I'm not opposed to the idea of natural language search, my point is the tools should be on the user side. Right now, I can ask questions about plain text pages that haven't been updated in 30 years directly in Firefox with no effort from the site operator. If an agent needs to have direct access to documentation, it's trivial for it to download pages autonomously (or even set up its own MCP server). There's literally no reason to demand that millions of sites independently add features that browsers and agents already have better and more uniform versions of.