Glossary
A2A protocol: what it is and how it works
In short
The A2A (Agent2Agent) protocol is an open standard that lets AI agents from different makers find each other, hand over tasks and get results back.
AI agents · Agent2Agent protocol
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine children from different schools at one contest. They do not know each other, but each wears a badge with a name and a skill. The badges show who can help with a drawing task and who can help with counting. The protocol gives smart programs such badges and shared rules for talking, so they can help each other.
In plain words
Companies now buy AI agents from different vendors: one handles email, another works with accounting, a third searches for job candidates. Without a shared language, each pair needs its own manual link. A2A is that shared language. Each agent publishes a card that says who it is, what it can do and how to connect to it. Another agent reads the card, sends a task and follows its status: accepted, in progress, waiting for a human answer, done. The result comes back in a form both sides understand, such as a file or a table. The agent’s internal instructions and memory stay private.
For business
A shared standard reduces dependence on one vendor: an agent bought from one company can give work to an agent from another without a separate integration for each pair. Google launched the protocol in April 2025 with more than 50 technology partners, and in June 2025 the project moved to the Linux Foundation. For a business, this means no single corporation sets the rules. The risks are about trust: an outside agent gets your data and your tasks. Before you connect it, check how it proves its identity, what rights it gets and where it stores the messages. Teams handle these questions together with the CRM and accounting integrations, not separately.
For a specialist
Agent2Agent is an open protocol for communication between independent agent systems that do not share their internal state, memory or tools with each other. The latest released version of the specification is 1.0.0. A server publishes an Agent Card: a JSON document at /.well-known/agent-card.json with the agent’s skills, service address and authentication schemes. The unit of work is a Task with a life cycle: submitted, working, input-required, auth-required, and the final states completed, failed, canceled and rejected. Agents exchange Messages made of Parts, and results come back as Artifacts. The specification describes JSON-RPC 2.0, gRPC and HTTP+JSON bindings, streaming updates and push notifications to the client’s webhook. Appendix B calls the protocols complementary: MCP connects an agent to tools and data, while this standard links agents to each other as equal participants.
Example
A purchasing agent finds a new supplier and sends the legal team’s agent a task to check the contract. That agent returns a report with comments, and the task status stays visible in the purchasing system the whole time.
Related terms
Questions and answers
How is A2A different from MCP?
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map