Generate a spec-correct llms.txt and llms-full.txt for any site, then publish them at your site root.
Free, no login, no email. Nothing stored beyond a 24 hour cache. Markdown copy · API and MCP
Paste your site URL
We read the homepage, then robots.txt, then the sitemaps it points at. Sites without a sitemap fall back to the links on the homepage.
We build the inventory
Discovered pages are grouped into sections from their URL paths, and titled from their slugs. Top level pages sit in a section called Core.
Copy or download both files
llms.txt is the index. llms-full.txt is the same site with page content converted to Markdown and inlined. Each has its own copy and download button.
It is a proposal, published at llmstxt.org, for a plain text file at your site root that tells a language model what your site contains. The format is deliberately small: one H1 with the site name, a blockquote summary, then H2 sections of Markdown links. A companion file, llms-full.txt, inlines the page content so a model can read the site without fetching anything.
The idea it borrows from is robots.txt, which everyone understands. The difference matters though. robots.txt is honoured by every major crawler because search engines agreed to honour it over twenty five years. llms.txt has no such agreement behind it.
No major AI provider has publicly committed to reading llms.txt. Publishing one does not get you into training data, does not get you cited in an answer, and does not rank you anywhere. Anyone selling it as an AI SEO technique is well ahead of the evidence. Some documentation platforms and AI coding tools do look for the file, which is a real if narrow audience.
The honest case for publishing one is cheapness against odds, not certainty. The file takes minutes to produce and cannot harm you: it does not change how your site renders, does not affect search, and does not expose anything that is not already public. If the convention gains adoption you already have one. If it does not, you spent five minutes. That is a reasonable bet, and it is a different claim from the one usually made about it.
Writing an index of your own site is a mechanical job. The titles are already your titles. The sections are already your URL paths. The content is already your content. Putting a language model in the middle of that would add cost, add latency, and add the one thing a site index must not have: variation between runs, and the possibility of a sentence about your site that nobody at your company wrote.
So there is no model here. The output is a deterministic transform of your own inventory, which means two runs over an unchanged site produce the same bytes, and every line in the file traces back to something your site already said. Read it before you publish it, reorder the sections so the pages you care about come first, and delete anything that is noise. It is your file.
A proposed convention, published at llmstxt.org, for a plain text file at your site root that tells a language model what your site contains. The format is small: one H1 with the site name, a blockquote summary, then H2 sections of Markdown links. A companion file, llms-full.txt, inlines the page content so a model can read the site without fetching each page.
The same run this page does, as an HTTP endpoint and as the MCP tool generate_llms_txt. No account, no API key. Full reference.
curl -sS https://usesuperflow.ai/api/tools/llms-txt-generator \
-H 'Content-Type: application/json' \
-d '{"url":"example.com"}'Returns { ok, report: { siteName, llmsTxt, llmsFullTxt, pagesDiscovered, pagesIncluded, truncated }, cached, ageSeconds }. 10 runs per hour per IP. Allow up to 75 seconds.
claude mcp add --transport http superflow https://usesuperflow.ai/api/mcpThen ask your agent for generate_llms_txt. Setup for Claude Desktop, Cursor and VS Code is on the reference page.
No login, no email, no ads. Same as this one.
Superflow agents watch every page you ship and tell you when the content changes, so the picture an AI has of your site does not quietly go stale.