MCP tools

a11y-catscan includes an MCP stdio server for Claude Code and other MCP clients. It is for quick one-page scans and report analysis, not long-running crawls.

Start the server

Either command starts the same tool server:

python3.12 mcp_server.py
./a11y-catscan.py --mcp

Example MCP configuration:

{
  "mcpServers": {
    "wcag-audit": {
      "type": "stdio",
      "command": "python3.12",
      "args": ["/path/to/a11y-catscan/mcp_server.py"]
    }
  }
}

Tools

Tool	Purpose
scan_page	Scan one HTTP/HTTPS page and return flattened findings
analyze_report	Group findings from a report
list_engines	Show engine versions and install status
lookup_wcag	Look up a Success Criterion
find_issues	Search a report by SC, URL, selector, outcome, or engine
check_page	Check one URL inside a report
compare_scans	Structured diff between two reports
manage_scans	List, get, or delete registry entries

Safety boundaries

The MCP server may be driven by an LLM, so it validates inputs more strictly than the CLI:

• scan_page accepts only http and https.
• Private, loopback, link-local, reserved, multicast, and unspecified scan targets are rejected by default.
• Report tools resolve report-like files (.json, .jsonl) or registered scan names, not arbitrary local files.

For trusted local testing only, you can allow private addresses:

A11Y_CATSCAN_MCP_ALLOW_PRIVATE=1 python3.12 mcp_server.py

Do not use that override for an MCP server exposed to untrusted prompts or shared clients.

When to use the CLI instead

Use the CLI for:

• full crawls
• authenticated scans
• resume state
• periodic flushing
• signal handling
• output file generation

MCP tools are best for quick answers while you are working in a codebase: "does this page still fail?", "which reports mention SC 1.4.3?", or "what does this criterion mean?"