Glossary · Protocols and standards
A2A (Agent2Agent protocol)
A2A is an open protocol that lets AI agents from different vendors discover each other and exchange messages, tasks and results over standard web technology.
A2A (Agent2Agent) is an open protocol that lets independent AI agents, built on different frameworks and run by different organizations, discover each other and work on tasks together without exposing their internal memory, prompts or tools.
Origin and governance. Google announced A2A in April 2025. In June 2025 the Linux Foundation announced it as the project’s new home under vendor-neutral governance. Version 1.0.0 of the specification was published on GitHub in March 2026.
How it is built. The specification has three layers. A data model is defined in Protocol Buffers, and a2a.proto is the normative source. Abstract operations such as Send Message, Get Task and Cancel Task sit on top of it. Protocol bindings map those operations to JSON-RPC 2.0, gRPC and HTTP+JSON/REST.
A typical exchange. The client agent fetches the remote agent’s Agent Card, usually from /.well-known/agent-card.json, to learn its skills, endpoints and authentication requirements. It then sends a message. The remote agent answers with a message, or with a task that moves through states such as TASK_STATE_WORKING, TASK_STATE_INPUT_REQUIRED and TASK_STATE_COMPLETED. Results arrive as artifacts made of parts: text, structured data or files. Clients follow long tasks by polling, streaming or push notifications.
A minimal v1.0 request over the JSON-RPC binding, sent with the header A2A-Version: 1.0 (a server treats a missing header as 0.3):
{
"jsonrpc": "2.0",
"id": 1,
"method": "SendMessage",
"params": {
"message": {
"messageId": "msg-1",
"role": "ROLE_USER",
"parts": [{"text": "What is the status of order 1042?"}]
}
}
}
What A2A leaves open. Security relies on standard web practice: TLS, authentication schemes declared in the Agent Card, and optional JWS signatures on the card. The specification does not say how to find an agent when you only have a phone number or a brand name, how to decide whether a provider is trustworthy, or what a credential obtained through TASK_STATE_AUTH_REQUIRED covers. Implementations and A2A extensions fill those gaps.
Neighbouring terms. MCP gives one agent access to tools and data, while A2A connects agents to each other. AP2 and x402 each define a way to carry payments over A2A.
Questions
- Who maintains A2A?
- Google created A2A and announced it in April 2025. Since June 2025 the Linux Foundation has hosted it as a vendor-neutral open source project, developed in the open at github.com/a2aproject/A2A.
- Does A2A replace MCP?
- No. MCP connects an agent to tools and data sources. A2A connects independent agents to each other. The A2A specification describes the two as complementary, and one agent can use both.
Sources
- A2A Protocol Specification (accessed )
- A2A protocol definition (a2a.proto) (accessed )
- A2A releases on GitHub (v1.0.0) (accessed )
- Linux Foundation Launches the Agent2Agent Protocol Project (23 June 2025) (accessed )