Home / Glossary / A2A protocol
Glossary
What is the A2A (Agent2Agent) protocol?
The A2A (Agent2Agent) protocol is an open standard that lets AI agents built by different teams and companies talk to each other and work on tasks together. One agent can find another, hand it a task and get the result back. MCP connects an agent to tools, and A2A connects an agent to other agents.
By the AIBound team · Updated 17 November 2026
Key takeaways
- A2A is an open protocol that lets AI agents from different makers talk to each other and share tasks.
- MCP connects one agent to its tools and data, and A2A connects that agent to other agents.
- An agent that can hand work to other agents has more reach than its own tools give it.
Why the A2A protocol matters
AI agents are built by many companies on many frameworks. Without a shared protocol, every pair of agents needs its own custom link. A2A gives them one common way to describe what they can do, pass a task and return a result. The agents do not have to share their internal memory, tools or logic to work together.
For security, A2A changes how far a task can travel. An agent on an employee laptop can pass work to an agent run by another team or another company. The task then moves through software the security team may never have reviewed. That makes it more important to know which agents employees run and what each one may do.
How the A2A protocol works
A2A has two sides. The client agent starts the request for a user. The remote agent receives it and does the work.
- 1 The remote agent publishes an Agent Card, a JSON document that lists its identity, its address, its skills and how to sign in.
- 2 The client agent reads the card and decides whether that agent fits the task.
- 3 The client sends a message over HTTPS using JSON-RPC 2.0, and the remote agent opens a task with its own ID.
- 4 The remote agent returns the result in one reply, as a stream, or as a push notification for long tasks.
A2A vs MCP
The two protocols do different jobs and are built to be used together. MCP gives one agent its tools. A2A lets that agent work with other agents, and each of those agents keeps its own tools.
| Question | A2A | MCP |
|---|---|---|
| What does it connect? | An agent to other agents | An agent to tools, APIs and data |
| What is on the other end? | An agent that plans, keeps state and can hold a long exchange | A tool with set inputs and outputs |
| What is passed across? | A task for the other agent to work on | A request to run one tool |
Sources
- A2A Protocol documentation · a2a-protocol.org
- A2A Protocol: key concepts · a2a-protocol.org
- A2A Protocol: A2A and MCP · a2a-protocol.org
- A2A project on GitHub · github.com
Sources checked 17 November 2026
Examples
A2A protocol examples
FAQ
Frequently asked questions
What is the A2A (Agent2Agent) protocol?
The A2A (Agent2Agent) protocol is an open standard for communication between AI agents. It lets an agent find another agent, read what it can do, hand it a task and receive the result. The agents can come from different makers, and they do not share their internal memory or tools.
What is the difference between A2A and MCP?
MCP connects one AI agent to tools, APIs and data. A2A connects an agent to other agents. A tool does one set job when called, and an agent plans its own steps and can hold a long exchange. The two are made to work together, and many agent systems use both.
Who runs the A2A protocol?
A2A is an open source project under the Linux Foundation, and its code and specification are published under the Apache 2.0 licence. The project site lists version 1.0 of the specification. Anyone can read the specification and build an agent that follows it.
How does A2A handle sign-in between agents?
Each remote agent states its sign-in requirements in its Agent Card. The client agent sends credentials, such as an OAuth token or an API key, in the HTTP headers, apart from the A2A messages. The protocol covers how agents sign in to each other. Each company still decides which agents its employees may run.