Accessible by Default

AI coding tools forget accessibility. Our agents make sure they never do. 47 specialized agents, 14 skills, and 52 ready-to-use prompts enforce WCAG 2.2 AA compliance inside Claude Code, GitHub Copilot, and Claude Desktop—so every line of code ships inclusive.

View on GitHub Install in 30 Seconds

GIT Going with GitHub

A free, two-day, hands-on workshop where blind and low vision participants learn to navigate, contribute to, and build real open source projects—using a screen reader, a keyboard, and nothing else.

When March 7 & 8, 2026 · 12–8 PM Eastern

Where Online via Zoom · Free

Seats loading... registered out of 75 spots

By the end of Day 2, your name will be in the commit history of a live, public open source project. Not a sandbox. Not a simulation. The real thing.

No coding experience required. NVDA, JAWS, and VoiceOver instructions included for every step. Sighted participants welcome.

Specialized Agents: 47
Platforms: 3
Ready-to-Use Prompts: 52
AA Compliance: WCAG 2.2

The Problem We Solve

AI coding tools generate inaccessible code by default. ARIA gets misused, keyboard navigation gets skipped, contrast fails silently, and modals trap screen reader users.

Automatic Enforcement

Agents evaluate every prompt, every response, every code suggestion. Accessibility is checked before code reaches your editor, not after deployment.

Screen Reader First

Built by and for the blind and low vision community. Every agent understands real-world assistive technology including NVDA, JAWS, and VoiceOver, not just spec compliance.

One-Command Install

A single command installs all agents across Claude Code, GitHub Copilot, and Claude Desktop. No configuration files to write, no YAML to edit.

Beyond Web Code

Scan Word, Excel, PowerPoint, PDF, EPUB, and Markdown documents for accessibility. 46 rules for Office, 56 rules for PDF/UA, plus full documentation auditing.

GitHub Workflow Agents

11 agents that manage repos, triage issues, review PRs, run analytics, and keep your team informed—all from natural language.

VPAT Generation

Generate VPAT 2.5 Accessibility Conformance Reports directly from agent scans. Compliance documentation from your codebase, not a spreadsheet.

axe-core Integration

Agents call axe-core directly to scan rendered HTML for WCAG violations. Results feed back into the agent workflow so issues are detected, reported, and fixed in a single pass.

Markdown Documentation

Agents audit Markdown files across 9 accessibility domains: heading structure, image alt text, link quality, language, tables, lists, readability, and more.

Framework-Aware

Built-in fix templates for React, Vue, Angular, Svelte, and Tailwind CSS. Agents suggest idiomatic fixes in your framework, not generic HTML.

Cognitive & Mobile

Full WCAG 2.2 cognitive accessibility criteria coverage plus React Native, iOS, and Android mobile accessibility auditing.

Meet the Agents

47 specialists across four teams. Each agent owns one domain and enforces it relentlessly. Expand any group to see every agent, and look for “How it works” to see exactly how each one analyzes your content.

Web Accessibility Agents (16)

Agent	Description
Accessibility Lead	Orchestrates the full team—routes tasks to specialists, merges results, and ensures nothing gets missed.
How the web audit works The lead wizard runs a multi-phase guided audit by delegating to specialist sub-agents in parallel: Code review — Each specialist reads your source files (HTML, JSX, Vue, Svelte) and applies its rule set against WCAG 2.2 AA criteria. axe-core scan — The `run_axe_scan` tool launches axe-core against a live URL (e.g. `localhost:3000`) via Chromium, returning violations grouped by severity with affected elements and fix suggestions. Framework-aware fixes — Specialists suggest idiomatic fixes for React, Vue, Angular, Svelte, and Tailwind CSS rather than generic HTML. Aggregation — Results from all specialists are merged, deduplicated, severity-scored, and compiled into a prioritized action plan.
ARIA Specialist	Makes sure interactive elements like buttons, menus, and tabs are properly announced to screen readers.
Modal Specialist	Verifies that dialogs and modals trap focus correctly, return focus on close, and announce themselves.
Contrast Master	Checks that text and backgrounds have enough color difference for people with low vision.
How contrast checking works The `check_contrast` MCP tool calculates the WCAG luminance contrast ratio between any two hex colors using the sRGB-to-linear conversion formula, then reports pass/fail for: Normal text: 4.5:1 minimum Large text (18px+ or 14px+ bold): 3:1 minimum UI components and focus indicators: 3:1 minimum The agent also scans CSS for common Tailwind failures (e.g. `text-gray-400` on white = 2.85:1 FAIL) and checks dark mode colors independently.
Keyboard Navigator	Ensures every interactive element is reachable and operable with keyboard alone—no mouse required.
Live Region Controller	Manages dynamic content announcements so screen readers hear updates like toasts, alerts, and loading states.
Forms Specialist	Validates that every form input has a visible label, proper autocomplete, and clear error messaging.
How form checking works The `check_form_labels` MCP tool parses HTML to verify every `<input>`, `<select>`, and `<textarea>` has a programmatic label via `<label for>`, `aria-label`, or `aria-labelledby`. Also checks: `aria-labelledby` references point to existing IDs Identity fields (email, tel, password) have `autocomplete` attributes Radio/checkbox groups are wrapped in `<fieldset>` with `<legend>`
Alt Text & Headings	Checks image alt text quality and heading hierarchy so pages make sense when read linearly.
How heading analysis works The `check_heading_structure` MCP tool extracts every `<h1>`–`<h6>` from HTML, builds the heading outline, and flags: No headings at all, or no H1 Multiple H1 elements on one page Skipped levels (e.g. H1 → H3 with no H2) Empty headings that confuse screen readers
Tables Specialist	Ensures data tables have proper headers, captions, and scope attributes for screen reader navigation.
Link Checker	Flags vague link text like “click here” and ensures links describe their destination.
How link checking works The `check_link_text` MCP tool scans every `<a>` in HTML and flags: Ambiguous text (“click here”, “read more”, “learn more”, etc.) URLs used as link text (screen readers spell out the full URL) Repeated identical text pointing to different destinations Links opening in new tabs without warning Links to non-HTML files (PDF, DOCX) without file type indication
Web Accessibility Wizard	Interactive guided audit—walks you through a full accessibility review step by step.
Testing Coach	Teaches you how to test with screen readers, keyboard, and browser dev tools.
WCAG Guide	Explains any WCAG 2.2 success criterion in plain language with practical examples.
Cognitive Accessibility	Audits for WCAG 2.2 cognitive criteria: reading level, consistent navigation, error prevention, and timeouts.
Mobile Accessibility	Checks touch targets, viewport scaling, orientation, and mobile screen reader patterns.
Design System Auditor	Reviews component libraries and design tokens for accessibility patterns and consistency.

Document Accessibility Agents (9)

Our Word, Excel, and PowerPoint agents are built on the public-facing rules used by Microsoft Office’s built-in accessibility checkers. We study the published documentation for these rules and map them to WCAG 2.1 AA criteria so the agents can surface the same issues—and more—directly inside your AI workflow.

Agent	Description
Word Accessibility	Scans .docx files for heading structure, alt text, reading order, table headers, and language.
How Word analysis works A `.docx` file is a ZIP archive containing XML. The agent (and the `scan_office_document` MCP tool) unzips the file and reads the XML directly: word/document.xml — Parses `<w:pStyle w:val="Heading1">` to verify heading hierarchy, checks `<wp:docPr descr="...">` for image alt text, scans `<w:tblHeader>` for table header rows, and flags `<w:gridSpan>`/`<w:vMerge>` for merged cells. docProps/core.xml — Reads `<dc:title>` and `<dc:language>` to verify document metadata. word/_rels/document.xml.rels — Resolves hyperlink targets so link text can be checked for ambiguity. Rule set: 9 errors, 6 warnings, 3 tips—mapped to WCAG 2.1 AA and derived from the public-facing rules of Microsoft Word’s built-in Accessibility Checker.
Excel Accessibility	Checks spreadsheets for sheet names, table formatting, merged cells, chart alt text, and data validation.
How Excel analysis works An `.xlsx` file is also a ZIP of XML. The agent reads: xl/workbook.xml — Checks `<sheet name="...">` for default names like “Sheet1”. *xl/worksheets/sheet.xml — Scans for `<mergeCell>` regions, empty sheets, conditional formatting (color-only data), and hyperlink display text. xl/tables/table.xml* — Verifies `headerRowCount` is not “0”. *xl/drawings/drawing.xml — Checks `<xdr:cNvPr descr="...">` for chart and image alt text. Rule set: 8 errors, 5 warnings, 3 tips—derived from the public-facing rules of Microsoft Excel’s built-in Accessibility Checker**.
PowerPoint Accessibility	Audits slide reading order, alt text on images, slide titles, table structure, and color contrast.
How PowerPoint analysis works A `.pptx` file is a ZIP of XML. The agent reads: *ppt/slides/slide.xml — Locates `<p:ph type="title">` placeholders to verify every slide has a title. Checks `<p:cNvPr descr="...">` for image and shape alt text. Flags `gridSpan`/`rowSpan` in tables. ppt/presentation.xml — Checks for `<p:sectionLst>` in presentations over 10 slides. Duplicate detection — Flags slides with identical title text, which breaks screen reader slide navigation. Media — Flags `<p:vid>` and `<a:audioFile>` elements that need captions. Rule set: 5 errors, 4 warnings, 4 tips—derived from the public-facing rules of Microsoft PowerPoint’s built-in Accessibility Checker**.
Office Scan Config	Configures which rules to enable or disable for Office document scans.
PDF Accessibility	Validates PDF/UA compliance using the Matterhorn Protocol—56 checkpoint rules.
How PDF analysis works The agent (and `scan_pdf_document` MCP tool) parses the PDF internal structure to check conformance with PDF/UA (ISO 14289-1) via the Matterhorn Protocol: Structure tree — Verifies `/StructTreeRoot` exists and `/MarkInfo /Marked true` is set (tagged PDF). Semantic tags — Checks that content uses standard tags (`/P`, `/H1`–`/H6`, `/Table`, `/Figure`, `/Link`, `/L`) with correct nesting. Figures — Every `/Figure` element must have an `/Alt` text entry. Decorative images must be marked as `/Artifact`. Tables — `/TH` cells must have `/Scope`. Complex spanning requires `/Headers` attribute. Language — Catalog must have `/Lang` in valid BCP 47 format. Inline language switches need `/Span` tags. Bookmarks — Documents over 20 pages must have `/Outlines` for navigation. Three rule layers: PDFUA (conformance), PDFBP (best practices), PDFQ (quality). Supports SARIF output for CI integration.
PDF Scan Config	Configures which PDF/UA rules to enable or disable for PDF scans.
Document Wizard	Interactive guided document audit—walks you through a full review step by step.
EPUB Accessibility	Scans EPUB publications for accessibility metadata, navigation, alt text, reading order, and semantic structure.
How EPUB analysis works An EPUB is a ZIP containing XHTML content documents, a package OPF, and navigation files. The agent reads: Package OPF — Checks `<dc:title>`, `<dc:identifier>`, `<dc:language>`, and `schema.org` accessibility metadata (`accessMode`, `accessibilityFeature`, `accessibilitySummary`). Navigation document — Verifies `<nav epub:type="toc">`, `page-list`, and `landmarks` exist. Content documents — Scans each XHTML file for image alt text, heading hierarchy, table headers, ambiguous link text, and color-only information. Spine — Checks `<itemref>` order for logical reading sequence. Conforms to EPUB Accessibility 1.1 mapped to WCAG 2.1 AA. Rule set: 7 errors, 6 warnings, 4 tips.
EPUB Scan Config	Configures which rules to enable or disable for EPUB accessibility scans.

Markdown Accessibility Agents (3)

These agents audit Markdown documentation for accessibility across 9 domains.

Agent	Description
Markdown Assistant	Orchestrates Markdown audits—coordinates the scanner and fixer for full documentation reviews.
Markdown Scanner	Scans .md files for heading structure, image alt text, link quality, tables, and readability. Runs automatically.
How Markdown analysis works The scanner reads each `.md` file and runs two analysis passes: markdownlint — Runs `npx markdownlint-cli2` to catch structural violations (MD001 heading hierarchy, MD025 multiple H1s, MD034 bare URLs, MD045 missing image alt text, MD055/056 table formatting). 9-domain accessibility scan — Checks each line for: Descriptive links (WCAG 2.4.4) — flags “click here”, bare URLs, repeated identical text Image alt text (WCAG 1.1.1) — empty or decorative alt, oversized alt text Heading hierarchy (WCAG 2.4.6) — skipped levels, missing top-level heading Table accessibility — alignment-only headers, missing header rows Emoji (cognitive accessibility) — excessive or decorative emoji that confuse screen readers Mermaid/ASCII diagrams — no text alternative for visual diagrams Language markers, readability, and list structure Each issue gets a severity, line number, and auto-fix classification. The fixer then applies safe fixes automatically.
Markdown Fixer	Applies automated fixes for accessibility issues found by the scanner. Runs automatically.

GitHub Workflow Agents (11)

Agent	Description
GitHub Hub	Central router for all GitHub operations—directs requests to the right specialist.
Daily Briefing	Gives you a morning summary of what happened in your repo overnight.
PR Review	Reviews pull requests for accessibility issues, code quality, and adherence to contribution standards.
Issue Tracker	Manages issue triage, labeling, assignment, and status tracking.
Analytics	Generates repository health reports: PR velocity, issue resolution time, contributor activity.
Insiders Tracker	Tracks VS Code Insiders accessibility issues and cross-references with your project.
Repo Admin	Handles repository settings, branch protection, and administrative tasks.
Team Manager	Manages team members, permissions, and collaboration workflows.
Contributions Hub	Tracks contributor activity, recognitions, and community engagement.
Template Builder	Creates issue templates, PR templates, and other repository scaffolding.
Repo Manager	Manages repository lifecycle: creation, archival, settings, and maintenance.

Nexus & Hidden Sub-Agents (10)

These agents run behind the scenes. You never invoke them directly—other agents call them as needed.

Agent	Description
Nexus	Top-level orchestrator that coordinates across all agent teams for complex multi-domain tasks.
Document Inventory	Catalogs all documents in a project before a batch scan begins.
Cross-Document Analyzer	Finds patterns and inconsistencies across multiple documents in a batch.
Cross-Page Analyzer	Finds accessibility patterns that span multiple web pages (navigation, headers, footers).
Web Issue Fixer	Applies automated code fixes for web accessibility issues found during audits.
Web CSV Reporter	Exports web audit results to CSV format for spreadsheet analysis.
Document CSV Reporter	Exports document audit results to CSV format for spreadsheet analysis.
Markdown CSV Reporter	Exports Markdown audit results to CSV format for spreadsheet analysis.
Markdown Scanner	Scans Markdown files for accessibility issues (invoked by Markdown Assistant).
Markdown Fixer	Applies fixes to Markdown files (invoked by Markdown Assistant).

How It Works Across Platforms

Same agents, same rules, three different platforms. Pick your editor and get started.

Claude Code (CLI)

Agents live in .claude/agents/ as Markdown files. Each agent has a dedicated system prompt that defines its identity and expertise.

Specialists run in parallel groups for fast audits—the Accessibility Lead orchestrates the full team so you get comprehensive results from a single command.

Install with one command. The installer copies all 47 agents and configures everything automatically.

GitHub Copilot (VS Code)

Agents load as .agent.md files in .github/agents/. Accessibility rules are injected into every Copilot chat via copilot-instructions.md.

Same agents, same rules, same parallel execution as Claude Code. All 52 prompts are available as .prompt.md files you can invoke directly in Copilot Chat.

14 skills provide shared knowledge modules that keep all agents consistent.

Claude Desktop (Standalone App)

Uses MCP (Model Context Protocol) since Desktop does not have an agent system. 11 tools provide contrast checking, heading audits, form validation, axe-core scanning, document scanning, and VPAT generation.

6 built-in prompts cover common accessibility reviews: full audit, ARIA review, modal review, contrast review, keyboard review, and live region review.

Install by downloading the .mcpb extension bundle from the releases page.

Skills — Shared Knowledge Modules

Skills are reusable knowledge packs that multiple agents draw from. They ensure consistency—every agent that checks contrast uses the same rules.

Accessibility Skills (6)

Skill	Description
Accessibility Rules	Core WCAG 2.2 AA rules shared by all accessibility agents.
Framework Accessibility	React, Vue, Angular, Svelte, and Tailwind CSS fix templates and patterns.
Cognitive Accessibility	WCAG 2.2 cognitive criteria: reading level, error prevention, and consistent navigation.
Mobile Accessibility	Touch targets, orientation, viewport, and native mobile screen reader patterns.
Design System	Component library patterns, design tokens, and accessible component guidelines.
Web Severity Scoring	Consistent severity ratings for web accessibility issues (critical, serious, moderate, minor).

Scanning Skills (3)

Skill	Description
Document Scanning	Rules and patterns for Office, PDF, and EPUB document accessibility scanning.
Web Scanning	Patterns for axe-core integration, multi-page crawling, and result aggregation.
Markdown Accessibility	Rules for heading structure, alt text, link quality, tables, and readability in Markdown.

Reporting & Reference Skills (2)

Skill	Description
Report Generation	Severity scoring, VPAT/ACR export, scorecards, and CSV output formatting.
Help URL Reference	Curated links to WCAG documentation, techniques, and understanding documents.

GitHub Skills (3)

Skill	Description
GitHub Workflow Standards	PR review standards, issue management policies, and branching conventions.
GitHub Scanning	Repository health checks, security scanning patterns, and CI/CD integration.
GitHub Analytics Scoring	Metrics and scoring for repository health, contributor activity, and PR velocity.

Ready-to-Use Prompts

Don’t know where to start? Pick a prompt. 52 pre-built commands across four categories get you results immediately.

Web Accessibility Prompts (6)

Prompt	Description
Audit Web Page	Full WCAG 2.2 AA audit of a single web page.
Quick Web Check	Fast accessibility scan for the most critical issues.
Audit Multi-Page	Crawl and audit multiple pages across a site.
Compare Web Audits	Compare two audit results to track progress over time.
Fix Web Issues	Automatically apply fixes for discovered accessibility issues.
Export Web CSV	Export web audit results to a CSV spreadsheet.

Document Accessibility Prompts (10)

Prompt	Description
Audit Single Document	Full accessibility audit of one Office or PDF document.
Audit Document Folder	Batch audit all documents in a folder with an aggregated scorecard.
Audit Changed Documents	Audit only documents that changed since last scan.
Quick Document Check	Fast scan for the most critical document accessibility issues.
Generate VPAT	Create a VPAT 2.5 Accessibility Conformance Report from audit results.
Generate Remediation Scripts	Generate automated fix scripts for discovered document issues.
Compare Audits	Compare before-and-after document audit results.
Setup Document CI/CD	Configure continuous accessibility checking for documents in your CI pipeline.
Create Accessible Template	Generate a new document template with accessibility best practices built in.
Export Document CSV	Export document audit results to a CSV spreadsheet.

GitHub Workflow Prompts (31)

Organized by workflow area. Each prompt invokes the right agent and gets you results in one step.

Prompt	Description
Review PR	Accessibility-focused pull request review.
PR Report	Summary report of a pull request.
My PRs	List your open pull requests and their status.
PR Author Checklist	Pre-submission checklist for PR authors.
PR Comment	Add a formatted comment to a pull request.
Address Comments	Address review comments on your PR.
Manage Branches	List, create, or delete branches.
Merge PR	Merge a pull request with proper checks.
Explain Code	Get a plain-language explanation of code.
My Issues	List issues assigned to you.
Create Issue	Create a well-structured issue from a description.
Triage	Triage and label incoming issues.
Issue Reply	Draft a response to an issue comment.
Manage Issue	Update labels, assignees, and milestones on issues.
Refine Issue	Improve issue descriptions and acceptance criteria.
Project Status	Overview of project health and progress.
React	Add emoji reactions to issues or PRs.
Daily Briefing	Morning summary of overnight repo activity.
CI Status	Check CI pipeline status and recent failures.
Notifications	Review and manage your GitHub notifications.
Security Dashboard	Overview of security alerts and vulnerabilities.
Onboard Repo	Set up a new repository with best practices.
Draft Release	Create a release draft with changelog.
Release Prep	Pre-release checklist and preparation steps.
My Stats	Your personal contribution statistics.
Team Dashboard	Team-wide activity and performance metrics.
Sprint Review	Sprint retrospective and metrics summary.
A11y Update	Accessibility-specific project status update.
Add Collaborator	Add a contributor to the repository.
Build Template	Create reusable issue or PR templates.
Build A11y Template	Create accessibility-focused issue and PR templates.

Markdown Accessibility Prompts (5)

Prompt	Description
Audit Markdown	Full accessibility audit of Markdown documentation.
Quick Markdown Check	Fast scan for the most critical Markdown accessibility issues.
Fix Markdown Issues	Automatically apply fixes for Markdown accessibility issues.
Compare Markdown Audits	Compare before-and-after Markdown audit results.
Export Markdown CSV	Export Markdown audit results to a CSV spreadsheet.

Tools & Integrations

The MCP server exposes 11 tools and 6 prompts that Claude Desktop can call directly. Every tool runs locally—no data leaves your machine.

MCP Tools (11)

Tool	What It Does
check_contrast	Calculates WCAG contrast ratio between two hex colors. Reports pass/fail for normal text (4.5:1), large text (3:1), and UI components (3:1).
get_accessibility_guidelines	Returns WCAG AA requirements, code examples, and common mistakes for a specific component type (modal, tabs, accordion, combobox, carousel, form, live-region, navigation, or general).
check_heading_structure	Analyzes HTML for heading hierarchy issues—single H1, skipped levels, and empty headings.
check_link_text	Scans HTML for ambiguous link text like “click here” or “read more”, URLs used as link text, repeated identical text, missing new-tab warnings, and links to non-HTML resources without file type indication.
check_form_labels	Checks that every input, select, and textarea has a programmatic label (via <label for>, aria-label, or aria-labelledby). Also checks required attributes, fieldset/legend grouping, and autocomplete on identity fields.
run_axe_scan	Runs axe-core against a live URL (e.g. localhost). Returns WCAG 2.2 AA violations grouped by severity with affected elements and fix suggestions. Optionally writes a Markdown report.
generate_vpat	Generates a VPAT 2.5 / Accessibility Conformance Report (ACR) in Markdown from audit findings. Outputs per-criterion conformance levels and remarks.
scan_office_document	Scans .docx, .xlsx, or .pptx files using Microsoft Accessibility Checker rule sets. Supports configurable rules, severity filters, and SARIF output for CI.
scan_pdf_document	Scans PDF files against PDF/UA (Matterhorn Protocol) with checks for tagging, structure tree, language, alt text, bookmarks, and forms. Supports SARIF output.
extract_document_metadata	Extracts accessibility-relevant metadata (title, author, language, page count, accessibility property health) from Office or PDF files.
batch_scan_documents	Scans up to 50 Office and PDF files in a single call. Returns an aggregated summary with per-file results, cross-document patterns, and an overall accessibility scorecard.

MCP Prompts (6)

Prompt	What It Does
accessibility-audit	Runs a comprehensive audit covering structure, ARIA, keyboard navigation, contrast, focus management, and live regions.
aria-review	Reviews ARIA roles, states, and properties. Enforces the first rule of ARIA: don’t use it if native HTML works.
modal-review	Reviews a modal, dialog, drawer, or overlay for focus trapping, focus return, escape behavior, and heading structure.
contrast-review	Reviews color choices and CSS for WCAG AA contrast compliance, including dark mode and Tailwind utilities.
keyboard-review	Reviews keyboard accessibility: tab order, focus management, skip links, keyboard traps, and arrow-key patterns.
live-region-review	Reviews dynamic content for screen reader announcements: live regions, toasts, loading states, and form feedback.

axe-core Integration

The run_axe_scan tool calls axe-core against any live URL. Results feed directly into agent audit reports and VPAT generation.

VPAT 2.5 Generation

The generate_vpat tool produces Accessibility Conformance Reports in Markdown—no spreadsheet required.

Document Scanning

Scan Office files (.docx, .xlsx, .pptx) and PDFs locally. Supports configurable rule sets, severity filters, and SARIF output for CI pipelines.

CSV Export

Export audit results from web, document, and Markdown scans to CSV for analysis in Excel, Google Sheets, or any data tool.

The Repository

Community-Access/accessibility-agents

47 accessibility agents, 14 skills, and 52 prompts for Claude Code, GitHub Copilot, and Claude Desktop. Enforce WCAG 2.2 AA compliance so AI coding tools stop generating inaccessible code.

MIT License
47 agents
14 skills
52 prompts

View Repository Getting Started Research Sources

Install in One Command

Works on macOS, Linux, and Windows. Installs all agents for every supported platform. Supports --global, --project, and --copilot flags.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/Community-Access/accessibility-agents/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/Community-Access/accessibility-agents/main/install.ps1 | iex

Safe, additive installation. Your existing files are never overwritten. Auto-update support included. Read the full guide.

Build With Us

Community Access is a community project. The more perspectives, lived experiences, and domain knowledge that go into it, the better it serves the blind and low vision community.

Every contribution ships as a pull request. Fork the repo, make your change, open a PR. Agents, skills, and prompts are all plain Markdown—no special tooling required.

Fork the Repo

Fork accessibility-agents to your GitHub account. Clone it locally or use github.dev.

Make Your Change

Fix an agent gap, add a new rule, improve docs, or build an entirely new specialist. Everything is Markdown.

Open a Pull Request

Submit your PR against main. The team reviews every submission and provides feedback.

Ship It

Once merged, your contribution is available to every user on next install or update. You are now a contributor.

Found an agent gap? Know a pattern we should catch? Have lived experience with assistive technology?

We want your pull request.

Fork the Repository Browse Open Issues Contributing Guide

Special Thanks

A special thanks goes out to the following contributors and to those people that have truly shaped what it means to engage with the blind and low-vision community. Thank you everyone for your partnership, mentorship, commitments, abilities, and most importantly your empathy centered around digital accessibility and inclusion. Go team!

Taylor Arndt
Michael Babcock
Jeff Bishop
Clint Covington
Michael Doise
Doug Geoffray
Maria Lamardo
Ty Littlefield
Roberto Perez
Megan Rogge

See It in Action

Here is a real example. An AI tool generates the code on the left. Our agent catches the issues and rewrites it on the right.

Before — AI-Generated Code

<!-- Missing alt, generic link text,
     no form label, wrong heading order -->
<h1>Welcome</h1>
<h3>Our Team</h3>
<img src="team.jpg">
<a href="/about">Click here</a>
<input type="email" placeholder="Email">

After — Agent-Fixed Code

<!-- Heading hierarchy, alt text,
     descriptive link, labeled input -->
<h1>Welcome</h1>
<h2>Our Team</h2>
<img src="team.jpg" alt="Our team at the 2025 summit">
<a href="/about">Learn about our team</a>
<label for="email">Email</label>
<input id="email" type="email" autocomplete="email">

Frequently Asked Questions

Does this work with VS Code?

Yes. The agents work inside GitHub Copilot in VS Code and VS Code Insiders. They also work in Claude Code (terminal) and Claude Desktop (standalone app). The install command sets up all three platforms automatically.

Does any data leave my machine?

The MCP tools all run locally. Document scanning, contrast checking, heading analysis, and axe-core scans happen entirely on your machine. The only network calls are the ones your AI platform already makes to its own API.

Can I use only some agents?

Yes. Each agent is independent Markdown. You can delete any agent file you do not need, or use the --project flag to install only at the project level. You can also disable individual agents in your platform settings.

How do I update?

Run the same install command again. It will pull the latest agents, skills, and prompts. Your existing configuration is never overwritten. You can also enable auto-update during installation.

What WCAG version do the agents target?

All web agents target WCAG 2.2 Level AA. Document agents map to WCAG 2.1 AA plus Microsoft Office Accessibility Checker rules. PDF agents follow PDF/UA (ISO 14289-1) via the Matterhorn Protocol.

Does this replace manual accessibility testing?

No. Automated tools catch many issues but cannot replace testing with actual screen readers (NVDA, JAWS, VoiceOver) and keyboard-only navigation. These agents are a powerful first pass, not a substitute for real-world testing with assistive technology.

Is this free?

Yes. Community Access is open source and licensed under MIT. Every agent, skill, prompt, and tool is free to use, modify, and distribute.

Latest News

Press releases, announcements, and updates from the project.

Loading news...

View All News

AI and automated tools are not perfect. They miss things, make mistakes, and cannot replace testing with real screen readers and assistive technology. Always verify with VoiceOver, NVDA, JAWS, and keyboard-only navigation. This tooling is a helpful starting point, not a substitute for real accessibility testing.