We built a free llms.txt validator — open-source, no signup, MIT-licensed

Live today at bridgetoagent.com/tools/llms-txt-validator. Also on npm as @bridgetoagent-com/llms-txt-validator and on GitHub at github.com/bridgetoagent/llms-txt-validator. MIT license. No telemetry. No external dependencies beyond fetch.

Why

We catalogued six failure modes that show up in roughly 80% of free-generator llms.txt output last week — link rot, hallucinated URLs, malformed Markdown, missing sections, generic placeholders. The post ended with "run a 30-second spot-check before you deploy." That post should have ended with "run this tool." We built it.

Three audiences this is for:

1. Anyone who used a free generator and wants to verify the file before shipping.
2. Developers writing llms.txt by hand who want a CI-runnable lint instead of eyeballing the spec.
3. AI agent / spec community who want a reference implementation of the llmstxt.org spec as a parser they can fork, audit, or wrap.

What it checks

Three layers, top to bottom by enforcement strictness:

1 · Structure (always on)

• # Title is present and is the first non-blank content
• No duplicate H1 headings
• Optional > blockquote description captured immediately after the title
• ## Section headings used for resource groups (H2 level)
• H3+ headings flagged as info (uncommon in llms.txt)
• Empty sections flagged

2 · Link bullets

• - [text](url) or - [text](url): description syntax — anything else inside a section's bullet list flagged as malformed-link
• Empty link text or URL
• Relative URLs (/docs/foo) — llms.txt is consumed by external agents, must be absolute
• Fragment-only URLs (#anchor) — meaningless to agents
• mailto: URLs flagged as unusual
• http:// URLs flagged (prefer https:// so agents that refuse insecure fetches don't drop the link)
• Duplicate URLs as a warning, duplicate link text as info

3 · Reachability (opt-in)

Check the box, or pass --check-links on the CLI, to actually request every link:

• Bounded-concurrency HEAD requests (8 in parallel by default)
• Falls back to GET when HEAD returns 405 or 501
• link-non-2xx for 4xx/5xx responses
• link-unreachable for network errors and timeouts (5 second default)
• link-slow warning for links above the 3-second threshold

User-agent: bridgetoagent-llms-txt-validator/0.1 (+https://github.com/bridgetoagent/llms-txt-validator) — identifies itself so server logs aren't anonymous.

How to use it

Hosted

Paste, upload, or fetch by URL at /tools/llms-txt-validator. Result renders inline. No data leaves the validation request — we don't store the content you paste.

CLI

# Local file
npx @bridgetoagent-com/llms-txt-validator ./public/llms.txt

# Fetch from URL
npx @bridgetoagent-com/llms-txt-validator https://example.com/llms.txt

# With reachability check
npx @bridgetoagent-com/llms-txt-validator ./public/llms.txt --check-links

# Machine-readable for CI
npx @bridgetoagent-com/llms-txt-validator ./public/llms.txt --json > report.json

Exit codes: 0 pass, 1 warnings only, 2 errors. Drop it into CI:

- name: Validate llms.txt
  run: npx @bridgetoagent-com/llms-txt-validator ./public/llms.txt --check-links

Library

import { validate } from "@bridgetoagent-com/llms-txt-validator";

const source = await fs.readFile("./llms.txt", "utf8");
const report = await validate(source, { checkReachability: true });

console.log(report.status);   // "pass" | "pass_with_warnings" | "fail"
console.log(report.issues);   // [{ severity, code, message, line, ... }]
console.log(report.parsed);   // parsed document tree

Use parse(source) if you just want the parsed document tree without validation — useful if you want to write your own custom checks downstream.

How this differs from the homepage audit

The homepage audit (bridgetoagent.com) is a lead-gen tool. It captures email, runs against five audits in 90 seconds, and routes you toward the $49 kit.

This validator is the opposite shape: no email, no signup, no time limit, no recommendation. You paste a file, you get a report. It does one thing — validate llms.txt content — and exposes the result as both UI and machine-readable JSON.

The two share validation logic. The same engine runs inside the kit's generator before every output ships, so customers don't get malformed files. We extracted it, added a CLI + npm package wrapper, hardened it for arbitrary input, and made it public.

Issue codes

Stable identifiers for every kind of finding. Pin on these in CI or downstream tooling — we don't rename them once they're shipped.

Code	Severity	Meaning
missing-title	error	No # Title heading found
title-not-first	error	Content appears before the title
title-not-h1	error	First heading is not H1
duplicate-title	warning	Multiple H1 headings
section-wrong-level	info	H3+ heading where H2 is conventional
empty-section	warning	Section has no link bullets
malformed-link	error	Bullet line is not a valid Markdown link
link-missing-url	error	Link [text]() has no URL
link-empty-text	warning	Link [](url) has no display text
link-relative-url	warning	Root-relative URL — must be absolute
link-hash-only	warning	Fragment-only URL
link-mailto	info	mailto: URL — unusual
link-non-https	warning	http:// — prefer https://
duplicate-url	warning	Same URL appears more than once
duplicate-link-text	info	Same link text appears more than once
link-unreachable	error	Network error or timeout
link-non-2xx	error	HTTP 4xx or 5xx response
link-slow	warning	Response above slow threshold
no-content-after-title	warning	Title exists but nothing else
trailing-whitespace	info	Line has trailing whitespace
tabs-instead-of-spaces	info	Line contains tab characters

The full list lives in the README.

What it deliberately doesn't do

A list of things people will ask for and our reasoning for not shipping them:

• Score the content quality. We can verify the parser-level conformance and link health. We can't verify that the descriptions are good or that the sections are useful without making a judgment call. That's a different tool — one that the BridgeToAgent kit does because it can read your real DOM to ground the answer.
• Auto-fix the file. Validators report. Fixers rewrite. They're different products with different risk profiles. We may add a --fix flag for safe rewrites (trailing whitespace, http→https) in a future version. Anything beyond that is the kit's territory.
• Crawl the URLs. Reachability check verifies the URLs resolve. It doesn't verify the linked content is what you said it is. We considered MIME-type assertions but decided they'd add noise without proportional signal — llms.txt doesn't constrain what type of resource each link points at.

What's next

Two more tools from the same track ship in the coming weeks:

• agents.json validator (target: week 6 of the Q3 campaign) — schema conformance for the agents.json spec, typed-parameter completeness check that maps directly to the Lighthouse agents-json-actions-typed audit, endpoint reachability.
• Standalone Lighthouse Agentic scorer (target: week 9) — runs all 9 Lighthouse Agentic Browsing audits against a URL, gives per-audit pass/fail + score breakdown, no signup. Will use the validators above as primitives.

If you have a real-world llms.txt file we miss-parse, open an issue — that's the fastest way to make the validator better.

Related reading

• Chrome just made Agentic Browsing default in Lighthouse — why this validator matters more right now: Lighthouse 13.3.0 (May 7, 2026) moved Agentic Browsing into the default config, so llms-txt-well-formed is now an audit every PageSpeed Insights run returns
• What free llms.txt generators don't tell you — the failure modes this validator catches
• llms.txt skepticism, half right — context on why the spec is worth taking seriously despite the critics
• llms.txt vs robots.txt — how the two files relate
• Every Lighthouse Agentic Browsing audit, every fix — the llms-txt-well-formed audit this validator helps you pass