WCAG Accessibility MCP

A WCAG accessibility testing MCP for AI agents, Codex, Claude Code, and automated UI audits.

WCAG Accessibility MCP gives AI coding and design agents a real accessibility feedback loop. It renders web interfaces in Chromium, runs axe-core, checks color contrast, maps findings to WCAG guidance, and returns structured evidence that an agent can use to fix accessibility issues before a UI is treated as complete.

It is designed for people building with AI: product designers, design engineers, frontend developers, AI product teams, and anyone using agents to generate or review web interfaces.

This is an automated testing aid, not a WCAG certification service. It catches many common failures, but full accessibility review still requires human testing for keyboard behavior, focus order, screen readers, zoom, motion, content quality, and assistive-technology behavior.

Why This Exists

AI-generated interfaces often look polished but fail basic accessibility checks: weak contrast, missing labels, broken semantic structure, poor ARIA usage, inaccessible forms, and hidden keyboard issues.

This MCP helps agents stop guessing. Instead of saying “follow WCAG,” it gives them testable feedback:

what failed
where it failed
which element was affected
which WCAG-related rule applies
what needs to be fixed
which checks still require manual review

What It Can Do

Tool	What it does
`audit_url`	Audits a rendered web page URL
`audit_html`	Audits supplied HTML before it is hosted
`audit_file`	Audits a local `.html` or `.htm` file inside an allowed project root
`audit_matrix`	Experimental: audits one target across named viewports and groups findings by stable fingerprints
`compare_audits`	Experimental: compares baseline and candidate audit results without rerunning the browser
`browser_diagnostics`	Experimental: reports browser selection and setup guidance without launching Chromium
`inspect_screenshot`	Experimental: captures screenshot evidence and optionally runs local OCR
`inspect_design_snapshot`	Experimental: inspects exported Figma, Claude Design, or generic design-node data before implementation
`check_contrast`	Calculates WCAG contrast for a foreground/background color pair
`suggest_contrast_fix`	Suggests a passing color adjustment
`explain_issue`	Explains an accessibility rule and how to fix it
`get_wcag_checklist`	Returns the complete WCAG checklist with automated/manual coverage

Standards Coverage

The server supports these profiles:

wcag2a
wcag2aa
wcag2aaa
wcag21aa
wcag21aaa
wcag22aa
wcag22aaa
best-practice

AA profiles include Level A and AA criteria. AAA profiles include Level A, AA, and AAA criteria.

The default profile is wcag22aa.

For WCAG 2.2:

AA includes 55 required success criteria.
AAA includes 86 required success criteria.

An automated axe mapping means partial automated coverage. It does not mean the whole WCAG success criterion has been fully tested.

Install

Requirements:

Node.js 20 or newer
Chrome or Edge recommended
Windows, macOS, or Linux

Install from npm:

npm install -g wcag-accessibility-mcp

Or run directly:

npx -y wcag-accessibility-mcp

If the server starts correctly, you should see:

wcag-accessibility-mcp is running over stdio

Use With Claude Code

Add the MCP server:

claude mcp add --scope user wcag-accessibility -- npx -y wcag-accessibility-mcp

Check that it is connected:

claude mcp list

Test prompt:

Use the wcag-accessibility MCP tool check_contrast. Check #999999 on #ffffff for WCAG AA normal text. Do not calculate manually.

Expected result:

The contrast ratio is about 2.85:1, which fails WCAG AA for normal text.

Use With Claude Desktop

Open the Claude desktop config file on Windows:

notepad "$env:APPDATA\Claude\claude_desktop_config.json"

Add this inside mcpServers:

{
  "mcpServers": {
    "wcag-accessibility": {
      "command": "C:\\Program Files\\nodejs\\npx.cmd",
      "args": [
        "-y",
        "wcag-accessibility-mcp"
      ]
    }
  }
}

If you already have another MCP server, keep it and add wcag-accessibility as a second entry.

Then fully quit Claude Desktop and reopen it.

Use With Codex

Add the MCP server:

codex mcp add wcag-accessibility -- npx -y wcag-accessibility-mcp

Test prompt:

Use the wcag-accessibility MCP to check the contrast of #999999 on #ffffff at WCAG AA.

Example Agent Prompts

Audit this HTML for WCAG 2.2 AA issues. Prioritize critical and serious issues, explain each fix, then list what still needs manual review.

Use wcag-accessibility to audit this local HTML file at desktop and mobile widths. Fix the accessibility issues in the code and rerun the audit.

Inspect this Figma or Claude Design snapshot for accessibility risks before implementation, then verify the rendered HTML with audit_matrix.

Capture a screenshot of this page and run OCR only if local OCR is configured. Use OCR findings as visual evidence, not as WCAG conformance.

Check whether #FF3B12 on #F5F2EC passes WCAG AA for normal text. If it fails, suggest the nearest passing foreground color.

Give me the WCAG 2.2 AAA checklist and separate automated checks from manual checks.

Local Development

Clone the repository:

git clone https://github.com/aditya-ariosity/wcag-accessibility-mcp.git
cd wcag-accessibility-mcp

Install dependencies:

npm install

Build:

npm run build

Run tests:

npm run test:unit

Run browser integration and end-to-end tests:

npm run test:browser

npm test is the deterministic browser-free unit suite. Use npm run test:browser when the machine has a working Chrome, Edge, Chromium, or bundled browser setup.

The CLI can write JSON, Markdown, or SARIF reports and compare a baseline:

npm run audit -- --url http://localhost:3000 --format json --out report.json
npm run audit -- --url http://localhost:3000 --baseline baseline.json --format md

Start the stdio MCP server:

npm start

Start the optional HTTP transport:

npm run start:http

The local HTTP endpoint is:

http://127.0.0.1:3000/mcp

Health check:

http://127.0.0.1:3000/health

Environment Variables

All environment variables are optional for normal local use.

Variable	Default	Purpose
`A11Y_MCP_BROWSER_PATH`	Auto-detected	Absolute path to Chrome, Chromium, or Edge
`A11Y_MCP_ALLOWED_ROOT`	Current working directory	Restricts local file audits to a specific folder
`A11Y_MCP_ALLOW_PRIVATE`	`true` on stdio, `false` on HTTP	Allows localhost/private-network URL audits in trusted environments
`A11Y_MCP_ENABLE_FILE_AUDIT`	`false` on HTTP	Enables local file audits over HTTP in trusted deployments
`A11Y_MCP_BROWSER_CONCURRENCY`	`2`	Maximum concurrent Chromium audits
`A11Y_MCP_AUDIT_TIMEOUT_MS`	`25000`	Timeout for the audit phase
`A11Y_MCP_NO_SANDBOX`	`false`	Disables Chromium sandboxing only when the runtime requires it
`A11Y_MCP_OCR_COMMAND`	`tesseract`	Local OCR executable used by `inspect_screenshot` when `runOcr` is true
`A11Y_MCP_ENABLE_EXPERIMENTAL_TOOLS`	`false`	Advertises the in-progress matrix, screenshot, diagnostics, and design-snapshot tools
`A11Y_MCP_ALLOW_LOCAL_FILE_SUBRESOURCES`	`false`	Allows local file audits to load file-based images/stylesheets beneath the allowed root
`HOST` / `A11Y_MCP_HOST`	`127.0.0.1`	HTTP bind address
`A11Y_MCP_ALLOWED_HOSTS`	Localhost names	Allowed Host headers for HTTP mode
`A11Y_MCP_BODY_LIMIT`	`4mb`	HTTP JSON body limit
`PORT`	`3000`	HTTP transport port

Security Notes

The stdio server is meant to run locally with the same permissions as the agent using it.

The HTTP transport is a deployment building block, not a public hosted service by itself. Do not expose it directly to the public internet without authentication, TLS, rate limits, request-size limits, outbound network controls, and a low-privilege runtime.

By default:

tools are read-only
HTTP mode blocks private network URL targets
local file audit over HTTP is disabled
local file audit is restricted to the allowed root
Chromium sandboxing remains enabled unless explicitly disabled

Published Links

GitHub: https://github.com/aditya-ariosity/wcag-accessibility-mcp
npm: https://www.npmjs.com/package/wcag-accessibility-mcp
MCP Registry name: io.github.aditya-ariosity/wcag-accessibility

Current Scope

The stable release currently audits rendered web interfaces and HTML. The evidence core reports criterion-level coverage and keeps unresolved manual requirements visible.

The stable release does not yet inspect:

native mobile apps
PDFs
canvas-only interfaces
video captions
raw screenshots without HTML
raw Claude Design pixels

Matrix, screenshot/OCR, and design-snapshot modules are experimental and require A11Y_MCP_ENABLE_EXPERIMENTAL_TOOLS=true. Direct Figma plugin inspection, persistence, automatic fixes, and deeper design-tool adapters remain future work.

License

MIT

wcag accessibility mcp