RSC payloads. Script tags. Font preloads. All to render a title, a price, and a description.
The fix is already built into HTTP: content negotiation. Check the Accept header, respond accordingly. JSON for apps, XML for legacy systems, Markdown for LLMs. Backend developers do this all the time - but Next.js has no built-in way to handle it.
I built next-md-negotiate to close that gap.
It lets you define Markdown versions of your pages alongside your existing routes. Requests with `Accept: text/markdown` get transparently rewritten to your Markdown definitions. Browsers still get your normal HTML. Same URL. No duplicate routing.
Browser -> GET /products/42 Accept: text/html -> normal Next.js page
LLM agent -> GET /products/42 Accept: text/markdown -> clean Markdown
257x smaller payload. Fewer tokens in, better context out.
This also has a massive impact on crawl budget. Markdown responses are tiny compared to full HTML, so AI crawlers can index far more pages in the same amount of time. For e-commerce sites with thousands of product pages, this is critical - it can mean the difference between an AI agent seeing a fraction of your catalog vs nearly all of it.
Works with both App Router and Pages Router. Type-safe route params. Plugs into existing middleware or uses Next.js rewrites.
npm install next-md-negotiate
npx next-md-negotiate init