TWO CAPTURES. TRACEABLE EVIDENCE.
Markdown Response Inspector.
Compare HTML and Markdown response captures before handing an agent-friendly page to your team. Every finding stays tied to the evidence you provide.
Bring your response captures
Sample data: this is not your website’s result. Clear all before starting your own capture.
HTML-request capture
Paste only response headers, or import a UTF-8 text file. Multiple status blocks are supported.
User-supplied context, never fetched. Matching uses the full URL; exports omit credentials, query and fragment.
Markdown-request capture
Paste only response headers, or import a UTF-8 text file. Multiple status blocks are supported.
User-supplied context, never fetched. Matching uses the full URL; exports omit credentials, query and fragment.
200 KiB each · 2,048 lines · 20 response blocks
Paste at least one capture to begin.
What the supplied evidence shows
Evidence and next checks
Take a focused report
Reports include only Content-Type, Vary, Cache-Control and derived findings. Other headers, including cookies and authorization, are omitted. This is not a guarantee that allowed values are secret-free: review before sharing.
Capture the response, not a guess
Run these examples yourself after replacing the example URL. They issue GET requests, follow HTTPS redirects, save response headers, and discard the downloaded body. Record each printed final URL and the request method / Accept values above.
curl --silent --show-error --location --max-redirs 5 --proto '=https' --proto-redir '=https' --dump-header html.headers --output /dev/null --header 'Accept: text/html' --write-out 'Final URL: %{url_effective}\nHTTP status: %{http_code}\n' -- 'https://example.com/docs'
curl --silent --show-error --location --max-redirs 5 --proto '=https' --proto-redir '=https' --dump-header markdown.headers --output /dev/null --header 'Accept: text/markdown' --write-out 'Final URL: %{url_effective}\nHTTP status: %{http_code}\n' -- 'https://example.com/docs'Do not use curl -I (HEAD) as proof of a GET body. Do not add credentials or bypass TLS warnings. Do not paste response bodies or verbose curl logs.
A narrow check, with useful boundaries
Does this test whether agents can access my website?
No. It only analyzes the headers and context you supplied. It does not contact the URL, check robots.txt, authenticate, render Markdown, verify body content, or predict indexing and citations. HTML remains a useful representation for many agents.
Is Vary: Accept always required?
HTTP guidance is qualified. Different Accept-selected representations often need Vary: Accept for cache reuse. Missing it is a review signal, not a universal failure. Vary: * has separate semantics; ordinary no-store prohibits storage, while private and no-cache do not mean no storage. Qualified directives and must-understand need separate interpretation.
Are token-count headers, Content-Signal or MCP cards required?
No. This inspector does not score them. Cloudflare stopped generating its two token-count headers in October 2026; Content-Signal is separate optional contextual metadata. MCP discovery is outside this tool’s scope.
Why can the comparison remain incomplete?
The strongest finding needs ordinary HTTP 200 captures, HTML and Markdown media types, and your recorded GET / exact Accept / matching final-URL context. Even then it is consistent with negotiation, not proof that Accept alone caused the change. HEAD, redirects, partial or conditional responses and ambiguous fields remain limited evidence.