The Open-Source AI Stack
RSS

The Stack · Protocols · Open source

Agent-to-Agent (A2A)

Google-seeded protocol for inter-agent communication; under Linux Foundation; 150+ org signatories.

Apache 2.0 · beta · Project site → · GitHub →

A2A (Agent-to-Agent) is a protocol for inter-agent communication seeded by Google in 2024 and donated to the Linux Foundation in 2025. Where MCP standardizes how an agent talks to tools and resources, A2A standardizes how two agents talk to each other: capability discovery (each agent publishes a structured "agent card" describing what it can do), task delegation, and result reporting. A2A matters because as agent ecosystems mature, agents will increasingly need to negotiate work with other agents rather than just call tools. The complementary positioning is the important framing: MCP is agent-to-tool, A2A is agent-to-agent, and both are needed for a multi-agent ecosystem that is not controlled by a single platform. Compared to siblings: MCP is the mature complement at the same wire layer; ACP (Agent Client Protocol) is an earlier-stage spec focused on agent-to-client surfaces (chat UIs, IDEs); various proprietary agent-platform protocols exist but lack the open-spec character of A2A. Production-readiness: shipping and adopted by a broad signatory list including Salesforce, SAP, ServiceNow. Real usage is still thinner than MCP usage; the boundary between "this should be an MCP server call" and "this should be an A2A agent call" remains fuzzy in practice and a topic of ongoing spec discussion. The strategic question is whether A2A becomes the durable inter- agent layer or whether agent-platform vendors fork their own.

Sources

Want a follow-up? Ask the chat about Agent-to-Agent (A2A) in context. It will compare to siblings at the same layer and ground every claim in the wiki.

Other projects at the Protocols layer

5 siblings · ordered open first