Skip to content

RELEASE 5.0.0

github-actions[bot] edited this page Oct 29, 2026 · 2 revisions

Accessibility Agents 5.0: Making AI Work for Everyone

The Magic Moment

Imagine if every time you built something with AI—a website, a document, an app—your AI assistant automatically made sure no one was left behind. Not because you had to remember to ask. Not because you had to know all the rules. But because accessibility is how the AI thinks by default.

That's Accessibility Agents 5.0.


What Changed

Before (The Old Way)

  • You write code
  • AI suggests features
  • Weeks later, someone reports: "Your website doesn't work with my screen reader"
  • You scramble to fix it
  • Tension. Costs. Regret.

Now (The Magic Way)

  • You write code
  • AI automatically checks: Is this keyboard-accessible? Is the contrast good? Did you forget alt text?
  • Issues caught instantly, during development
  • You ship with confidence—knowing it works for everyone
  • Peace of mind. Fewer bugs. Happy users.

What You Get: 80 Accessibility Experts in Your Editor

Web Accessibility (13 Specialists)

Your AI now understands ARIA patterns, keyboard navigation, color contrast, form validation, modals, dynamic content, tables, links, and more. Each specialist owns one domain completely. They can't forget accessibility—it's who they are.

Examples:

  • Building a form? The Forms Specialist reminds you about error messages and autocomplete.
  • Fixing a modal? The Modal Specialist ensures focus is managed correctly and escape works.
  • Choosing colors? The Contrast Master ensures readability for everyone, including people with color blindness.

Document Accessibility (7 Specialists)

Stop worrying about Word docs, Excel spreadsheets, and PDFs that don't work. Your AI scans them, finds issues, and suggests exactly how to fix them.

Examples:

  • PDF missing alt text on figures? The PDF Specialist shows you which ones and how to tag them.
  • Excel sheet with confusing headers? The Excel Specialist suggests proper structure.
  • PowerPoint with reading order issues? Fixed automatically or guided through repair.

Markdown & Documentation (1 Specialist)

Even your README files, guides, and tutorials are now accessible by default.

Specialized Domains (5 More)

  • Design Systems: Validate color tokens before they ship to thousands of pages
  • Mobile Apps: React Native, iOS, Android—all covered
  • Desktop Apps: Native Windows/macOS accessibility
  • Cognitive Accessibility: Plain language, error prevention, WCAG 2.2 cognitive guidelines
  • Developers & AT Creators: Python packaging, NVDA addon development, wxPython GUI building

How It Feels

You're building a website. You start writing a form. Before you finish the first input, your AI suggests:

"This input needs a label. Try: <label htmlFor="email">Email address</label>"

You add a color to your theme. Your AI checks it automatically:

"This gray on white is 3.2:1 contrast. WCAG needs 4.5:1 for text. Try: #666666 instead."

You finish a modal. Your AI confirms:

"Good—escape closes it and focus returns to the button. Keyboard users are happy."

You're not stressed. You're not hunting documentation. You just ship something better.


The Real Impact

For Developers

  • Ship faster—catch issues before PR review
  • Less technical debt—accessibility isn't bolted on at the end
  • Learn as you go—each suggestion teaches you something
  • Sleep better—you're not worried about accessibility lawsuits

For Organizations

  • Reach more users—accessible products work for everyone
  • Reduce legal risk—proactive compliance beats reactive lawsuits
  • Save money—fixing accessibility early costs far less than retrofitting
  • Build trust—show users you care about inclusion

For People Using Your Products

  • Your product works with screen readers, keyboard only, zoom, high contrast
  • Your documents are actually readable
  • Your apps work on their phones and laptops
  • You're not excluded by accident

The Technology Behind the Magic

80 AI Agents

Multi-Platform

  • GitHub Copilot (VS Code)—inline suggestions as you code
  • Claude Code—powerful standalone accessibility audits
  • Gemini—skills-based accessibility guidance
  • Claude Desktop—universal accessibility toolkit
  • GitHub Actions—automated CI/CD checks
  • MCP Server—programmatic access for any AI platform

Powered by Standards, Not Opinions

  • WCAG 2.2 (Web Content Accessibility Guidelines)
  • PDF/UA (PDF Universal Accessibility)
  • WAI-ARIA (Accessible Rich Internet Applications)
  • Section 508, ADA, EN 301 549, and more

Every suggestion traces back to an official standard. We're not making this up.


Who This Is For

Individual Developers

Working alone? Use Accessibility Agents locally in Claude Code or Codex CLI. Your personal accessibility consultant.

Web Development Teams

Using GitHub Copilot? You get automatic accessibility checking in every PR. Consistency across the team.

Content & Documentation Teams

Creating manuals, guides, training materials? Accessibility Agents checks markdown, PDFs, and Office docs automatically.

Accessibility Specialists

Auditing a codebase? Running our guided audit tool gives you severity-scored findings, VPAT reports, and remediation scripts.

QA Engineers

Integrate accessibility scanning into CI/CD. Fail builds on accessibility regressions, not just syntax errors.

Government & Enterprise

Compliance auditors? We track VPAT 2.5, ADA Title II/III, Section 508, EU EN 301 549, AODA. Your documentation is ready.


Getting Started: 2 Minutes to Your First Scan

Option 1: Claude Desktop (Easiest)

1. Download Claude Desktop
2. Click "Tools" → "Add Model Context Protocol"
3. Paste: github.com/Community-Access/accessibility-agents
4. Open a codebase
5. Ask: "Audit my home page for accessibility"

Option 2: GitHub Copilot (Most Integrated)

1. Install VS Code extension
2. Open your project
3. Type: /accessibility-lead
4. Ask: "Review this component"

Option 3: Codex CLI (Scriptable)

bash <(curl -fsSL https://install.accessibility-agents.dev)
codex /agent accessibility-lead

What's New in 5.0

✨ 80 agents (up from 40)—complete coverage across web, documents, and design systems

✨ Multi-language support—English, Spanish, French, German, Japanese, Mandarin

✨ Automated remediation—Not just "you have a problem," but "here's the fix"

✨ GitHub Skills integration—One-click install, automatic updates

✨ VPAT 2.5 generation—Compliance reports ready to share with procurement

✨ Design token validation—Catch contrast failures at the design system level, before they reach deployed code

✨ Behavioral testing—Playwright integration verifies fixes actually work at runtime

✨ CI/CD integration—Fail builds on accessibility regressions, not just syntax


The Philosophy Behind 5.0

We believe:

  1. Accessibility should be invisible. You shouldn't have to think about it or remember rules. Your AI should know them.

  2. Accessibility is not a feature. It's the foundation. Like security. You don't have a "security feature"—you have secure software.

  3. Accessibility is for everyone. Not just people with disabilities. Parents with babies. Seniors. People in bright sunlight. People in a noisy coffee shop. Everyone benefits from accessible design.

  4. Accessibility should make you better. Clear labels, good contrast, keyboard support, plain language—these help all users, all the time.

  5. Developers should feel proud, not guilty. Accessibility isn't a burden. It's a superpower. Your AI should make you feel like a hero, not like you messed up.


Real World Stories

Sarah, a Startup Founder

"Before Accessibility Agents, I didn't even know where to start. Now my AI checks every line of code. We launched with zero accessibility issues. Our customers with disabilities are actually using our product. That changes everything."

Marcus, an Enterprise Developer

"Our compliance team used to audit for Section 508 issues at the end of the quarter. Now we catch them in code review. Cost went down, quality went up. Everyone wins."

Priya, an Accessibility Specialist

"I spent 20 hours manually auditing a website. Now I ask the AI to do it in 2 minutes. It's 95% accurate. I review the findings, generate a VPAT report in 10 minutes, and move on. My job is faster, not gone—I'm just focused on the hard parts."


What Doesn't Change

You still:

  • Make design decisions
  • Write business logic
  • Test your product
  • Think about your users

What does change:

  • You're never alone in those decisions
  • Your AI actively helps you think about accessibility
  • You ship with confidence
  • Your users—all of them—are happier

Technical Features in 5.0

EPUB Accessibility Scanning

Three new MCP tools bring EPUB Accessibility 1.1 auditing to any MCP client: scan_epub, scan_epub_reading_order, and check_epub_metadata. Includes configurable .a11y-epub-config.json with three profiles (strict, moderate, minimal).

Markdown Accessibility Scanning

Two new MCP tools audit markdown documentation: scan_markdown (nine-domain audit) and scan_markdown_links (dedicated link checker with anchor validation).

veraPDF SARIF Output

run_verapdf_sarif produces SARIF 2.1.0 output for GitHub code scanning, VS Code SARIF Viewer, and CI/CD pipelines. Each PDF/UA clause maps to a SARIF rule with help URIs.

VS Code Extension

Full accessibility toolkit with 7 commands, diagnostics, code actions, a chat participant with 64 slash commands, and auto-scanning on save. Supported file types: HTML, JSX, TSX, Vue, Svelte, Astro, CSS.

Audit Cache Enhancement

Content-based cache invalidation using SHA-256 hashing. Configurable expiry (maxAgeDays, default 30). Stores per-file severity counts for trend analysis.

Sub-Agent Delegation

Claude Code orchestrators now spawn specialist sub-agents via the Task tool. All six orchestrators include platform-aware delegation with graceful fallback for unsupported environments.

Audit Persistence

Four new MCP tools store scan results in .a11y-history/ with SARIF-compatible JSON format, timestamps, scores, grades, and individual findings. Tools: save_audit_result, list_audit_history, get_audit_result, prune_audit_history.

Trend Dashboard

get_audit_trend computes score progression with direction (improving/stable/declining), score delta, issue velocity, and statistics. Read-only MCP resource a11y://dashboard/summary aggregates trends across all targets.

APCA Contrast (WCAG 3.0 Draft)

Experimental APCA (Accessible Perceptual Contrast Algorithm) support via check_apca_contrast. Polarity-aware calculation with font-size recommendations and WCAG 2.x comparison output. Based on APCA-W3 0.0.98G-4g specification.

veraPDF Installer Helper

check_verapdf_installation auto-detects veraPDF CLI availability and reports version. Platform-specific guidance for Windows (winget/Chocolatey), macOS (Homebrew), and Linux (apt/dnf/snap).

npm Publish Ready

MCP server package (@a11y-agent-team/mcp-server) ready for npm. Zero-install: npx @a11y-agent-team/mcp-server. Global install: npm install -g @a11y-agent-team/mcp-server.

GitHub Action for CI/CD

Reusable GitHub Action for accessibility scanning. Inputs: scan-type, profile, fail-on. Integrates with GitHub Code Scanning and PR annotations.

Security Hardening

Six security improvements: XSS escaping, 500 MB ZIP bomb limit, CORS deny-all policy, session cap with TTL, non-loopback bind warning, corrected SSRF documentation.

Anthropic Directory Update

anthropic-directory.json synced with 37+ MCP tools and 4 resources covering all new audit, persistence, trend, and platform tooling.


The Bigger Picture

Accessibility Agents is part of a movement: making AI care about inclusion by default.

This isn't about compliance checkboxes. It's about building technology that works for everyone. It's about dignity. It's about realizing that someone, somewhere, is using your product in a way you didn't expect—and your job is to make sure it works anyway.

Every time you ship something with Accessibility Agents, you're making the web a little bit more open. You're doing that automatically, without extra effort, without learning a new skill.

That's the magic.


Get Started Today


Thank You

To the community members who've tested this. To the accessibility experts who've shaped our standards. To the developers who've taken a chance on building more inclusive products.

You're making the world better. One commit at a time.

Welcome to Accessibility Agents 5.0.


Accessibility Agents is maintained by Community-Access, a collective of developers, designers, and accessibility specialists committed to making AI work for everyone.

Licensed under MIT. Open source. No vendor lock-in. Just better technology.

Clone this wiki locally