The A2A protocol, short for Agent2Agent, is an open standard from Google, now governed by the Linux Foundation, that lets independently built AI agents discover each other's capabilities, exchange tasks and results, and coordinate work across organizations and frameworks without a shared codebase or a single vendor's tooling.
What problem does the A2A protocol actually solve?
Before a shared protocol, getting two agents built by different teams to work together meant a custom integration for every pair. A2A replaces that with one shared contract: each agent publishes an agent card describing what it can do, a caller reads that card to find the right agent, and the two exchange structured task requests over standard web protocols.
A2A protocol vs MCP: what is the actual difference?
They solve adjacent but different problems. MCP standardizes how a model calls a tool, a connection to something that is not itself an agent. A2A standardizes how one agent calls another autonomous agent as a peer. A multi-agent system built inside one company usually does not need A2A; it earns its complexity once agents cross an organizational boundary and cannot share code.
Where does A2A fit into a real agentic system?
A2A is one option for cross agent coordination, alongside simpler in-process patterns like the orchestrator-worker pattern used inside a single system. For most companies building one internal agentic AI system, A2A only becomes relevant once it needs to call an agent someone else built and does not control.
Frequently asked questions
How is the A2A protocol different from MCP?
MCP standardizes how a model calls tools and data sources. The A2A protocol standardizes how two independent agents talk to each other as peers, publishing an agent card describing what they can do, then exchanging tasks and results. Most production systems end up using both, MCP for tools, A2A for cross agent coordination.
Who governs the A2A protocol?
Google introduced it in 2025 and later transferred the specification and SDKs to the Linux Foundation, which now runs it as a vendor neutral open source project with contributions from a large group of technology partners rather than a single company.
Do both agents need to run the same framework to use A2A?
No, that is the point of the protocol. A2A is built on standard HTTP, Server Sent Events, and JSON-RPC, so an agent built on one framework can discover and call an agent built on a completely different one as long as both speak A2A.