Accessibility MCP Server
# Accessibility MCP Server
An MCP (Model Context Protocol) server that provides accessibility auditing tools for LLMs through Cursor. Integrates axe-core (via Playwright), Lighthouse CLI, and optional WAVE API to deliver deterministic, machine-actionable accessibility audit results.
## Features
- **axe-core Integration**: Automated accessibility testing via Playwright with full WCAG 2.1/2.2 coverage
- **Lighthouse CLI**: Comprehensive accessibility audits with performance insights
- **WAVE API**: Optional integration for additional accessibility validation
- **Deterministic Output**: All results normalised to explicit pass/fail/unknown outcomes
- **WCAG Compliance**: Every finding mapped to WCAG 2.1/2.2 criteria with severity classification
- **Machine-Actionable**: Results include selectors and DOM context for remediation
- **Type-Safe**: Full TypeScript implementation with strict type checking
## Installation
```bash
# Clone the repository
git clone <repository-url>
cd accessibility-mcp
# Install dependencies
npm install
# Build the project
npm run build
# Install Playwright browsers (required for axe-core)
npx playwright install chromium
```
## Configuration
Create a `.env` file in the project root (see `.env.example` for template):
```env
# WAVE API Configuration (optional)
WAVE_API_KEY=your_wave_api_key_here
WAVE_API_URL=https://wave.webaim.org/api/request
# Lighthouse Configuration (optional)
LIGHTHOUSE_TIMEOUT=60000
LIGHTHOUSE_CHROME_FLAGS=--headless --no-sandbox
# Playwright Configuration (optional)
PLAYWRIGHT_BROWSER=chromium
PLAYWRIGHT_TIMEOUT=30000
```
## Usage
### As MCP Server in Cursor
1. Build the project: `npm run build`
2. Configure in Cursor's MCP settings (`~/.cursor/mcp.json` or Cursor settings):
```json
{
"mcpServers": {
"accessibility": {
"command": "node",
"args": ["/absolute/path/to/accessibility-mcp/dist/index.js"]
}
}
}
```
3. Restart Cursor to load the MCP server
### Available MCP Tools
#### `axe_audit`
Run an accessibility audit using axe-core via Playwright. Supports URLs, local file paths, and localhost URLs.
**Parameters:**
- `url` (required): The URL to audit. Can be:
- HTTP/HTTPS URL: `"https://example.com"` or `"http://localhost:3000"`
- Local file path: `"./index.html"` or `"/path/to/file.html"`
- File protocol URL: `"file:///path/to/file.html"`
- `options` (optional): Configuration object
- `tags`: Array of WCAG tags (e.g., `["wcag2a", "wcag2aa"]`)
- `rules`: Object with rule-specific configuration
- `timeout`: Timeout in milliseconds (default: 30000)
- `browser`: Browser to use - `"chromium"`, `"firefox"`, or `"webkit"` (default: `"chromium"`)
**Examples:**
```json
{
"name": "axe_audit",
"arguments": {
"url": "https://example.com",
"options": {
"tags": ["wcag2a", "wcag2aa"],
"browser": "chromium"
}
}
}
```
```json
{
"name": "axe_audit",
"arguments": {
"url": "./src/index.html"
}
}
```
```json
{
"name": "axe_audit",
"arguments": {
"url": "http://localhost:3000"
}
}
```
#### `lighthouse_audit`
Run an accessibility audit using Lighthouse CLI. Supports URLs, local file paths, and localhost URLs.
**Parameters:**
- `url` (required): The URL to audit. Can be:
- HTTP/HTTPS URL: `"https://example.com"` or `"http://localhost:3000"`
- Local file path: `"./index.html"` or `"/path/to/file.html"`
- File protocol URL: `"file:///path/to/file.html"`
- `categories` (optional): Array of Lighthouse categories to include
- `options` (optional): Configuration object
- `onlyCategories`: Array of categories to include
- `skipAudits`: Array of audit IDs to skip
- `timeout`: Timeout in milliseconds (default: 60000)
**Examples:**
```json
{
"name": "lighthouse_audit",
"arguments": {
"url": "https://example.com",
"categories": ["accessibility"]
}
}
```
```json
{
"name": "lighthouse_audit",
"arguments": {
"url": "./dist/index.html"
}
}
```
#### `wave_audit`
Run an accessibility audit using WAVE API (requires API key). Supports URLs and localhost URLs. Local files are automatically served via a temporary local server.
**Parameters:**
- `url` (required): The URL to audit. Can be:
- HTTP/HTTPS URL: `"https://example.com"` or `"http://localhost:3000"`
- Local file path: `"./index.html"` (will be served via temporary server)
- `apiKey` (optional): WAVE API key (uses `WAVE_API_KEY` env var if not provided)
**Examples:**
```json
{
"name": "wave_audit",
"arguments": {
"url": "https://example.com",
"apiKey": "your_api_key"
}
}
```
```json
{
"name": "wave_audit",
"arguments": {
"url": "./src/index.html"
}
}
```
**Note:** For local files, the MCP server automatically creates a temporary HTTP server to serve the file, as WAVE API requires HTTP/HTTPS URLs.
### Output Format
All tools return normalised results in the following format:
```typescript
{
url: string;
tool: 'axe' | 'lighthouse' | 'wave';
timestamp: string;
results: Array<{
rule_id: string;
wcag_ref: string[];
severity: 'critical' | 'serious' | 'moderate' | 'minor';
confidence: 'high' | 'medium' | 'low';
outcome: 'pass' | 'fail' | 'unknown';
selector: string;
dom_context: string;
message: string;
reason_code?: string; // Required when outcome is 'unknown'
}>;
summary: {
total: number;
pass: number;
fail: number;
unknown: number;
by_severity: {
critical: number;
serious: number;
moderate: number;
minor: number;
};
};
}
```
## Development
### Prerequisites
- Node.js 18.0.0 or higher
- npm or yarn
### Scripts
```bash
# Development mode (with hot reload)
npm run dev
# Build TypeScript to JavaScript
npm run build
# Run all tests
npm test
# Run unit tests only
npm run test:unit
# Run integration tests only
npm run test:integration
# Type check without building
npm run type-check
# Lint code
npm run lint
```
### Project Structure
```
accessibility-mcp/
├── .cursor/
│ ├── rules/ # Cursor behavioural constraints
│ └── commands/ # Cursor repeatable actions
├── src/
│ ├── adapters/ # External tool adapters (axe, Lighthouse, WAVE)
│ ├── services/ # Business logic (audits, normalisation)
│ ├── tools/ # MCP tool definitions
│ ├── types/ # TypeScript type definitions
│ ├── utils/ # Utility functions (WCAG mappings, selectors)
│ ├── index.ts # Entry point
│ └── server.ts # MCP server setup
├── tests/
│ ├── fixtures/ # Test data fixtures
│ ├── unit/ # Unit tests
│ └── integration/ # Integration tests
└── dist/ # Compiled JavaScript output
```
### Architecture
The server follows a clean architecture pattern:
1. **Tools Layer** (`src/tools/`): MCP tool definitions that expose audit capabilities
2. **Services Layer** (`src/services/`): Business logic for audit execution and result normalisation
3. **Adapters Layer** (`src/adapters/`): Interface layer for external tools (axe, Lighthouse, WAVE)
4. **Types Layer** (`src/types/`): Shared type definitions for audit results and normalised outputs
### Adding a New Audit Tool
See `.cursor/commands/add-new-audit-tool.md` for detailed instructions on adding a new accessibility audit tool.
### Testing
```bash
# Run all tests
npm test
# Run with coverage
npm test -- --coverage
# Run specific test file
npm test tests/unit/normaliser.test.ts
```
## CI/CD Integration
The project includes commands for CI/CD integration. See `.cursor/commands/ci-gate-accessibility.md` for details on setting up accessibility gates in your CI pipeline.
## Contributing
1. Follow the architectural constraints defined in `.cursor/rules/`
2. Ensure all code passes type checking: `npm run type-check`
3. Write tests for new features
4. Ensure deterministic outputs (see `.cursor/rules/determinism.md`)
5. Map all findings to WCAG criteria (see `.cursor/rules/accessibility.md`)
## License
MIT
## Acknowledgments
- [axe-core](https://github.com/dequelabs/axe-core) - Accessibility testing engine
- [Lighthouse](https://github.com/GoogleChrome/lighthouse) - Web performance and accessibility auditing
- [WAVE](https://wave.webaim.org/) - Web accessibility evaluation tool
- [Playwright](https://playwright.dev/) - Browser automation framework
- [Model Context Protocol](https://modelcontextprotocol.io/) - Protocol for LLM tool integration
TDQS
Scored across 3 tools
All three tools perform accessibility audits with overlapping purposes, making them highly ambiguous. An agent would struggle to choose between axe_audit, lighthouse_audit, and wave_audit since they all audit accessibility, differ only in the underlying engine (axe-core, Lighthouse, WAVE), and have nearly identical input support (URLs, local files). This overlap creates significant confusion without clear guidance on when to use each.
Tool names follow a perfectly consistent pattern: all use snake_case with the format 'engine_audit' (axe_audit, lighthouse_audit, wave_audit). This predictable naming makes it easy to understand each tool's purpose at a glance, with no deviations or mixed conventions.
With only 3 tools, the count feels thin for an accessibility server, as it lacks broader functionality like reporting, remediation suggestions, or compliance checks. However, it's borderline reasonable for a focused audit toolset, though it could benefit from additional tools to enhance utility beyond just running audits.
The server is severely incomplete for accessibility testing, offering only audit execution without any tools for analysis, reporting, or follow-up actions. There are no tools to generate reports, track issues, suggest fixes, or validate compliance, leaving significant gaps that will hinder agents in performing comprehensive accessibility workflows.