Agent2Agent Protocol (A2A)

The Agent2Agent Protocol (A2A) is an open standard for communication between independent AI agents. It lets an agent discover what another agent can do, hand it a task, follow its progress and receive the results, even when the two agents were built by different companies on different frameworks. Where the Model Context Protocol connects one agent to tools and data, A2A connects agents to each other.

Google introduced A2A in April 2025 and later contributed it to the Linux Foundation, which now hosts the project. Version 1.0, the first stable release, shipped on March 12, 2026, guided by a technical steering committee with representatives from eight technology companies. At its one-year mark in April 2026 the Linux Foundation said more than 150 organizations supported the standard and that it was integrated across Google, Microsoft and AWS platforms. Agents built with frameworks such as LangGraph or CrewAI can use it to delegate subtasks to one another (AI agent frameworks).

The protocol has a few core pieces (A2A specification):

• The Agent Card, a JSON document usually published at /.well-known/agent-card.json, describes an agent's name, skills, endpoints and authentication requirements. Since v1.0, cards can be cryptographically signed so a client can verify who published them.
• Tasks are units of work with a defined lifecycle, suited to jobs that take minutes or hours. Agents exchange messages and return outputs as artifacts.
• Streaming and push notifications report progress on long-running tasks.
• Transport bindings for JSON-RPC, gRPC and HTTP+JSON, with clients sending an A2A-Version header so agents supporting different versions can negotiate.

Version 1.0 tightened the spec rather than reinventing it. It separated the application protocol from its transport mappings, added multi-tenancy and modernized security flows, and introduced breaking changes in the interaction protocol while letting Agent Cards advertise both v0.3 and v1.0 support during migration.

Other standards build on or align with A2A. Google's Universal Commerce Protocol is designed to be compatible with A2A, as well as with MCP and the Agent Payments Protocol, and the first version of AP2 shipped as an A2A extension. Microsoft's NLWeb project says every NLWeb instance acts as an MCP server and plans A2A support.

For GEO, A2A matters because multi-agent systems change who evaluates your brand. A research or shopping agent may delegate part of a task, such as checking availability or comparing prices, to a specialist agent. The system that decides whether your product or content makes the final answer can be several steps removed from the person who asked, and it may never visit your site directly. This extends the logic of query fan-out: visibility now has to survive delegation between agents as well as parallel searches.

Brands engage with A2A in two ways. Companies that offer bookable or queryable services (travel, retail inventory, B2B software) can publish Agent Cards so other agents can find and call them. Content-focused brands rarely implement the protocol, but they benefit from the same groundwork as agent experience optimization: facts, prices, availability and policies that machines can read and that match across every source.