Agent2Agent protocol
In short
The Agent2Agent protocol, or A2A, is an open standard that lets independent AI agents discover each other, negotiate an exchange of information and collaborate on tasks without access to each other's memory, tools or internal state.
- Also called
- A2A, Agent2Agent Protocol
- Topic
- AI agents
- Level
- Some background helps
- First described
- Google introduced the Agent2Agent protocol on 9 April 20252.
The Agent2Agent protocol is an open standard for communication between independent, potentially opaque AI agents, including agents built with different frameworks or by different vendors1. It is meant to let agents discover capabilities, negotiate text, files or structured data, manage collaborative tasks and exchange information without access to each other's internal state, memory or tools1. The latest released version of the specification is 1.0.01.
A remote agent publishes an Agent Card, a JSON document with its identity, capabilities, skills, service endpoint and authentication requirements1. A message is one turn, from the user or the agent, made of parts that can hold text, a file reference or structured data1. The output of a task is an artifact built from those parts1. Send Message returns either a task or a direct message1. One passage says that call must return immediately, while a later field, returnImmediately, defaults to waiting until the task finishes or pauses for input or authentication1. Streaming, and an HTTP POST to a client webhook, are separate ways to receive updates1.
Google announced the protocol on 9 April 2025 as a complement to the Model Context Protocol, which provides tools and context to agents2. Project documentation treats the Model Context Protocol as agent-to-tool communication and A2A as agent-to-agent communication3. A2A is not a kit for building agents, it does not specify how an agent calls its own tools or sub-agents, and it is not a chat application3. Google developed A2A and donated it to the Linux Foundation, where a technical steering committee maintains it3.
Example
In Google's April 2025 example, a hiring manager asks an agent to find candidates for a job listing, a location and a skill set2. That agent then uses A2A to work with other agents that source candidates, schedule interviews and run background checks2.
Why it matters
Agents built in different ways still need a common description of how to discover capabilities and pass work without sharing private memory or tools1.
Questions people ask
What does A2A stand for?
A2A stands for Agent2Agent1. Google announced it on 9 April 2025 as an open protocol that lets agents communicate, exchange information and coordinate actions2.
Is A2A the same as the Model Context Protocol?
No. Project documentation says the Model Context Protocol connects an agent to tools, APIs and resources, while A2A lets agents discover each other, delegate tasks and share results3.