A2A Protocol Joins the Agentic AI Foundation Alongside MCP: What It Means for Agent Projects

On August 17, 2026, the Agentic AI Foundation (AAIF) announced that the Agent2Agent (A2A) protocol is joining the foundation as a hosted project. A2A is the open standard that defines how one AI agent finds another, hands it a task and gets the result back, even when the two were built by different vendors on different frameworks.

The move puts A2A under the same roof as the Model Context Protocol (MCP), the standard for connecting agents to tools and data. For a company planning custom software development that involves AI agents, this is a governance story with a practical consequence: the two protocols most likely to sit underneath an agent project are now managed in one neutral foundation instead of being steered by individual vendors.

This article explains what A2A does, what was announced, what has not changed, and how to factor the news into a software decision.

What was announced

The AAIF post, dated August 17, 2026, says A2A is joining as a hosted project. The A2A project followed with its own announcement on August 27, 2026, stating that the protocol has been accepted as a Growth Stage project at the foundation.

The AAIF describes itself in the post as hosted by the Linux Foundation. It lists the projects it now holds as layers of one stack: AGENTS.md for giving instructions and context to agents, goose as an agent runtime, MCP for connecting agents to tools, agentgateway for routing, policy and monitoring at the boundary of agent systems, and A2A for communication between independent agents.

What A2A does, in plain terms

Many business processes cross system boundaries. A procurement agent in your company may need a quote from a supplier's agent. A support agent may need a shipping agent to rebook a delivery. Without a shared standard, each of those handoffs is a custom integration that someone has to build and maintain.

According to the AAIF post, that integration work, not the agents themselves, was the significant expense teams kept running into. A2A addresses it with a published description called an agent card: a structured statement of what an agent can do and how to reach it. Another agent reads the card, discovers the capabilities and delegates a task in a standard format.

The post gives a short history. Google launched A2A in April 2025, IBM's Agent Communication Protocol merged into it in August 2025, and version 1.0 shipped in March 2026. The AAIF says v1.0 added multi-protocol bindings, version negotiation, multi-tenancy and signed agent cards, which let one agent verify the identity of another cryptographically.

How A2A and MCP fit together

The two protocols are often confused, and the announcements are useful for separating them. The AAIF post assigns MCP the job of connecting an agent to tools, data sources, applications and services. It assigns A2A the job of letting independent agents discover each other, communicate, delegate tasks and exchange results.

The A2A project's post describes the same split as vertical and horizontal: MCP reaches down from an agent into the systems it uses, while A2A reaches across to other agents. A typical design would use both. Your order-handling agent might use MCP to read your own inventory database and A2A to ask a logistics partner's agent for a delivery slot.

A2A is only relevant once more than one agent is involved. If you are not sure whether your project needs an agent at all, our guide to AI agents, chatbots and workflow automation covers that earlier decision.

What has not changed

This is a change of home, not a new version. The A2A project's post describes v1.0 as the stable specification and refers to SDKs in Python, Go, Java, JavaScript, .NET and Rust. Neither announcement describes a new protocol release, a breaking change or a migration that existing implementations must carry out as a result of the move.

A2A was already an open project before this announcement. What the two posts emphasize is that it now shares governance and location with the other layers of the agent stack, and that no single participant controls it.

Who is using A2A today

Both posts state that more than 150 organizations back the protocol. The AAIF post names production uses in several areas, which we repeat here as the foundation's claims and have not verified independently:

• Cloud platforms: Google Cloud, Microsoft Azure AI Foundry and AWS Bedrock AgentCore are described as supporting the building, hosting or discovery of A2A agents.
• Mobile: Huawei is described as using A2A between its HarmonyOS assistant and in-app agents, with Tencent's WeChat integrating with phone assistants through the protocol.
• Commerce: Google Cloud and PayPal are described as extending A2A into agent-driven shopping through the Agent Payments Protocol (AP2), which handles payment authorization while agents communicate over A2A.

For most US and UK businesses, the cloud platform support is the relevant point. It suggests that an agent hosted on one major cloud can be reached in a standard way by an agent hosted on another.

What this means for your business

The views in this section are our interpretation, offered as guidance.

Governance sounds abstract, but it affects a real risk in a software investment. When a protocol is controlled by one vendor, that vendor's product priorities can change the rules your software depends on. The AAIF post makes this argument directly, saying open governance means the protocol your agents rely on is not subject to a single vendor's product decisions. That does not remove all risk. Standards under neutral governance still change, and vendors can implement them unevenly. It does make a multi-year commitment easier to justify.

The news is most relevant in three situations. The first is a company that expects its agents to work with agents run by suppliers, customers or partners. The second is a company buying agent products from more than one vendor and wanting them to cooperate. The third is a software company whose product may need to be reachable by customers' agents in future, where publishing an A2A interface could become an expected feature in the way a public API is today.

If your plan is a single assistant or a single agent working only inside your own systems, A2A is not something you need now. Adding it early would mean extra build and security work without a matching benefit.

What to do next

If you are evaluating agent platforms or vendors, add two questions to your list: does the product support A2A version 1.0, and does it support signed agent cards? The answers tell you how easily that product will cooperate with agents you do not control.

If you are commissioning custom agents, ask your developers to keep agent-to-agent communication behind a clear interface, so that A2A can be adopted without rewriting the agent's core logic. Treat any agent exposed to outside parties like a public API: decide who may call it, what it is allowed to do on their behalf, how each request is authenticated, and how activity is logged and reviewed.

Finally, keep watching both projects. With A2A and MCP in the same foundation, closer alignment between them is a reasonable expectation, but neither announcement commits to specific joint changes or dates, so do not plan around features that have not been published.

Conclusion

On August 17, 2026, the Agentic AI Foundation announced that the A2A protocol is joining as a hosted project, placing the standard for agent-to-agent communication alongside MCP under Linux Foundation-hosted, vendor-neutral governance. The protocol itself remains at version 1.0. For businesses, the practical effect is more confidence that agents built on these standards will keep working across vendors over the life of a project.

Entrant Technologies builds websites, web applications, mobile apps and custom software. If you are considering a system in which several agents, or agents from several companies, need to cooperate, you can request a quote and outline what you want to build.