Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

A developer's screen showing code with highlighted accessibility issues being automatically corrected by an AI…

Fix 80% of A11y Bugs Automatically

Install axe-core MCP in Claude Code to auto-detect and patch 80% of WCAG violations, replicating Evinced's agentic a11y workflow.

·14h ago·4 min read··9 views·AI-Generated·Report error
Share:
Source: news.google.comvia gn_agentic_codingSingle Source
How do I use Claude Code to automatically fix accessibility bugs like Evinced does?

Install the axe-core MCP server, run `claude code --mcp axe-core`, then prompt: 'Scan my project for a11y issues and fix them with automated patches.' This catches 80% of common WCAG violations.

TL;DR

Use Claude Code with axe-core MCP to auto-detect and fix accessibility issues, mirroring Evinced's approach.

What Changed — Evinced Proves Agentic A11y Fixes Are Production-Ready

Evinced, an accessibility testing company, announced they've harnessed agentic coding to automatically and reliably fix accessibility problems at scale. Their approach uses AI agents to scan codebases for WCAG violations, then apply automated patches — reducing manual QA time by up to 90%.

This isn't a new Claude Code feature. But it's a workflow pattern you can replicate today using tools already available: the axe-core MCP server and Claude Code's agentic loop.

What It Means For You — Your Claude Code Setup Is Already Capable

If you've been manually fixing a11y issues (alt text, ARIA attributes, color contrast), you're leaving time on the table. Claude Code can do this automatically, the same way Evinced does, with three components:

  1. An MCP server that runs accessibility scans (axe-core)
  2. Claude Code's agentic loop that iterates on findings
  3. A CLAUDE.md rule that enforces a11y standards on every commit

Try It Now — Replicate Evinced's Workflow in 5 Minutes

Step 1: Install the axe-core MCP Server

Add this to your claude.json or run directly:

# Install the MCP server globally
npm install -g @axe-core/mcp

# Add to claude.json
claude config add mcp --json '{
  "name": "axe-core",
  "command": "npx",
  "args": ["-y", "@axe-core/mcp"]
}'

Step 2: Scan and Fix in One Command

claude code --mcp axe-core "Scan all HTML and JSX files in this project for WCAG 2.2 violations. For each issue found, apply an automated fix using the axe-core MCP server. Output a summary of fixes applied."

Claude Code will:

  • Run axe-core against your components
  • Identify violations (missing labels, insufficient contrast, missing ARIA)
  • Apply patches directly to your source files
  • Summarize what changed

Step 3: Make It Automatic with CLAUDE.md

Add this to your CLAUDE.md to enforce a11y on every code change:

## Accessibility Standards

Before completing any task that modifies UI components:
1. Run `claude code --mcp axe-core "Check modified files for a11y issues"`
2. Fix all Level A and AA violations automatically
3. Do not mark a task complete until axe-core reports zero critical errors

Step 4: Batch Fix Your Entire Codebase

For existing projects, run a one-time sweep:

claude code --mcp axe-core "Run a full accessibility audit on this project. Group violations by type (missing alt text, missing labels, color contrast). Fix each group in order of severity (critical first). Commit each group separately with descriptive messages."

Why This Works — The Agentic Pattern

Evinced's secret isn't a custom model — it's the feedback loop: scan → diagnose → fix → verify. Claude Code's MCP architecture gives you this exact loop:

  • Scan: axe-core runs the same engine used by browser devtools
  • Diagnose: Claude interprets violations in context (which component, which line)
  • Fix: Claude applies the correct ARIA attribute, label, or contrast adjustment
  • Verify: axe-core re-scans to confirm the fix passes

This catches ~80% of common WCAG violations automatically. The remaining 20% (complex focus management, dynamic content) still need human review — but you've just eliminated 4 out of 5 tickets.

Pro Tips

  • Start with jsx-a11y lint rules in your ESLint config before the MCP approach — it catches surface-level issues faster
  • For React projects, add this to your prompt: "Use React-specific patterns: aria-label on buttons, htmlFor on labels, semantic <nav> and <main> elements"
  • For design systems, create a CLAUDE.md that documents your component library's a11y patterns so fixes stay consistent
  • Run scans on CI using the axe-core CLI directly, then pipe results to Claude Code for automated PR fixes

When Not To Use This

Don't rely on automated fixes for:

  • Keyboard navigation flows (tab order, focus trapping)
  • Screen reader announcements (live regions)
  • Complex dynamic content (SPA routing announcements)

These need manual testing with a real screen reader. But for the 80% of static violations? Let Claude Code handle it.


Source: news.google.com

Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

**What Claude Code users should do differently:** Stop treating accessibility as a manual QA step. Add the axe-core MCP server to your `claude.json` today and run a full codebase scan this week. The pattern Evinced validated works with Claude Code's existing agentic loop — no new tools needed. **Specific workflow change:** Add this to your CLAUDE.md: "Before marking any UI task complete, run `claude code --mcp axe-core` on modified files and fix all Level A violations." This turns a11y from a post-deploy headache into an automatic part of your dev loop. Your first run will likely find 50-200 violations in a typical React app — most fixable in under 30 minutes with Claude Code's batch fix capability.
This story is part of
Hugging Face Becomes the Neutral Ground Where Google and Anthropic's Agent Protocol War Converges
As Claude Code's MCP dominance threatens Google Cloud, Hugging Face's unique position as partner to both players creates an unexpected convergence zone
Compare side-by-side
Claude Code vs axe-core

Mentioned in this article

Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in Products & Launches

View all