On April 9, 2025, Google published a post titled "Announcing the Agent2Agent Protocol (A2A)" and, with it, a new open protocol for AI agents. Most of the coverage filed it under developer news. The plain reading is bigger: agents built by different companies now have an agreed way to find each other and coordinate work, and that quietly changes what a website has to do to stay reachable.
What Google announced
Google announced A2A, an open protocol that lets AI agents talk to each other across different vendors and frameworks. The announcement at developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability arrived with support and contributions from more than 50 technology partners, names like Atlassian, Box, Intuit, PayPal, Salesforce, SAP, and ServiceNow among them. The premise is simple: an agent your software vendor builds and an agent someone else builds should be able to exchange information and coordinate a task without a developer hand-wiring every possible pair.
A2A does not compete with the other agent standard already in circulation. Google is explicit about how the two fit together:
A2A is an open protocol that complements Anthropic's Model Context Protocol (MCP), which provides helpful tools and context to agents.
Under the hood the protocol is deliberately ordinary, and that is the point. It runs on standards businesses already use every day: HTTP, SSE, and JSON-RPC. The one new artifact worth learning by name is the Agent Card.
Agents can advertise their capabilities using an “Agent Card” in JSON format, allowing the client agent to identify the best agent that can perform a task and leverage A2A to communicate with the remote agent.
In plain English
Picture two agents. One is a "client" agent working for a person: book the table, source the candidate, reconcile the invoice. The other is a "remote" agent that holds a capability the first one needs. A2A is the etiquette they use to introduce themselves, agree on a task, and pass results back.
The Agent Card is the introduction. It is a small JSON file that states, in terms software can read, what an agent is called, what it can do, and where to reach it. A client agent reads the card, decides this is the right counterpart for the job, and opens a conversation. Nobody sat down in advance to connect those two systems.
Once the conversation starts, the work is organized as a task with a lifecycle. A quick request finishes in one exchange. A long one, the kind that runs for hours with a human checking in, stays in sync through status updates until it produces what A2A calls an artifact, the finished output. That is the whole shift in a sentence: discovering a capability stops being something a developer hard-codes and becomes something agents do at runtime by reading published cards.
Why this matters for your business
Today your customers reach you through a screen. They read your pages, fill your forms, click your buttons. A2A describes a second channel forming beside that one, where the visitor is software acting on a customer's behalf, and it arrives expecting a card and a conversation rather than a landing page.
Two things follow. Being reachable by agents is a different job from being readable by people, and almost no website does the first one yet. And because A2A is an open standard rather than one company's walled system, the agent that comes knocking could be built by anyone who follows it. You describe your business once, in the shared format, and every compliant agent can understand you.
The partner list carries more weight than it first appears. When more than fifty platforms your business already touches agree on a single protocol on day one, the format stops looking like a bet and starts looking like the default your tools will soon assume.
Where Trinzik fits
This protocol family is what we build websites around. When we deliver a site, we treat the agent channel as a real surface, not an afterthought: a published Agent Card that states what the business offers, and a site agent grounded in the business's own content so it answers accurately instead of improvising. Our Apex Domain Agents work exists to make a domain legible to exactly the kind of agents A2A is built to carry, so that when one arrives on a customer's behalf, there is something there to talk to.
We build it this way now because the alternative ages fast. A website made only for human eyes is invisible on the channel A2A opens, and bolting that channel on later is harder than designing it in from the start.
If you want to see what an agent actually finds when it reaches your domain, start a conversation with us and we will walk you through it.
Questions this raises
Does A2A replace Anthropic's MCP?
No, A2A does not replace MCP. Google describes A2A as an open protocol that complements Anthropic's Model Context Protocol, which supplies tools and context to a single agent. A2A handles the layer above that: letting separate agents, potentially built by different vendors, discover each other and coordinate a task. The two protocols are designed to work together rather than compete.
What is an Agent Card in A2A?
An Agent Card is how an A2A agent advertises itself. It is a JSON description of the agent's capabilities that a client agent reads to identify which agent can perform a task and where to reach it. Capability discovery happens by reading these cards at runtime, so a new agent can be added without a developer wiring the connection in advance.
How many partners launched with A2A?
A2A launched with support and contributions from more than 50 technology partners, according to Google's April 2025 announcement. The group spans technology platforms and service providers, among them Atlassian, Box, Intuit, PayPal, Salesforce, SAP, and ServiceNow. Because A2A is built on existing standards like HTTP, SSE, and JSON-RPC, those partners can integrate it with IT stacks they already run.
Sources
- Announcing the Agent2Agent Protocol (A2A) · Google for Developers Blog, April 9, 2025
- A2A draft specification on GitHub
- A2A code samples and example scenarios