Markdown Twin
A clean markdown version of an HTML page, served from the same URL via content negotiation when a client requests text/markdown.
- term
- Markdown Twin
- category
- protocols
- short_def
- A clean markdown version of an HTML page, served from the same URL via content negotiation when a client requests text/markdown.
- long_def
- Markdown carries the meaning of a page in roughly a tenth of the tokens of the equivalent HTML. Agent fetchers (Claude Code's WebFetch, Cursor, Cloudflare's edge) request it with an Accept: text/markdown header; humans still get the styled HTML.
- see_also
content-negotiationllms-txttoken-economics- etymology_origin
- — verify-against-primary-at-build ↗ https://datatracker.ietf.org/doc/html/rfc7763 — 'markdown twin' is an agentic-web practitioner term with no single coiner/standards body; text/markdown is registered in RFC 7763 (2016)
- related_to
content-negotiationllms-txttoken-economicsrag- contrast_with
- Unlike a separate /page.md URL, a markdown twin lives at the SAME URL as the HTML and is selected by the Accept header via content negotiation — one canonical URL, two representations.
- example
- Claude Code's WebFetch and Cursor request the markdown representation with an Accept: text/markdown header, receiving roughly a tenth of the tokens of the HTML.
- source
- https://datatracker.ietf.org/doc/html/rfc7763
- status
- emerging
- why_it_matters
- The markdown twin is the canonical way to serve agents a cheap, clean, low-token version of a page without maintaining a duplicate site.
- sameAs
- —
- bridge_entity
- agent-readiness
- last_verified
- 2026-06-15
- md_twin
- /glossary/markdown-twin.md