits existing infrastructure
- Complements MCP: Covers the agent-to-agent axis, while MCP covers the agent-to-tool axis