Google's A2A Protocol Takes Aim at Agent Task Handoffs

A2A is a response to a structural limit in earlier multi-agent frameworks such as LangChain, AutoGen, CrewAI and LangGraph. Those tools could coordinate agents, but mainly within the same repository, language or SDK, leaving no common way to safely delegate work to an outside company's agent. Calling a traditional API requires the caller to work out every parameter and step in a prompt, which makes full task delegation impossible.

A2A's design aims to change that by defining roles of Client Agent and Remote Agent. A Client Agent receives a broad request and delegates a goal, while a Remote Agent publishes its skills and works autonomously, reporting progress and results. This mirrors the idea of handing work to a specialist colleague rather than operating a tool.

There is a clear line between A2A and MCP. MCP is described as a North-South standard for an agent to reach databases, files and APIs safely in its own environment, while A2A is an East-West standard for handing tasks across different models, organizations and frameworks. A remote agent can use MCP internally for tools and report back to a client agent via A2A, making the two complementary.