MyA11yReport MCP

The MyA11yReport MCP is a local Model Context Protocol server that gives an AI agent the tools to audit a website: it drives a real Chromium browser with Playwright, runs axe-core, and reads the page’s structure, tab order, images and colors. An agent connects to it over stdio and can then navigate, inspect and report on pages on its own.

It runs entirely on your machine — no account, no API key, no data leaves the host. For the full step-by-step workflow an agent should follow, see the accessibility auditor skill.

What it gives your agent

Automated audits — axe-core scans with violation counts, per-rule detail and target selectors.
Page review — landmarks, headings, nested lists and frames; tab/focus order with out-of-order tabindex flags; every image and its alt text.
Contrast — WCAG 2.2 ratio for any two colors, no browser needed.
Browser control — navigate, click, type, check, select, scroll, screenshot and run arbitrary page JavaScript, so the agent can reach content behind accordions, modals and menus.

Sessions are explicit: the agent calls start_session, passes the returned sessionId to every later call, and calls close_session when done. Every action is logged to a session history file.

Install

Requires Node.js 20 or newer.

# Install the server
npm i -g mya11yreport-mcp

# Download the matching Chromium build (about 150 MB, one time)
mya11yreport-mcp install chromium

Quick check that it runs:

echo '{}' | mya11yreport-mcp

If you would rather not install globally, replace the command with npx -y mya11yreport-mcp everywhere below — but still run the browser step once with npx -y mya11yreport-mcp install chromium.

Add it to your agent

opencode

Add this to your project’s opencode.json, or to ~/.config/opencode/opencode.json for every project, then restart opencode.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "mya11y-audit": {
      "type": "local",
      "command": ["mya11yreport-mcp"],
      "enabled": true
    }
  }
}

If you are not installing globally, use the npx form instead:

"command": ["npx", "-y", "mya11yreport-mcp"]

Claude Desktop

Edit claude_desktop_config.json — on macOS at ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows at %APPDATA%\Claude\claude_desktop_config.json — then restart Claude Desktop.

{
  "mcpServers": {
    "mya11y-audit": {
      "command": "npx",
      "args": ["-y", "mya11yreport-mcp"]
    }
  }
}

If you installed globally, use "command": "mya11yreport-mcp" with "args": [].

Claude Code

Add the server from the terminal:

claude mcp add mya11y-audit -- npx -y mya11yreport-mcp

Anything else

Any MCP client that supports local stdio servers works. Configure the command as mya11yreport-mcp (global install) or npx -y mya11yreport-mcp, with no arguments and no environment variables.

Tools

Tool	What it does
`start_session` / `close_session`	Open and close a browser session
`get_audit_guide`	The full audit workflow, as markdown (no session needed)
`run_a11y_audit`	Run an axe-core scan on the current page or a URL
`get_structure`	Landmark regions, headings, nested lists and frames
`get_tab_order`	Tabbable order, with positive-`tabindex` flags
`list_images`	Every `img` and inline `svg`, with alt text and source
`check_color_contrast`	WCAG 2.2 ratio for two colors (no session needed)
`navigate`	Load a URL
`click` / `type` / `press_key` / `check` / `uncheck` / `select_option`	Interact with the page
`scroll_by`	Scroll the page
`screenshot`	Capture the current viewport
`get_page_snapshot`	Accessibility-tree snapshot of the page
`evaluate`	Run JavaScript in the page for computed styles or geometry
`get_viewport_size` / `get_scroll_position`	Read layout state

Tools that read the page (get_structure, get_tab_order, list_images) cover the top frame only and pierce open shadow roots, matching the MyA11yReport browser extension.

Running an audit

Give your agent a URL or a sitemap and ask for an accessibility audit. It will call get_audit_guide first, then work through the automated pass and the structure, tab-order, image and contrast reviews. The guide is explicit about the checks a machine cannot make — keyboard navigation, screen-reader compatibility and image meaning — which a human must verify.