A2A protocol
Google's 2025 open Agent2Agent protocol — a standard for agents from different vendors to discover each other, exchange tasks, and stream results.
A2A is the multi-agent counterpart to MCP. Where MCP standardizes how an agent talks to tools, A2A standardizes how an agent talks to another agent. The spec covers capability discovery, task delegation, streaming responses, and authentication.
Announced by Google in April 2025 with 50+ launch partners, A2A is built on standard web primitives (HTTP, JSON, Server-Sent Events). An A2A-compliant agent exposes a "capability card" describing what it can do; other agents read the card and decide whether to delegate.
In 2026, A2A is gaining traction in enterprise multi-agent stacks where agents are built by different teams or vendors. The MCP-A2A pair is becoming the assumed backbone for cross-org agent orchestration.
Frequently asked
How is A2A different from MCP?+
MCP: agent ↔ tool. A2A: agent ↔ agent. Most production stacks in 2026 use both — MCP to give individual agents tools, A2A to let those agents coordinate.
Do I need A2A if my agents share a codebase?+
No. A2A solves the cross-vendor case. For agents you build yourself, in-process function calls or a message queue are simpler.