Agent2Agent

The Agent2Agent (A2A) Protocol is an open standard for communication and interoperability between independent AI agents that may be built by different teams, on different frameworks, and hosted by different vendors. Rather than requiring one agent to know the internals of another, A2A gives them a shared interaction model — a common way to advertise capabilities, exchange tasks, and stream results — so that opaque agents can still cooperate across organizational boundaries.

• Agent Cards - A machine-readable description an agent publishes so others can discover what it does and how to reach it.
• Task-based interaction - Work is framed as tasks that are submitted, tracked, and returned, with support for long-running and streaming responses.
• Framework and vendor neutral - The protocol assumes agents are black boxes, letting participants keep their own stacks and models private.
• Built on familiar web foundations - It leans on HTTP, JSON, and server-sent events so it fits existing API infrastructure rather than replacing it.

In real API operations A2A sits one layer above the tool-calling world of MCP: where MCP wires a single agent to tools and data, A2A is how whole agents delegate to and coordinate with one another. That makes it a governance surface as much as an integration one — the Agent Card becomes a contract to review, and the task exchange becomes traffic to authenticate, authorize, and audit as agent-to-agent interactions start crossing company lines. It is a close sibling of standards like AGENTS.md in the emerging agentic stack.