BYD TechThe short answer
A2A, short for Agent2Agent, is an open protocol that lets AI agents built by different teams or vendors discover each other, exchange messages and hand off tasks. Google introduced it in April 2025, and it is now a Linux Foundation project. It complements MCP: MCP connects an agent to tools and data, while A2A connects an agent to other agents. Most products need MCP-style tool access first, and A2A only when their agent must work with agents they do not control.
Key points
- A2A lets agents from different builders work together.
- Introduced by Google in April 2025; now governed under the Linux Foundation.
- MCP is agent-to-tool. A2A is agent-to-agent. They fit together.
- Agents describe what they can do in an Agent Card.
- Most products do not need it on day one.
What A2A is
As companies deploy agents, those agents increasingly need to work with agents somewhere else: a purchasing agent asking a supplier’s agent for a quote, or a support agent handing a case to a billing agent built by another vendor. A2A gives them a shared way to do it, without either side exposing its internal prompts, memory or tools.
Google announced A2A on April 9, 2025, with more than 50 technology partners. In June 2025 the Linux Foundation launched the Agent2Agent project to govern it, with companies including Google, AWS, Microsoft, Salesforce, SAP, ServiceNow and Cisco involved.
How it works
- Each agent publishes an Agent Card: a description of what it can do, where to reach it and how to authenticate.
- A client agent finds a remote agent and reads its card.
- It sends a task as a message, over standard web technologies such as HTTP and JSON-RPC.
- The remote agent works on the task, sending updates for long jobs, and returns results called artifacts.
A2A vs MCP
| MCP | A2A | |
|---|---|---|
| Connects | An agent to tools and data | An agent to other agents |
| Introduced by | Anthropic, November 2024 | Google, April 2025 |
| Typical use | Read a database, call an API, search files | Delegate a task to another team’s or vendor’s agent |
| Other side is | A tool that does what it is told | An agent that decides how to do the task |
The A2A project itself describes the two as complementary, not competing. An agent can use MCP for its own tools and A2A to work with other agents. See What is an MCP server?.
Does your product need A2A?
- Not if your agent only uses your own tools and data. Plain tool calls or MCP are enough.
- Not if all your agents run in one system you control. See Multi-agent systems.
- Possibly if partners or customers will run agents that need to call yours.
- Possibly if you sell to companies standardizing on agent platforms that support it.
Treat messages from outside agents as untrusted input, with the same permission checks as any other caller. We connect agents to the tools and systems they need, and to other agents where it earns its place, in AI agent development.
Sources
Checked when this guide was written.
