An llms.txt validator checks the structure and links of a Markdown file intended to guide agents toward useful website information. Screpy reads a live file or pasted text, identifies its project heading and sections, and reports malformed, repeated or unresolved links with source lines. Optional HTTP checks inspect a bounded selection of destinations.
Validate the file you actually publish
Enter a domain to check its root /llms.txt, or supply the exact file URL, including an explicit subpath such as /docs/llms.txt. In text mode, paste the Markdown and optionally provide a base URL for relative references. That base should represent the published file's address when you want the same resolution as the live file. Use llms.txt Generator if you need to prepare a draft first.
The live-file limit is 512 KiB; pasted UTF-8 text is limited to 256 KiB. Documents may contain up to 10,000 lines. The report retains at most 500 links and 1,000 findings. These are operational bounds, not limits imposed by the proposal.
Separate required structure from optional context
The llms.txt v2 proposal requires an initial H1 with the project or site name. A blockquote summary, additional explanatory text and H2 file-list sections are optional. The conventional Optional section groups secondary references; its absence is not an error. An H1-only document can therefore pass the structure check without providing a useful reading guide.
Screpy parses Markdown rather than counting heading markers inside code examples. It supports reference-style links, resolves relative addresses when context is available and flags duplicate destinations. A relative link without a base is unresolved, not confirmed broken. Review the link's section, text and source line before editing it.
Read HTTP checks within their scope
Link response checking is off by default. When enabled, the API checks the first 20 unique resolved HTTP or HTTPS destinations, ignoring fragments for the network request. It uses GET requests but stops after the response headers; it does not evaluate the target content. Each target has four seconds, the source fetch six seconds, and the overall API deadline is 25 seconds.
Restricted, failed and unavailable responses are distinct. A timeout does not prove that a page is broken for visitors. The table and CSV contain retained links, while JSON includes the bounded report. Read partial-coverage notices when a deadline or limit prevents a complete check.
A valid file does not establish AI visibility
llms.txt is a proposed information format, not an access-control mechanism. Use Robots.txt Tester or AI Bot Access Tester to investigate declared crawling rules separately. This validator does not measure citations, execute JavaScript or confirm that a particular agent reads your file. Google also says that its AI search features do not require special AI text files. See Google's AI features guidance.
For a broader source review, use AI Search Readiness Checker to inspect page HTML, robots policies and bounded discovery files together. Its technical findings do not measure AI visibility or guarantee citations.