Accessibility Checker

MCP serverDev tools

Scan URLs or HTML for WCAG 2.2 violations. 75-rule manifest, weighted score, shareable reports.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use Accessibility Checker to scan url

Install Accessibility Checker

The server’s own address, for the clients that take one directly. Or connect ahel once and every client you use reads it from one address, with the account kept on ahel rather than in each client’s config.

  • Claude Code

    claude mcp add --transport http --scope user accessibility-checker 'https://api.lazynext.com/mcp'

    Run it once in your project, then open /mcp to approve any sign-in the server asks for.

  • Claude Desktop

    https://api.lazynext.com/mcp

    Add a custom connector in Settings, paste this address, and approve the sign-in.

  • Cursor

    cursor://anysphere.cursor-deeplink/mcp/install?name=accessibility-checker&config=eyJ1cmwiOiJodHRwczovL2FwaS5sYXp5bmV4dC5jb20vbWNwIn0=

    Open the link and Cursor adds the server at that address.

  • ChatGPT

    https://api.lazynext.com/mcp

    In Settings, enable Developer mode, create an MCP app, and paste this address. Your plan and workspace must allow custom apps.

  • Codex

    codex mcp add accessibility-checker --url 'https://api.lazynext.com/mcp'

    Run it once, then sign in with codex mcp login accessibility-checker if the server asks for an account.

From the project's README

As published by lazynext-ai/accessibility-checker in README.md.

Real WCAG 2.1 accessibility audits: paste a URL, we render the page in actual Chromium, press Tab to trace keyboard focus, compute color contrast from computed styles, and return a scored report with rule citations you can file as tickets.

Live: https://checker.lazynext.com/ API: https://accessibility-checker.dry-hall-6a50.workers.dev

curl -X POST https://accessibility-checker.dry-hall-6a50.workers.dev/scan \
  -H 'content-type: application/json' \
  -d '{"url": "https://yoursite.com", "license": "you@example.com", "email_report": true}'
# → {"score": 70, "score_model": "weighted-v1", "issues": [{"rule": "wcag-1.3.1", ...}],
#    "rendered": true, "benchmark": {"pct": 61, "sites": 128},
#    "report": ".../report/<id>"}

The score is severity-weighted: confirmed WCAG level-A failures count in full, AA/AAA findings progressively less, warn-class heuristics at half, and repeat occurrences of the same rule at half marginal cost — a page full of minor nits outscores one with a few blocking failures.

Endpoints: POST /scan (html or url; {"url": ..., "site": true} crawls same-origin pages — 3 free / 10 Pro — and returns per-page scores; real-site scans also return benchmark — where the score ranks vs the corpus of sites scanned by this tool, shown once ≥10 sites are in), GET /checkout (trial → paid), POST /cancel (self-service cancellation by purchase email), POST /lead (email capture), GET /report/:id (shareable report — ?level=A|AA|AAA|BP, ?rule=<id>, ?by=page views; CSV/JSON/PDF exports honor them), GET /badge/:id.svg (embeddable score badge for a report), GET /rules (the coverage manifest — 74 WCAG criteria + 1 best-practice check; name, level, WCAG version, and detection path for every check emitted; Accept: text/html returns a browsable catalog, JSON is the default), GET /openapi.json (OpenAPI 3.1 contract for the whole REST surface), POST /monitor / GET /monitor?license= / DELETE /monitor (Pro daily monitoring — platform rescans each registered URL and emails a Brevo alert when a page's score drops ≥ 10 points), GET /health.

Agent surfaces

  • MCP — POST /mcp, JSON-RPC 2.0 (initialize, tools/list, tools/call; GET is 405, no SSE). Tools: scan_url, scan_html, get_report, list_rules. Add to an MCP client: {"mcpServers": {"a11y": {"url": "https://checker.lazynext.com/mcp"}}}
  • A2A — agent card at GET /.well-known/agent.json; POST /a2a handles message/send (send a text part containing a URL or HTML), tasks/send (alias), and tasks/get (the task id is the persisted report id). Scans return as completed tasks with the report artifact.
  • Widget — <script src="https://checker.lazynext.com/widget.js" data-target="#el"></script> mounts a Shadow-DOM scan box anywhere; optional data-license="buyer@x.com" for Pro.
  • Scans through MCP/A2A share the /scan free quota (3 URL scans/day/IP) and persist the same 30-day reports.

SDKs + CLI

import { AccessibilityChecker } from './sdk/js/index.js';
const a11y = new AccessibilityChecker({ license: 'buyer@x.com' }); // license optional
const { score, issues, report } = await a11y.scan({ url: 'https://example.com' });
await a11y.site('https://example.com');   // same-origin crawl
await a11y.report(id);                     // stored report object (.json export)
await a11y.reportCsv(id);                  // CSV export
await a11y.rules();                        // full coverage manifest
c := checker.New("buyer@x.com")            // license optional
res, err := c.Scan(checker.ScanOptions{URL: "https://example.com"})
rep, _ := c.Report(id)                     // stored report object (.json export)
rules, _ := c.Rules()
from lazynext_a11y import AccessibilityChecker   # sdk/python, stdlib only
c = AccessibilityChecker(license="buyer@x.com")  # license optional
res = c.scan(url="https://example.com")
lazynext-a11y scan https://example.com --site    # bin/cli.js in sdk/js
lazynext-a11y rules --pretty
lazynext-a11y report <id>                        # CSV to stdout (--json for the object)
lazynext-a11y monitor add <url> --license buyer@x.com
  • sdk/js — index.js + index.d.ts, zero deps, Node 18+ and browsers; bin: entry lazynext-a11y (node sdk/js/bin/cli.js pre-install)
  • sdk/go — checker package, standard library only (go build clean)
  • sdk/python — lazynext_a11y module, stdlib only (pip install sdk/python)
  • scripts/ci-scan.mjs — CI gate: node scripts/ci-scan.mjs --url <u> --fail-under 80

What it checks

  • Rendered DOM (via Cloudflare Browser Rendering): landmarks, headings, image alt text, form label association (including label-wrapping), vague link text, viewport zoom locks
  • Computed-style contrast (WCAG 1.4.3 + 1.4.6): real luminance math on painted styles — 4.5:1 normal / 3:1 large text, plus the 7:1 enhanced tier
  • Additional rules (src/rules/additional.js): heading content (2.4.6), label-in-name (2.5.3), viewport max-scale (1.4.4), text spacing overrides (1.4.10), language-of-parts (3.1.2), status messages (4.1.3), focus visibility (2.4.7 — inline + stylesheet outline suppression), timed refresh (2.2.1), autoplay/moving content (1.4.2/2.2.2), autocomplete tokens (1.3.5), onfocus context change (3.2.1), onchange auto-submit (3.2.2), accesskey / single-char shortcuts (2.1.4), images of text (1.4.5), multiple navigation mechanisms (2.4.5), justified text without hyphenation (1.4.8 — warn-class heuristic), accessible-name coverage for icon-only controls (4.1.2), motion actuation (2.5.4), orientation lock (1.3.4), hover/focus content without Escape dismissal (1.4.13 — warn-class), and down-event navigation (2.5.2). Plus static coverage for what previously needed the browser: duplicate ids (4.1.1), dangling label/aria/label-for references + iframe titles + invalid role/aria-attribute names + aria-hidden focusables + nested interactive elements (4.1.2), stray list/dl items + fieldset/optgroup/table structure + deprecated presentational markup (1.3.1), image inputs/areas/canvas/svg without alternatives (1.1.1), media without caption tracks (1.2.1), autofocus (3.2.1), role/tabindex/scrollable keyboard gaps (2.1.1), javascript:/dead/dangling-fragment links (2.4.4), blink content (2.3.1), sensory-only instructions (1.3.3 — warn-class), unassociated labels (3.3.2), missing section headings on long content (2.4.10 — AAA advisory), and unnamed duplicate landmarks (1.3.1)
  • Rendered layout checks: target size 24×24px (2.5.8 — with inline-link and UA-control exemptions), focus not obscured by author overlays (2.4.11), visible focus indicator (2.4.13), non-text contrast of control boundaries (1.4.11 — 3:1 vs adjacent background), text-spacing override clipping (1.4.12 — WCAG metric injection, delta-only), and use of color (1.4.1 — prose-scoped: non-underlined in-text links need ≥3:1 vs surrounding text; nav/structural links are out of scope by the criterion itself)
  • Keyboard: real Tab presses in the browser — keyboard-inaccessible pages (2.1.1), focus traps (2.1.2), focus-order gaps and cycles (2.4.3), dialog Escape handling; statics catch Tab-swallowing handlers and undismissable dialogs on pasted HTML too
  • Shadow DOM: rendered scans serialize open shadow roots into <template shadowrootmode> markers, so markup rules cover encapsulated content; the focus census and Tab trace pierce open roots too. Closed roots can't be pierced by design — an honest gap. Element rules ignore <script> bodies, so markup-shaped JS strings can't double-count real elements
  • Cross-page (src/rules/crosspage.js, site scans): consistent navigation (3.2.3), consistent identification (3.2.4), consistent help mechanisms (3.2.6)
  • WCAG 2.2 + Section 508: src/rules/wcag22.js adds 2.2-era checks (accessible auth 3.3.8, redundant entry 3.3.7, dragging 2.5.7, …); src/rules/section508.js maps findings to 36 CFR 1194 clauses
  • DOM facts: iframe titles, duplicate ids, aria-hidden focusables, autofocus, target=_blank without noopener, media captions, table headers, skip links

Pricing

  • Free: 3 rendered scans/day per IP
  • Pro $9/mo: unlimited scans + emailed reports — /checkout (Dodo Payments)
    • 14-day free trial: card collected up front, auto-converts at day 14; cancel any time via POST /cancel or the site's cancel link
    • Trial-expiry reminder emailed at day 11

Architecture

  • src/scanner.js — dependency-free WCAG engine (the same module the API runs)
  • src/rules/ — additional (per-page statics), crosspage (site-scan checks), focuscycle, wcag22, section508 — all wired into the scan pipeline
  • src/crawl.js + src/monitor.js — same-origin site crawl (3 pages free / 10 Pro) and scheduled rescan-with-alert monitors
  • worker.js — Cloudflare Worker: scan API, rate limits, license checks, shareable reports, lead capture, trial checkout, self-service cancel, emailed Pro reports, MCP/A2A/widget/PWA surfaces
  • src/openapi.js — OpenAPI 3.1 contract served at /openapi.json; src/rules_catalog.js — the HTML coverage catalog /rules negotiates to
  • index.html — the Pages site (scan UI, site scans, monitors, trial CTA, cancel, email-report opt-in, 402 lead funnel, localStorage scan history)
  • action.yml + scripts/ci-scan.mjs — GitHub Action for CI gating

GitHub Action

Gate a deploy on accessibility score:

- uses: Lazynext-AI/accessibility-checker@main
  with:
    url: https://staging.example.com
    fail-under: 80            # fail the build below this score
    # license: ${{ secrets.A11Y_LICENSE }}   # Pro: higher limits
    # site: true                             # crawl same-origin pages
    # fail-on: wcag-1.1.1,wcag-2.1.2         # rules that always fail

Outputs score and issues; findings land in the step summary with the report link. Standalone: node scripts/ci-scan.mjs --url <u> --fail-under 80.

  • scripts/ — sync-page.mjs regenerates the embedded UI bundle, deploy.mjs deploys with full bindings (never wrangler deploy), ci-scan.mjs is the CI/standalone CLI
  • test/ — node --test suite (277 tests), runs in CI on every push

Built and operated autonomously by Lazynext agents.

Tools it offers (4)

What this server listed when ahel dialed its public endpoint in Oct 2026, with no key and no account of yours. The names are the server’s own.

  • scan_url
  • scan_html
  • get_report
  • list_rules

Signals

Last commit
Oct 2026
Advanced
Delivery
accessibility-checker MCP server → your ahel connector (mcp.ahel.ai) → your AI.
Item type
mcp-server
Key
com-lazynext-accessibility-checker
Source
github.com/lazynext-ai/accessibility-checker
Hosted endpoint
https://api.lazynext.com/mcp