You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 7174152
Browse filesBrowse the repository at this point in the historyBrowse files
ChatGPT Web's fetcher can read /mcp/ but not /skill.md or /mcp.md. The
only difference between them was the content type: text/html versus
text/markdown. text/markdown is a registered media type but most
non-browser fetchers have no renderer for it and refuse the document,
and because we send nosniff there is no fallback for the client either.
Markdown is plain text; text/plain loses nothing and is read inline
everywhere. GitHub raw serves .md exactly this way, which is why LLM
fetchers can read README files at all.
The headers were copied across the worker and both Astro routes, so
they now come from one place that carries the reason. Three locks keep
it from drifting back: the source check in the site tests, and the live
contract auditor, which now measures the content type in production and
covers mcp.md at all — it did not before.
Also rename the connector label on /mcp to Equinox Orbit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
0 commit comments