WebMCP landed in ChatGPT today Websites expose tools, your agent can use them, or you can just use the app like a human. Probably both! I'm calling it Clicks-n-clankers™

Honestly I don’t see the benefit of WebMCP over /.well-known/mcp, feels silly loading a human built interface just to know how an agent should interact with it

WebMCP makes websites more legible to agents without hiding the human interface.

huge potential. right now obviously it is limiting to need a special chrome flag or to run your website in Codex desktop. :\

sounds like a fun way to interact with web tools. curious how it all works in practice

I think it’s a good way of making new and existing web apps accessible for agents within off-the-shelf harnesses. Complimentary to that though, I’m most excited about projects like @eve that give you more control over your own harness. So much more power where you can integrate with APIs directly

So what's the difference with playwright mcp or computer use ?

😂good one

Using Cloudflare Pages and GitHub MCP is magical. It just changes everything it needs almost instantly.

WebMCP is a big deal—agents finally get a standard way to interact with web tools. Curious how it handles auth and state across sessions, or is it mostly stateless?

webmcp's integration with chatgpt is huge, but let's be real: most ppl still don't know what they're doing with their agents. webmcp's just making the noise a little louder

Clicks-n-clankers lmao. First thing I thought was stream overlays and chat tools just running themselves while Im live. Websites exposing tools to the agent is actually sick.

Nice part is the page registers tools through navigator.modelContext, so the agent acts inside your logged-in session already.

now supports WebMCP as well as traditional MCP, with the pour engine built inside too :)

cool, 但我第一反应是 auth 和权限边界：网站暴露 tools 之后，怎么防止 agent 被诱导执行意外操作？是不是还是得保留一层 human-in-the-loop 确认？想听你实际用下来的体感

"Probably both" is the part that worries me. Two surfaces, versioned separately, and only one has users who file bugs. Agent hits a stale tool, retries, falls back to clicking. Task still completes. The tool path can rot for months and nothing tells you.

Clicks n clankers is actually a great name lol

recipe websites 🤣

does webmcp define retry semantics because partial tool success will break production flows

big for shopping. millions of Shopify stores becoming agent-ready overnight is such a huge shift

Add Kitesurf from Cloudflare on top of the webmcp Wes, and let the magic kick in.

Would love your thoughts on