How WebMCP/Site Tools currently work in ChatGPT Desktop

  1. When a page exposes a WebMCP tool that matches the user’s intent, does ChatGPT prefer WebMCP over regular browser interaction (DOM/screenshots), or is tool selection decided dynamically?

  2. If a user asks about a WebMCP-enabled site but does not already have that page open, can ChatGPT automatically navigate to the site in the built-in browser to discover and invoke its WebMCP tools, or must the user open the page first?

This topic was automatically closed after 24 hours. New replies are no longer allowed.