AI coding agents are notorious for prioritizing speed over substance, often leaving developers with visually appealing but functionally broken interfaces. A new Model Context Protocol (MCP) server called WebAbility aims to fix this by integrating automated WCAG 2.2 AA compliance checks directly into the workflow of tools like Claude Code, Cursor, and VS Code. The tool addresses the common pitfall where LLMs generate forms and modals that look correct but fail screen readers due to missing labels or poor contrast.

The Scan-Fix-Verify Loop

The core innovation of WebAbility is its tripartite workflow that moves beyond simple reporting. Unlike traditional accessibility tools that generate static lists of errors, WebAbility provides machine-readable fixes that agents can execute. The process begins with scan_page, which identifies high-confidence violations and separates them from incomplete findings that require human judgment, such as context-dependent alt text. This distinction prevents agents from blindly applying incorrect semantic fixes.

Multi-Engine Detection and Verification

Under the hood, the server aggregates data from three distinct engines: WebAbility’s own 60+ detectors, axe-core’s 104 rules, and HTML_CodeSniffer’s 200+ rules, deduplicating results to avoid noise. Once an agent applies a fix, the verify_fix tool re-scans the specific element to confirm resolution. This closed-loop verification is critical; if an agent claims a contrast issue is fixed but the tool reports a ratio of 3.1:1 instead of the required 4.5:1, the agent is forced to iterate rather than hallucinate success.

Integration and Local Development

Setup is frictionless, requiring no API keys or accounts for basic use via the hosted endpoint at https://mcp.webability.io/mcp. For local development, the tool includes a tunneling mechanism (webability-tunnel) that allows the hosted server to scan localhost instances, ensuring compliance checks happen before code is pushed. The integration supports major clients including Claude Code, Cursor, Windsurf, Zed, and Gemini CLI, with specific commands provided for each environment.

Key Takeaways

  • WebAbility MCP enables AI agents to automatically fix mechanical accessibility issues while flagging contextual ones for human review.
  • The tool uses a 'fixability' tag (mechanical, contextual, visual) to determine if an AI can safely apply a fix or if it needs human oversight.
  • Verification is automated via verify_fix, preventing agents from falsely reporting that an accessibility issue has been resolved.
  • The service is free for anonymous users with fair-use limits, lifting restrictions for signed-in users who need advanced visual audits.

The Bottom Line

This is a necessary evolution for AI coding agents; without automated verification loops, LLM-generated code remains a liability for enterprise compliance. WebAbility effectively turns accessibility from a manual audit into a programmatic constraint that agents can respect and resolve in real time.