Agent2Agent and MCP: The Plumbing of the Agent Economy

Every time a new standard shows up, someone tells you it will change everything, and someone else tells you it is a nothing burger. Both are usually wrong. The useful question is smaller: what job does this thing do, and does it touch my business yet?

On April 9, 2025, Google announced the Agent2Agent protocol, A2A for short. The headlines made it sound like a rival to Anthropic's Model Context Protocol. It is not. It is the other half of the plumbing. Here is how to tell the two apart, in plain English, so you can ignore the noise and watch the right things.

What did Google actually announce on April 9?

Google released A2A as an open protocol, meaning a published set of rules any company can build against. Its goal is to let agents built by different vendors on different frameworks find each other, trade information securely and coordinate actions.¹

How two agents talk to each other

The announcement describes two roles. A client agent formulates and communicates a task. A remote agent acts on it.¹ Before that can happen, the client has to know who it is talking to. So an agent can publish an "Agent Card," a small description in JSON format of what it can do, which lets the client find a suitable agent for the job.¹

Think of it like a contractor directory. You do not hire a plumber by walking into a house and shouting. You look up who does plumbing, read what they handle, then send them a work order. The Agent Card is the listing. The task is the work order.

Who signed on

Google said more than 50 technology partners were working with the protocol at launch, naming companies such as Atlassian, Box, Cohere, Intuit, LangChain, MongoDB, PayPal, Salesforce, SAP and ServiceNow.¹ It also said it was working with partners on a production-ready version "later this year."¹ Read that as a signal, not a finished product.

What is MCP, and how is it different?

The Model Context Protocol came first. Anthropic announced it on November 25, 2024 as an open standard for secure, two-way connections between data sources and AI tools.² The problem it targets is a messy one: every new data source needed its own custom connection, which made connected systems hard to scale.²

I will keep this short, because MCP deserves its own deep dive and this piece is about the new half. The one-line version: MCP is how an agent uses tools. Your calendar, your files, your CRM, your database. One agent, many tools.

A2A in the same one line

A2A is how agents work with each other. One agent, many agents. Google says this directly: A2A complements MCP, which supplies tools and context to agents, while A2A handles the collaboration between them.¹

Why should a business owner care about agents talking to agents?

Because the way you will probably use AI in two years looks less like one chatbot and more like a small staff. One agent triages your inbox. Another prepares proposals. Another watches your ad account. Today those are separate islands. Each one has its own login, its own memory, its own rules.

The promise of a protocol like A2A is that those islands could trade work without a human copying and pasting between them. Google frames the bigger idea as agents collaborating in a multi-agent ecosystem across siloed data systems and applications.¹ If that works, the thing that stops being scarce is the connection. The thing that stays scarce is judgment.

The part nobody mentions: permissions

When agents hand work to other agents, mistakes travel too. An agent that can send email, spend money or edit your website, and that accepts instructions from another agent, needs a short leash. Google's own announcement lists "secure by default" among the design principles for A2A.¹ That is the right instinct. It is still your job to decide what each agent is allowed to touch.

What should you do about A2A and MCP right now?

Less than the hype suggests. Here is the order I would work in.

Is this the beginning of an "agent economy"?

Maybe. The pieces are lining up: models that can use tools, one open standard for tools, and now an open proposal for agents to coordinate. I am not going to tell you where it lands. I have watched plenty of platform announcements promise the moon and deliver a settings page.

What I will say is that the businesses that benefit are the ones whose house is already in order. Clean offers. Clean data. One owner for each process. If your business runs on five disconnected tools and a pile of copy and paste, adding agents just makes the mess faster. Getting that foundation right is the work I take on.

Keep building.Ekai