Summary
At Google Cloud Next 2025, Google launched the Agent2Agent (A2A) protocol — an open HTTP/SSE/JSON-RPC standard for cross-vendor agent coordination. With more than 50 ecosystem partners on launch day and an explicit design to complement rather than compete with MCP, A2A addressed a layer of the agent stack that MCP had not: how agents coordinate with other agents.
What Happened
Google unveiled A2A at its Cloud Next conference alongside a partner coalition that included Salesforce, SAP, Atlassian, ServiceNow, and more than 45 other enterprise software vendors. The protocol defines how AI agents discover each other, delegate tasks, and exchange results — all over standard HTTP with Server-Sent Events for streaming and JSON-RPC as the message format.
Where MCP defines how an agent connects to tools and data sources (the model-to-tool relationship), A2A defines how agents coordinate with other agents (the agent-to-agent relationship). Google positioned the two protocols as complementary layers in an agent stack: an orchestrator agent uses A2A to route tasks to specialist agents, and those specialist agents use MCP to access the tools they need.
The reference implementation was open-sourced immediately, and the specification was submitted to a community governance process with participation from the launch partners.
Why It Matters
A2A filled a genuine gap in the emerging agent infrastructure. Multi-agent workflows — where one AI coordinates several specialized subagents — had become a common architecture in enterprise deployments, but each implementation was proprietary. A2A gave that coordination layer an open specification, enabling cross-vendor orchestration where a Salesforce agent could, in principle, delegate to a ServiceNow agent without either vendor writing custom integration code.
The timing, on the same day Google confirmed MCP adoption, was notable: Google was simultaneously accepting a competitor's standard (MCP) and proposing its own (A2A). The two-layer framing — MCP for tools, A2A for agents — proved influential and was widely adopted as a conceptual model by the developer community in subsequent months.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.