The Agent Protocol Landscape: MCP, ACP, And A2A

If you have been half-following AI news, you have seen a soup of three-letter protocols fly past: MCP, A2A, ACP. They sound interchangeable and they are not. Each solves a different problem, and the reason there are three rather than one is that connecting a model to a tool, letting two agents talk, and letting an agent spend money are genuinely different jobs with different failure modes.

This is the map. As of August 2026, all three are open standards moving toward vendor-neutral governance, which is the part that makes them worth learning rather than dismissing as one company’s land grab. Here is what each one does and where it fits.

MCP Connects One Model To Your Tools And Data

The Model Context Protocol is the vertical connection: it links a single AI model down to the tools and data it needs to do a job. Instead of writing a custom integration for every app a model touches, you expose a tool through MCP once and any MCP-capable model can use it.

Think of it as the port on the back of the model. A file store, a database, an internal API: MCP is the standard shape they plug into. It started at Anthropic and has since picked up a substantial 2026 spec revision that we covered in what the latest MCP spec actually changed. The primary source is the Model Context Protocol site. If your question is “how does my agent reach my systems,” MCP is the layer you are asking about.

A2A Lets Agents Talk To Each Other

The Agent2Agent protocol is the horizontal connection: it lets separate agents communicate, delegate, and coordinate. Where MCP wires one model to its tools, A2A handles the case where your agent needs to hand part of a task to someone else’s agent, built by a different team on a different stack.

A2A was created by Google and announced in April 2025, then donated to the Linux Foundation in June 2025, which is what put it on a neutral footing rather than a single vendor’s roadmap. By its first anniversary in April 2026 it reported more than 150 supporting organizations. The primary source is the A2A protocol repository. If your question is “how do two agents from two companies cooperate without a bespoke integration,” A2A is the answer that layer.

ACP Lets An Agent Complete A Purchase

The Agentic Commerce Protocol is the transaction layer: it defines how an agent completes a purchase on a person’s behalf. That is a narrower and riskier job than the other two, because it moves money and cannot be undone as cheaply as a bad summary.

ACP is co-developed by OpenAI and Stripe and released under the Apache 2.0 license. It defines building blocks like agentic checkout, a cart, a product feed, payment delegation, and delegated authentication, and it is designed to interoperate with MCP rather than replace it. The primary source is the ACP specification repository, and we go deeper in the Agentic Commerce Protocol, explained. If your question is “how does an agent actually buy the thing,” ACP is that layer.

They Stack, They Do Not Compete

The mistake is reading these as rivals fighting for one slot. They compose. A real agent workflow can use MCP to reach your data, A2A to coordinate with another agent, and ACP to close a purchase, all in one task. Picking MCP does not mean rejecting A2A any more than choosing a keyboard means rejecting a monitor.

Because all three are open and heading toward foundation or permissive-license governance, betting on the stack does not lock you to one vendor. That is the strategic reason to care. Interoperability standards you do not control are safer to build on than a single company’s closed API, because the standard outlives any one provider’s product decisions.

Explaining The Difference Without Blurring It

The everyday failure is a summary that flattens three distinct protocols into “AI agent stuff,” which helps nobody decide anything. Keep the jobs separate.

Before:

There are a bunch of new AI agent protocols now, like MCP and A2A and ACP, and they basically all help agents work together and do things automatically, so we should look into adopting them.

After:

Three open agent protocols do three different jobs. MCP connects a model to tools and data. A2A lets separate agents communicate and coordinate. ACP lets an agent complete a purchase. They compose rather than compete, and as of August 2026 all three are open and moving toward vendor-neutral governance. Next step: decide which layer our use case actually needs first.

The second version lets a reader assign each name to a job and choose, instead of nodding at a category.

A Wrivio Context for explaining a set of standards could say:

Rewrite this so each protocol or standard is defined by the one job it does, in its own sentence, without merging them. Keep every protocol name, organization name, license, and date exactly as written. Do not add protocols, capabilities, or adoption figures that are not in the original.

Press Ctrl+Shift+Space, paste the draft, and check the diff. Watch for the model swapping one protocol’s job onto another, because a confident sentence that says MCP handles payments is wrong in a way that looks polished.

Common Questions

What is the difference between MCP, A2A, and ACP?

MCP connects one AI model to tools and data, A2A lets separate agents communicate and coordinate with each other, and ACP lets an agent complete a purchase on a person’s behalf; each is an open standard for a different job.

Are these protocols competitors?

No. They compose rather than compete: a single agent workflow can use MCP to reach data, A2A to coordinate with another agent, and ACP to close a transaction.

Who governs these protocols?

As of August 2026 all three are open and moving toward vendor-neutral governance. A2A was donated to the Linux Foundation in June 2025, MCP is under foundation-style stewardship, and ACP is released under the Apache 2.0 license.

How widely adopted is A2A?

A2A was created by Google and announced in April 2025, then donated to the Linux Foundation in June 2025, and by its first anniversary in April 2026 it reported more than 150 supporting organizations.

Download Wrivio for Windows to rewrite a protocol explainer so each standard keeps its own job instead of blurring into one paragraph.