Is WebMCP worth paying attention to yet?
I keep seeing WebMCP mentioned as a way for websites to offer tools directly to AI agents in the browser. Is anyone actually using it, and should I plan around it now or keep using browser automation for agents?
WebMCP is a proposal for web pages to declare tools that an agent in the browser can call directly, instead of the agent reading the page and clicking buttons. The idea is sound: a site that says "here is a search tool, here is an add-to-cart tool" is faster and less fragile than an agent guessing from screenshots or page structure.
It is still early. Browser and site support are limited and the details are still moving, so very few sites you would want an agent to use expose tools this way today. Treat it as something to follow and experiment with, not something to build a product on.
If you run a site, the cheap way to prepare is to keep clear, stable actions behind your interface, the same ones you would expose through an API. Those map directly onto tools if WebMCP or something like it takes off.
If you need agents working on websites now, browser automation servers are the practical route. They let the model open pages, read them through the accessibility tree and act on them, and they work on any site without the site doing anything.
Listings mentioned
- Playwright MCP · agent by microsoftMicrosoft's browser automation server; the agent reads pages through the accessibility tree and clicks, types and fills forms on any site today.
- Chrome devtools MCP · agent by ChromeDevToolsDrives a real Chrome with DevTools access, useful for testing how your own site behaves when an agent uses it.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.