This is a submission for the Google I/O Writing Challenge
Remember Bitcoin?
Not the trading frenzy. Go back further — 2009, 2010, when it was just a whitepaper and a weird internet experiment that most people scrolled past. The ones who stopped, read it, and actually understood what it was trying to do were not lucky. They were early. And being early with genuine understanding is a different thing entirely from just hearing about something and moving on.
That is the energy I want you to bring to this article.
Google I/O 2026 dropped a lot. New Gemini models, Antigravity 2.0, AI glasses, a $100 subscription tier — there is genuinely a lot to explore and get excited about. But tucked inside the developer keynote was an announcement that most developers skimmed past on their way to the shinier headlines. It is called WebMCP. It is still a proposal. And understanding it now, before it fully lands, might be one of the most useful things you do this week.
So let's talk about it. What is it exactly, and why should you as a developer even pay attention to it?
Let's start with a scenario.
The Problem Nobody Is Solving Yet
When an AI agent needs to book a flight, check your account balance, or fill out a form on your website — how does it actually do that today?
It takes a screenshot. It reads your HTML like a treasure map. It guesses where the button is, clicks it, and hopes nothing breaks. If you changed your button colour last week, the agent might fail completely. If your page loads slowly, the agent might click the wrong thing. If you added a popup, congratulations — the agent is now stuck talking to your cookie banner forever.
This is not a small problem. This is how every AI agent interacts with every website on the internet right now. Fragile. Slow. Expensive to run. And completely invisible to you as the developer — your site has no idea an agent even visited.
Before WebMCP — this is the chaos happening right now
Based on that scenario, you probably have a guess at what WebMCP is trying to fix. Whether you do or you don't, let's get into it.
So What Is WebMCP?
WebMCP is a proposed open web standard that Google announced at I/O 2026.
The idea is straightforward. Instead of making the agent guess its way through your UI, you — the developer — tell the agent exactly what your site can do.
You expose your site's capabilities as structured functions. Think of it like writing a menu for the agent. Here is how you search for flights. Here is how you select a seat. Here is how you complete a booking. The agent reads the menu, calls the right function, and the task is done. Clean. Fast. Reliable.
No more screenshots. No more DOM scraping. No more praying your redesign does not break someone's automation.
After WebMCP — clean, direct, reliable
An Analogy That Might Help
Think about what REST APIs did for mobile apps.
Before clean API standards took hold, if you wanted your mobile app to talk to your backend, things got messy fast. Everyone did it differently. Everything broke.
Then developers started exposing clean endpoints. GET /flights. POST /booking. Suddenly any mobile app could talk to any backend that followed the standard. The whole ecosystem moved faster because of it.
WebMCP is that same moment but for AI agents and the web.
Right now every website is a mystery box to an agent. With WebMCP, your website becomes something an agent can reliably work with, the same way a mobile app reliably works with an API today.
The REST API moment happened once. This is that moment again.
Who Is Already Paying Attention
This is not a small developer experiment tucked in a GitHub repo somewhere. Booking.com, Expedia, Instacart, Shopify and Redfin are already experimenting with WebMCP.
Look at what those companies have in common. They are all sites where users come to do something — book, buy, search, transact. They are exactly the kind of sites AI agents are going to visit on behalf of users constantly over the next two years.
Big players move early for a reason. That reason is always advantage.
Where It Stands Right Now
The experimental origin trial starts in Chrome 149. Gemini in Chrome support is coming soon after.
You cannot ship a full production WebMCP implementation today and expect every agent to support it. This is early — deliberately early. And that is exactly why this conversation matters right now.
Go back to Bitcoin for a second. The people who won big were not just the ones who heard about it first. They were the ones who understood it early enough to be ready when the moment came. They had already done the thinking. When it mattered, they moved fast.
That is the position you want to be in with WebMCP. Not scrambling when every framework has adopted it and every job description lists it. Ready.
What You Should Be Doing Right Now
You do not need to drop everything today. But here is how to think about it.
If you are building a web app where users complete tasks — booking, buying, submitting, searching — start asking yourself:
- What are the three or four things an agent should be able to do on my site?
- What data does it need?
- What actions should it be able to take?
That thinking costs you nothing right now. But when WebMCP lands fully, the developers who already have those answers will move fast. The ones who have not thought about it will be starting from zero.
The Bigger Picture
Every major platform shift has had a moment where early movers defined the rules.
Mobile had it. REST APIs had it. This is that moment for the agentic web.
Your website is about to have two kinds of visitors — humans who see your UI, and agents who need your functions. WebMCP is how you serve both. Right now, most websites are built for only one of them.
The web is not going away. But how people and agents navigate it is changing fast. Understanding WebMCP now is not about being trendy. It is about not being caught off guard when it arrives.
You read this far. You already understood the assignment.
Top comments (0)