What is the A2A Protocol (Agent2Agent)?

The A2A Protocol (Agent2Agent) is the open standard announced by Google in April 2025 that lets AI agents built by different vendors discover each other, delegate tasks and coordinate work across distinct enterprise systems, without all of them having to sit on the same platform. A procurement agent built on one vendor's stack can task a logistics agent built on another vendor's stack, each keeping its own internal implementation and underlying model, with no rewrite needed to get them talking to each other. Communication happens through standardized JSON messages and public "agent cards" that describe each agent's capabilities, somewhat like an API manifest but designed to be read and negotiated by another agent rather than by a human developer. Google brought in more than 50 technology partners during the initial design precisely to keep it from becoming one vendor's proprietary standard.

Governance and current status

In June 2025 Google transferred the protocol to the Linux Foundation, which established the Agent2Agent project under the Agentic AI Foundation to keep governance vendor-neutral, not controlled by a single supplier. The current version is 1.2, which introduces cryptographically signed "agent cards": a digital business card that lets you verify an agent's domain and identity before exchanging tasks or data with it.

Enterprise adoption in 2026

A year after launch, A2A has surpassed 150 organizations in production, not pilots, routing real tasks between agents built on different platforms. Microsoft, AWS, Salesforce, SAP and ServiceNow already run it in production, a sign the protocol has moved past developer curiosity into the operational stack of large-scale enterprises.

The key distinction from MCP

A2A and MCP solve two different, complementary problems. A2A defines how agents communicate and coordinate with each other, including across different organizational boundaries (one vendor's agent talking to another vendor's agent). MCP defines how a single agent connects to internal tools and data (a database, a CRM, a repository). They are not competitors: together they form the base layer for interoperable multi-agent systems, one facing outward between agents, the other facing inward between an agent and its tools.

Why it matters for your business

For a company building on multiple vendors, A2A confirms a principle worth defending: you do not need to pick a single AI agent vendor if an open protocol lets different platforms interoperate. The same vendor-neutral logic applies here as for MCP and for AI agents in general: the best choice is not the most closed platform, but the one that integrates best with the rest of your stack.