axle — a11y / WCAG scanner for VS Code

Scan a URL — or your localhost dev server — for WCAG 2.2 AA accessibility violations without leaving VS Code, then click a violation to get an AI-suggested code fix. Powered by axe-core 4.11, the same engine used by axle, Pa11y, Lighthouse, and Microsoft Accessibility Insights.

What it does

• Command palette → "axle: Scan a URL" — type any URL, get a violations report in a side panel.
• Command palette → "axle: Scan localhost dev server" — scans http://localhost:3000 (or whatever you set as axle.defaultUrl). Local and private-network URLs are scanned on your machine by running npx axle-cli in a visible terminal — our servers can't reach your localhost. The first run needs a one-time npx playwright install chromium.
• "Generate AI fix" — a link on every violation card (or "axle: Generate AI fix for a violation…" in the palette). It asks Claude for a before/after change for one element and opens it as a Markdown document next to your code. The suggestion is never applied automatically — review it, then apply it yourself.
• "axle: Open last scan result" — re-render the previous result without re-scanning.

Results are grouped by severity (critical / serious / moderate / minor) with the offending HTML, WCAG tags and links to the axe-core rule docs. Issues below your axle.failOn threshold are collapsed, not hidden.

Why use this and not just open a browser

• No context switch — stay in your editor.
• Same engine as the axle CI — what you see locally is what the GitHub Action will catch on a pull request.
• Scans the rendered page — what your users actually see, including everything client-side JavaScript adds.

Settings

Pricing

The extension and scanning are free; no account is needed. AI fixes: the first few each day are free, then you need an axle API key — see pricing. To run the same checks on every pull request, the axle GitHub Action is free for one repository.

Privacy — exactly what leaves your machine

No telemetry, no analytics, no error reporting beyond what VS Code itself does. Requests only happen when you ask:

• Scan a public URL → the URL (and, if set, your API key) is sent to axlescan.com/api/scan, which loads the page in a headless browser and runs axe-core.
• Scan localhost / a private address → nothing is sent to us. The extension runs npx axle-cli locally with telemetry switched off. (It downloads the axle-cli npm package on first use.)
• Generate AI fix → after a one-time confirmation, the rule details plus the HTML of one element (up to 500 characters), its selector and axe's failure summary are sent to axlescan.com/api/fix, which passes them to Anthropic's Claude to produce the suggestion. Your API key, if set, is sent as a header. Nothing else from your workspace is sent.

The extension does not read your workspace files, your filesystem outside its own storage folder, or your git history. See the privacy policy.

How it differs from other a11y extensions

• Web Accessibility (Maxim Salnikov) and axe-linter (Deque) lint HTML source files statically. axle scans the rendered page, so it also catches what JavaScript, runtime ARIA and dynamic content introduce.
• Microsoft Accessibility Insights for Web is a browser extension; this one lives in your editor.

Automated tools find roughly 57% of WCAG issues. This is not a compliance certificate — manual review is still recommended.

Companion tools

License

MIT. Source: github.com/asafamos/axle/tree/main/packages/axle-vscode.