Skip to content
AdminKit

A2A (Agent2Agent Protocol)

What is A2A (Agent2Agent Protocol)?

A2A (Agent2Agent Protocol) is an open spec letting independent AI agents, built on different frameworks, discover each other's capabilities via a published Agent Card and delegate tasks between them. It solves a different problem than MCP: MCP connects an agent to tools, A2A connects one agent to another. AdminKit's MCP & Agent Gateway governs the MCP side.

By AdminKit team, Netbase JSC · Updated

A2A (Agent2Agent Protocol) is an open specification for how independent AI agents — built by different teams, on different frameworks — discover each other’s capabilities and collaborate on a task, instead of one agent needing custom, one-off integration code for every other agent it talks to. A server-side agent publishes an Agent Card, a metadata document describing its skills, service endpoint, and authentication requirements, which a client-side agent fetches to learn what the other agent can do before delegating work to it. Under the hood, A2A exchanges Tasks, Messages, and Artifacts over one of several interchangeable bindings — JSON-RPC 2.0, gRPC, or HTTP+JSON — so the choice of transport doesn’t change the protocol’s data model. Originally published by Google in April 2025 with support from over 50 technology partners, and now maintained by the Linux Foundation as an open, vendor-neutral specification — its full technical specification defines the Agent Card, Task lifecycle and bindings in detail — A2A solves a different problem than MCP: MCP connects one agent to the tools and data it uses, while A2A connects one agent to another agent it needs to delegate a task to. The Linux Foundation announced the launch of the Agent2Agent project on June 23, 2025, at Open Source Summit North America, taking on the protocol Google created.

See the full AdminKit glossary for other related terms.

Frequently asked questions

How does A2A differ from MCP?

MCP connects one agent to the tools and data it uses, while A2A connects one agent to another agent it needs to delegate a task to -- they solve different problems and are often used together.

What is an Agent Card?

A metadata document a server-side agent publishes describing its skills, service endpoint and authentication requirements, which a client-side agent fetches before delegating work to it.

What transports can A2A run over?

A2A exchanges Tasks, Messages and Artifacts over one of several interchangeable bindings -- JSON-RPC 2.0, gRPC, or HTTP+JSON -- so the choice of transport doesn't change the protocol's data model.

Who maintains the A2A specification?

Google originally published A2A in 2025, and it is now maintained by the Linux Foundation as an open, vendor-neutral specification.