What's this about?
A proposal, with a reference site, that web servers should return clean Markdown when the requester asks for it via an HTTP Accept header — so an agent fetching a page gets the content rather than a pile of layout and scripts.
What exactly is content negotiation?
It is a feature HTTP has always had: the client says what formats it can handle, and the server picks. It has been used for decades to serve different image formats or languages. The argument here is that agents are a new client type with a genuinely different need, and the mechanism to serve them already exists — no new standard required, just using the old one.
What's the community saying?
HN 96 points, 57 comments
- Appreciation for reusing an existing mechanism rather than inventing a standard, contrasted favourably with heavier proposals.
- The practical objection: the incentive is backwards, since many sites do not want agents reading them cleanly and cheaply.
- It sits alongside the WebMCP items as one of two competing answers to "how should the web expose itself to machines" — this one is minimal, WebMCP is structural.