What is the A2A protocol?
A2A (Agent2Agent) is an open protocol for communication between AI agents. Introduced by Google in April 2025 and governed by the Linux Foundation, it lets agents built on different frameworks discover each other through agent cards, delegate tasks, and receive results synchronously, as a stream, or as pushed notifications.
How it works
An A2A server is an agent that accepts work; an A2A client is whoever sends it, usually another agent. The server publishes an agent card describing its skills and endpoints. The client picks a skill, sends a message, and gets back a task with an ID.
The task then moves through a defined lifecycle: submitted, working, and eventually a terminal state of completed, failed, canceled, or rejected. Two extra states, input-required and auth-required, let a task pause and wait for the client to supply something, which is what makes multi-turn collaboration between agents possible. The 1.0 specification defines three interchangeable transports: JSON-RPC 2.0, gRPC, and plain HTTP+JSON.
Getting results back
Quick tasks return their result in the response. For anything longer, A2A offers streaming over Server-Sent Events, or push notifications: the client registers a webhook URL, disconnects, and the server POSTs status and artifact updates as they happen. Push is the mode built for asynchronous agent communication, where a task might run for hours and the client can't hold a connection open.
Push notifications are ordinary webhooks, which means A2A servers inherit the classic sender-side problems: retries when a receiver is down, authentication of the delivery, and observability into what was delivered where. The spec defines the payload but leaves that operational layer to the implementer; how A2A push notifications work covers it in detail, and it's the layer [Svix]https://www.svix.com/?utm_source=resources&utm_medium=content) provides off the shelf.
A2A and MCP
A2A connects agents to each other; MCP connects an agent to its tools. They're designed to be used together, with MCP inside each agent and A2A between them. A2A vs MCP has the full comparison, and agent to agent communication covers the mechanisms both protocols build on.
Frequently asked questions
Who created the A2A protocol?
Google introduced A2A in April 2025 with launch support from dozens of companies, then donated the specification and SDKs to the Linux Foundation in June 2025. Microsoft, AWS, Cisco, Salesforce, and SAP are among the members backing it.
What transports does A2A support?
The 1.0 specification defines three: JSON-RPC 2.0 over HTTP, gRPC, and HTTP+JSON/REST. An agent declares which it supports in its agent card, and all three carry the same task lifecycle and message types.
Are A2A push notifications the same as webhooks?
Functionally yes. The client registers a URL and the server POSTs task updates to it. The spec defines the payload and an authentication field but not retries, signing, or delivery tracking, so implementers handle the webhook operational layer themselves.
Ready to send webhooks?
Svix handles signing, retries, rate limiting, and delivery observability for the webhooks you send to your users, so your team can stay focused on your product.
Start sending webhooks with Svix or read the build vs. buy analysis