AccessScan

AccessScan is a free, automated accessibility checker for the EU Accessibility Act (EAA) and WCAG 2.x. It is a local-first tool: node server.js serves a self-contained web UI and JSON API bound to localhost. A public hosted instance is also maintained — a static frontend on Vercel proxies /api/* to a Render-hosted backend.

Honest disclaimer: automated scans detect roughly 30–57% of WCAG issues (per Deque's research). A clean scan is not a compliance guarantee, and EAA enforcement is live in the EU. Use AccessScan as a fast first pass; obtain a manual audit for real assurance.

Features

• Paste a URL, receive an automated WCAG 2.x A/AA report via headless Chromium and axe-core
• Optional same-origin crawl — up to 10 pages total (submitted URL + up to 9 linked pages), two pages scanned concurrently, ~3-minute wall-clock cap per scan
• Results re-openable in memory; export as self-contained HTML report, PDF, or JSON
• Scan comparison — diff a finished scan against the most recent finished scan of the same URL (addedRules, fixedRules, severityChanges)
• SSRF protection: private/internal IPs, local hostnames, and redirect hops are blocked before the browser connects
• Rate limiting on scan creation; polling and export endpoints are not limited

Quickstart (local)

Requirements: Node.js >= 18.

npm install        # installs deps + downloads Chromium (~170MB first time)
npm start          # serves on http://localhost:3000

• PORT overrides the port (default 3000); HOST overrides the bind address (default 127.0.0.1).
• If the browser download was skipped, run npx playwright install chromium once.
• Open http://localhost:3000, paste a URL, scan. No accounts, no database, no tracking.

API

All endpoints are JSON unless noted. Polling endpoints return the current scan state.

Method	Path	Description
POST	/api/scan	Body { "url": "https://…", "crawl": false, "maxPages": 5 } → 202 { id, status: "running" }. 429 if a scan is busy or rate-limited, 400 on invalid/private URL or non-numeric maxPages. maxPages is total pages including the submitted URL, clamped 1–10, default 6.
GET	/api/scan/:id	{ status, phase, result | error }
GET	/api/scan/:id/export?format=html|json|pdf	Downloadable report (409 until the scan finishes)
GET	/api/scans/:id/compare	Diff vs. most recent other finished scan of the same normalized URL (origin+pathname, query/hash ignored): { comparable, previousId, addedRules, fixedRules, severityChanges, totalsBefore, totalsAfter } or { comparable: false }
GET	/api/health	{ ok, busy, playwright }

Rate limits

Rate limiting applies only to POST /api/scan (accepted scan creations). Polling (GET /api/scan/:id, exports, compare) is deliberately unlimited — one scan legitimately generates many polls.

• Per-IP: 10 accepted scans per 10-minute sliding window (default)
• Global cap: 60 accepted scans per window across all IPs
• Excess requests get 429 with a Retry-After header
• Tunables via env vars: RATE_LIMIT_MAX, RATE_LIMIT_WINDOW_MS (ms), RATE_LIMIT_GLOBAL_MAX

In production the server runs with trust proxy = 2 (two hops: client → Vercel edge → Render load balancer), so req.ip resolves to the real client IP and buckets stay per-user. Known caveat: a direct-to-Render attacker could spoof X-Forwarded-For to rotate per-IP buckets, but the global cap still bounds total load.

Environment variables

Variable	Default	Purpose
PORT	3000	Port the Express server listens on
HOST	127.0.0.1	Bind address. Render sets HOST=0.0.0.0; locally leave the default
PROXY_SECRET	unset	Shared secret for the Vercel→Render proxy. When set, every request must carry a matching x-proxy-key header or it gets a bare 403 — the Render URL becomes unusable directly. GET /api/health is exempt (Render's health check can't send headers; it only returns {ok,busy,playwright}). Must be identical on Vercel and Render
RATE_LIMIT_MAX	10	Accepted scans per IP per window
RATE_LIMIT_WINDOW_MS	600000	Rate-limit window length (ms)
RATE_LIMIT_GLOBAL_MAX	60	Accepted scans globally per window

PROXY_SECRET and the RATE_LIMIT_* variables are unset locally, so node server.js dev is unaffected.

Deployment architecture

AccessScan is local-first by design — node server.js is meant to run on your own machine. The hosted public instance uses a split architecture because Playwright scans cannot run on Vercel serverless:

browser ──► Vercel (static frontend, deploy root = public/)
              public/vercel.json rewrites /api/* → serverless function
              public/api/index.js ── fetch + x-proxy-key: PROXY_SECRET ──►
Render (accessscan-9gcz.onrender.com, node server.js, HOST=0.0.0.0)
              gated: all requests except GET /api/health require x-proxy-key

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗