Post

💯 I've had similar conversations with Chrome that it's hard to "sell" WebMCP without a way to actually demo the consumer UX of the feature. DevTools are great, but not sufficient to get apps involved in this.

Nice — same space we hit building Latch. What bit us: declarative <form data-mcp-tool> is great for inputs, but agents also need a structured result/error schema back, plus a polyfill so non-Chrome agent browsers can call it now. Are you handling the return contract, or the form's success page?