If you’ve ever watched an AI agent “use” a website by guessing which button is “Checkout” and where the “Passenger Name” goes… you already know the...
For further actions, you may consider blocking this person and/or reporting abuse
For those who are searching for the flag,
The flag name is "Experimental Web Platform features"
Strange, in Canary it has its own flag.
I thought it's only on Canary build. Is this in prod chrome version?
The screenshot i shared is from canary.
didn't know about this. let me dive a little over the weekend. have you tried any agents that run in the browser (using this pattern). I know a couple of frameworks but never tried it.
I have tried only the demo website for plane tickets that is described in the article. Haven't seen any other demos. Drop here if you find them
the values on the demo is hardcoded (which you only see after trying a random query) and still it doesn't work for me somehow. Maybe it's happening because I didn't enable the flag.
They are hardcoded, but it doesn't really matter. AI agent called the Tool. it doesn't matter if it didn't find actual tickets.
But for sure you need to enable the flag. for me hardest was to find the right channel that has the right chrome version. I installed all of them, Beta, Dev, Canary. haha. worked on Canary!
This is really interesting. The "publish tools, not pixels" framing nails it.
The parallel to how we handle AI coding tools is striking. Right now with Cursor and similar tools, we give the AI context through rules files and documentation instead of letting it guess from the code. WebMCP is basically the same idea applied to browsers - give the agent structured context instead of making it reverse-engineer the UI.
The adoption challenge is real though. Getting websites to voluntarily expose structured APIs for agents to use is a chicken-and-egg problem. Sites won't add WebMCP until agents use it, and agents won't prioritize it until sites support it.
Thank you for sharing - I made a demo here: netsi.dk/experiments/webMCP.html
Works like a charm!
Good one! Really liked it.