Last reviewed:
What is the A2A (Agent2Agent) protocol? Definition and difference from MCP
A2A (Agent2Agent) is an open protocol that lets AI agents built by different vendors discover each other, delegate tasks to one another and exchange results securely. Launched by Google in April 2025, it has since been governed by the Linux Foundation and complements MCP: MCP connects an agent to its tools, A2A connects agents to each other.
Google introduced A2A on 9 April 2025 with over 50 partners, including Atlassian, Salesforce, SAP, ServiceNow and Workday. The problem addressed: an agent in your CRM and an agent in your HR tool, built by two vendors, have no common language to cooperate. A2A provides one, relying on web standards (HTTP, JSON-RPC, SSE streams). Each agent publishes an "Agent Card", a machine-readable sheet describing what it can do and how to call it. A client agent then sends a task to a remote agent, follows its progress, including on processes lasting several hours, and retrieves the result, without either one accessing the other's memory or internal tools. On 23 June 2025, Google handed the project to the Linux Foundation; a technical steering committee brings together AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP and ServiceNow. Version 1.0, the first stable specification, was released on 12 March 2026: cryptographically signed agent cards, multi-tenant support on a single endpoint, JSON-RPC, gRPC and HTTP+JSON bindings. In August 2026, A2A joined the Agentic AI Foundation, the Linux Foundation body that already hosts MCP; the project claims the support of over 150 organisations.
Concrete example
Illustrative case (fictitious company). A transport and logistics group with 900 employees, based in Lille, uses a route planning agent supplied by its TMS vendor and a customer relations agent deployed in its CRM. When a customer asks via chat to postpone a delivery, the customer relations agent queries the planning agent via A2A: it sends the request, receives a proposed slot, then confirms to the customer. Neither vendor had to develop a specific connector for the other, and each agent keeps its own access: the CRM agent never sees fleet data.
Comparison
| Criterion | A2A (Agent2Agent) | MCP (Model Context Protocol) |
|---|---|---|
| What it connects | An agent to other agents | An agent (or a model) to tools and data |
| Origin | Google, April 2025 | Anthropic, November 2024 |
| Governance | Linux Foundation (June 2025), Agentic AI Foundation (August 2026) | Agentic AI Foundation, Linux Foundation (December 2025) |
| Core building block | Agent Card: an agent's capability sheet; tasks tracked over time | MCP server: exposes tools, resources and prompts |
| Internal visibility | Opaque: each agent keeps memory and tools private | The model sees the exposed tools and their results |
| Example | The CRM agent asks the planning agent for a slot | The agent reads a customer record in the CRM |
FAQ
What is A2A?
A2A (Agent2Agent) is an open protocol that serves as a common language between AI agents. An agent can discover what another can do, give it a task, follow its progress and retrieve the result, even if the two come from different vendors.
What is the difference between A2A and MCP?
MCP connects an agent to its tools and data (database, files, APIs). A2A connects agents to each other, each keeping its tools and memory private. The official A2A documentation sums it up: MCP inside agents, A2A between agents. The two are complementary and hosted by the Agentic AI Foundation.
Who governs the A2A protocol?
Created by Google (April 2025), A2A was handed to the Linux Foundation in June 2025. A technical steering committee brings together AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP and ServiceNow. In August 2026, the project joined the Agentic AI Foundation, also hosted by the Linux Foundation.
Is A2A production-ready?
Version 1.0, the first stable specification, was published on 12 March 2026. It brings signed agent cards and several connection modes (JSON-RPC, gRPC, HTTP+JSON). The major clouds (Google Cloud, AWS, Microsoft Azure) offer native support.
Does an SME need A2A?
Not directly: A2A is built in by software and agent platform vendors. An SME should check, when choosing a tool, that its agents support A2A, to avoid lock-in with a single supplier once it uses agents from several vendors.
See also
Further reading
Sources
- Announcing the Agent2Agent Protocol (A2A), Google for Developers, 9 April 2025. https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/
- Google Cloud donates A2A to Linux Foundation, Google for Developers, 23 June 2025. https://developers.googleblog.com/en/google-cloud-donates-a2a-to-linux-foundation/
- A2A Protocol Ships v1.0: Production-Ready Standard for Agent-to-Agent Communication, official A2A blog, 12 March 2026. https://a2a-protocol.org/latest/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/
- A New Chapter for A2A: Joining the Agentic AI Foundation, official A2A blog, 27 August 2026. https://a2a-protocol.org/latest/blog/2026/08/27/a-new-chapter-for-a2a-joining-the-agentic-ai-foundation/