a11y

Agent A11y (Agent Accessibility)

Go accessibility auditing toolkit for WCAG 2.0, 2.1, and 2.2 compliance testing.

Features

• ✅ WCAG Compliance Testing - Supports WCAG 2.0, 2.1, and 2.2 at levels A, AA, and AAA
• 🔍 Multiple Audit Modes - Single page, site crawling, and user journey testing
• 🪓 axe-core Integration - Industry-standard accessibility testing via axe-core
• 🤖 LLM-as-a-Judge - Optional AI evaluation to reduce false positives
• 📊 Report Formats - JSON, HTML, Markdown, VPAT 2.4, WCAG-EM, CSV
• 🔌 MCP Server - Model Context Protocol integration for AI assistants
• 🌐 HTTP API - REST API for programmatic access
• 🤝 Multi-Agent Spec - Integration with multi-agent workflows

Installation

go install github.com/plexusone/agent-a11y/cmd/agent-a11y@latest

Prerequisites

• Go 1.25+
• Chrome/Chromium browser (for browser-based testing)

Quick Start

CLI Usage

# Audit a single page
agent-a11y audit https://example.com

# Audit with site crawling
agent-a11y audit https://example.com --crawl --depth 2

# Specify WCAG level and version
agent-a11y audit https://example.com --level AA --version 2.2

# Output to file with format
agent-a11y audit https://example.com -o report.html --format html

# Enable LLM evaluation (reduces false positives)
agent-a11y audit https://example.com --llm-provider anthropic --llm-model claude-sonnet-4-20250514

Go Library

package main

import (
    "context"
    "fmt"
    "log"

    a11y "github.com/plexusone/agent-a11y"
)

func main() {
    // Create auditor
    auditor, err := a11y.New(
        a11y.WithLevel(a11y.LevelAA),
        a11y.WithVersion(a11y.Version22),
    )
    if err != nil {
        log.Fatal(err)
    }
    defer auditor.Close()

    // Audit a page
    result, err := auditor.AuditPage(context.Background(), "https://example.com")
    if err != nil {
        log.Fatal(err)
    }

    fmt.Println(result.Summary())
    // WCAG 2.2 Level AA: Conformant (Score: 95/100, Issues: 0 critical, 0 serious, 1 moderate, 2 minor)
}

CLI Commands

Output Formats

Configuration

Create a config file at ~/.agent-a11y.yaml or specify with --config:

wcag:
  level: AA
  version: "2.2"

browser:
  headless: true
  timeout: 30s

llm:
  enabled: true
  provider: anthropic
  model: claude-sonnet-4-20250514

crawl:
  depth: 2
  maxPages: 50
  delay: 500ms

LLM Providers

agent-a11y supports multiple LLM providers for AI-assisted evaluation:

MCP Server

Run as an MCP server for integration with AI assistants:

agent-a11y mcp serve

Add to your Claude Desktop config:

{
  "mcpServers": {
    "a11y": {
      "command": "agent-a11y",
      "args": ["mcp", "serve"]
    }
  }
}

HTTP API

Start the API server:

agent-a11y serve --port 8080

Endpoints:

• POST /audit - Run an accessibility audit
• GET /health - Health check

Multi-Agent Integration

agent-a11y integrates with the multi-agent-spec for use in multi-agent workflows:

result, _ := auditor.AuditPage(ctx, url)

// Convert to AgentResult for multi-agent workflows
agentResult := result.AgentResult()

// Get narrative for reports
narrative := result.Narrative()

Examples

See the examples directory:

• basic - Simple single-page audit
• multi-agent - Multi-agent workflow integration

Development

# Run tests
go test -v ./...

# Run linter
golangci-lint run

# Build CLI
go build -o agent-a11y ./cmd/agent-a11y

License

MIT