axle · Model Context Protocol

axle MCP server

One npx away from giving Claude Desktop, Cursor, Cline, or any MCP-compatible agent the ability to scan a URL for WCAG 2.1 / 2.2 AA violations and return a structured report — same engine (axe-core 4.11) that powers the axle GitHub Action and the WordPress / VSCode / Storybook plugins.

What your agent gets

Install in Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "axle": {
      "command": "npx",
      "args": ["-y", "axle-mcp"]
    }
  }
}

Restart Claude Desktop. You'll see a 🔨 icon in the composer — click it and axle.scan_url is callable.

Install in Cursor

Settings → Features → Model Context Protocol → Add server:

Command: npx -y axle-mcp

Install in Cline (VS Code)

{
  "axle": {
    "command": "npx",
    "args": ["-y", "axle-mcp"]
  }
}

Install in Continue.dev

In ~/.continue/config.json:

{
  "mcpServers": {
    "axle": {
      "command": "npx",
      "args": ["-y", "axle-mcp"]
    }
  }
}

Try it

Once installed, in any MCP-connected chat just ask:

Removing the rate limit

Free tier is 5 scans / day per IP. To remove the limit, get an API key from /pricing and add it to the MCP config:

{
  "mcpServers": {
    "axle": {
      "command": "npx",
      "args": ["-y", "axle-mcp"],
      "env": {
        "AXLE_API_KEY": "sk_axle_..."
      }
    }
  }
}

Why axle-mcp vs other a11y MCP servers?

Honest scope

axe-core (and therefore this MCP server) catches roughly 57% of WCAG issues automatically. The other 43% require manual review by a qualified human auditor. Treat the output as a high-recall first pass, not a conformance certificate.


Source: github.com/asafamos/axle/packages/axle-mcp · npm: axle-mcp · Already use ChatGPT? Add axle to a GPT instead.