🩺 docsanity

One command for docs-site health — broken links, orphan pages, SEO, readability & structure in a single score.

Your docs site looks fine — until a reader hits a 404 from a link you moved, lands on a page Google shows with a duplicate title, or bounces off a wall of grad-school prose. These problems live between files (which page links where, which titles repeat, what's orphaned), so single-file linters miss them and no one notices until it ships.

docsanity audits a whole docs directory in one pass and unifies four health dimensions into one score and one report — 100% locally, no API key:

• 🔗 Links & references — broken relative links, dead #anchors, missing images, and orphan pages nothing links to.
• 🔎 SEO & frontmatter — missing/oversized title & description, plus site-wide duplicate titles and descriptions across every page.
• 📖 Readability — a per-page reading grade; flags pages that read too hard.
• ♿ Structure & a11y — single H1, no skipped heading levels, image alt text, fenced code blocks that declare a language.

npx docsanity scan ./docs

broken.md  50/100 (F)
  ✗ L5     Link target not found: "./nope.md"            links.broken-file-link
  ✗ L5     Anchor "#does-not-exist" not found in index.md links.broken-cross-anchor
  ⚠ L7     Image has no alt text: ./diagram.png           structure.image-alt
orphan.md  88/100 (B)
  ⚠        Orphan page — nothing links here               links.orphan-document
  ⚠        Duplicate title — also used by 1 other page(s) seo.duplicate-title

Overall  82/100 (B) · 5 page(s), 4 error(s), 10 warning(s)

Why a suite (and not four separate tools)?

The valuable checks need the whole-site view, which only a unified pass can give you:

• Orphan pages and broken cross-file anchors require the full page graph.
• Duplicate titles/descriptions require comparing every page to every other.
• A single score + grade you can gate in CI means one number to watch, not four tools to wire up and reconcile.

docsanity builds the page graph once and runs every dimension over it, then folds everything into a per-page and per-dimension report. Under the hood it leverages the focused engines linklint (link graph) and readlevel (readability), and adds the SEO-frontmatter and structure checks that tie them into a docs-site audit.

Why not just ask an LLM to "check my docs"? It can't hold 300 interlinked files in context, it can't run in CI deterministically, and "did this link break" is a graph lookup, not a guess.

Install

# run it now, no install
npx docsanity scan ./docs

# or add it
npm install -g docsanity      # global CLI
npm install -D docsanity      # CI dependency

Node ≥ 18. Works with Docusaurus, Astro, Nextra, Hugo, Jekyll, VitePress — any Markdown/MDX docs tree.

Quick start

docsanity scan ./docs                       # audit a folder
docsanity scan ./docs --min-score 85        # CI gate: fail under 85
docsanity scan ./docs --md docs-report.md   # write a Markdown report
docsanity scan ./docs --disable readability # turn a dimension off
docsanity init                              # write docsanity.config.json

See examples/sample-report.md and examples/sample-report.json for full reports.

Real scenarios

1. CI gate for your docs. A PR that moves a page and forgets to fix the inbound links — or ships a new page with no description — fails the build:

# .github/workflows/docs.yml
- run: npx docsanity scan ./docs --min-score 90 --md docs-report.md

2. Pre-launch audit of an inherited site. Point it at a docs tree you didn't write and instantly see the orphans, duplicate titles and dead anchors that accumulated over time — one report, ranked by page score.

3. Keep AI answer engines happy. Clear, well-structured, well-linked pages are what answer engines quote. Fix the readability and structure findings to make your docs the source they cite.

What each dimension checks

| Dimension | Checks | | --------- | ------ | | Links & references | broken relative links, dead #anchors, broken cross-file anchors, missing images, undefined references, orphan pages | | SEO & frontmatter | required title/description, length windows, H1-as-title fallback, site-wide duplicate titles & descriptions | | Readability | Flesch–Kincaid / Gunning Fog / SMOG average grade per page, flagged against a configurable ceiling | | Structure & a11y | single H1, no skipped heading levels, image alt text, code-fence language tags |

Scoring is transparent: each finding is a weighted error / warning / info; pages roll up to a 0–100 score and an A–F grade, averaged into the overall.

Configuration

docsanity init writes docsanity.config.json:

{
  "extensions": [".md", ".mdx", ".markdown"],
  "requireFrontmatter": ["title", "description"],
  "descriptionMin": 50,
  "descriptionMax": 160,
  "titleMax": 60,
  "maxGrade": 14,
  "minWordsForReadability": 80,
  "disable": [],            // e.g. ["readability"]
  "ignore": [],             // rule ids, e.g. ["structure.code-language"]
  "minScore": 0             // CI gate threshold
}

Library API

import { analyzeDocs } from "docsanity";

const report = analyzeDocs(root, inputs, config, {
  version: "1.0.0",
  generatedAt: new Date().toISOString(),
});
for (const page of report.pages) {
  console.log(page.path, page.score, page.findings.length);
}

Also exported: parsePage, parseFrontmatter, parseMarkdown, seoChecks, structureChecks, readabilityCheck, linkFindings, DEFAULT_CONFIG, and types.

Roadmap

• More frontmatter conventions (Hugo/Jekyll specifics, custom required schemas).
• HTML docs input (not just Markdown/MDX).
• Per-rule severity overrides and inline ignore comments.
• --baseline to fail only on new issues vs a committed snapshot.
• A web playground (drop a zip of docs, get the report — nothing uploaded).

💖 Sponsor

docsanity is free and MIT-licensed, built and maintained in spare time. If it caught a broken link before your readers did, please consider supporting it:

• ⭐ Star this repo — the simplest free way to help others find it.
• 🍋 Sponsor via Lemon Squeezy — one-time or recurring.

License

MIT © docsanity contributors