MyA11yReport MCP Brings Automated Accessibility Audits to AI Coding Agents
Building accessible websites requires more than visual inspection. Screen reader compatibility, keyboard navigation, color contrast, and semantic structure are all checks that developers often skip under deadline pressure. A new tool called MyA11yReport MCP aims to change that by giving AI coding agents the ability to audit a website's accessibility on their own.
MyA11yReport MCP is a local Model Context Protocol server. It connects to an AI agent over stdio and provides a set of tools for driving a real Chromium browser, running automated accessibility scans, and inspecting the structural and visual properties of web pages. The agent can navigate, click, type, and screenshot pages just as a human tester would, but it does so through code rather than manual interaction.
How the Audit Engine Works
The system combines two technologies under the hood. Playwright drives a real Chromium browser instance, which means the agent can interact with dynamic content behind accordions, modals, and menus that would be invisible to a simple HTTP request. axe-core performs automated scans against the WCAG standard, producing violation counts, per-rule details, and specific target selectors for every issue it finds.
The agent can inspect page structure including landmark regions, heading hierarchies, nested lists, and frames. It reads tab and focus order, flagging instances where a positive tabindex breaks the natural navigation sequence. For images, it extracts every img and inline SVG element along with its alt text and source. Color contrast checks run against WCAG 2.2 ratios without needing a browser at all, so the agent can verify text readability against any two colors instantly.
Session-Based Architecture
Every interaction with the browser happens through explicit sessions. The agent opens a session, receives a session ID, and passes that ID to every subsequent call until it closes the session when finished. Every action taken during a session is logged to a session history file, creating a complete record of what the agent did and what it found.
This logging is important for accountability. Unlike a black-box audit, the session history gives developers a full trace of how the agent navigated the page, what it clicked, what it typed, and which accessibility checks it performed. If a bug in the audit process is suspected, the history file provides the evidence to investigate.
Installation and Integration
Installing MyA11yReport MCP requires Node.js 20 or newer. The server is installed globally through npm with a single command, and a matching Chromium build of approximately 150 megabytes is downloaded once as a separate step. A quick echo command verifies that the server is running before any agent attempts to use it.
The tool integrates with several popular AI coding environments. For opencode users, a configuration entry in opencode.json registers the server as a local MCP client. Claude Desktop users edit claude_desktop_config.json on their system to add the same server definition. Claude Code users add it directly from the terminal with a single command. Any MCP client that supports local stdio servers can connect using the same command format.
For users who prefer not to install packages globally, the tool supports an npx-based invocation that runs the server without a permanent installation. The Chromium browser download step still needs to happen once, regardless of the installation method chosen.
What the Agent Does During an Audit
When given a URL or a sitemap, the agent follows a structured workflow. It calls the audit guide first, which provides the full step-by-step process as a markdown document. The guide walks the agent through an automated axe-core scan followed by separate reviews of page structure, tab order, image alt text, and color contrast.
Critically, the guide is honest about what machines cannot check. Keyboard navigation compatibility, screen reader behavior, and the semantic meaning of images all require human judgment. The agent performs the automated portions and flags the manual checks that remain. This division of labor means developers get the benefit of comprehensive automated testing without false confidence that every accessibility requirement has been satisfied.
Why Local Execution Matters
Running entirely on the host machine is not just a convenience feature. No account creation, no API key, and no data leaving the computer means that accessibility audits of proprietary or sensitive web applications can be performed without exposing any page content to external services. The agent processes everything locally, which also eliminates network latency and removes dependency on third-party availability.
For teams building accessibility into their development workflow, MyA11yReport MCP offers a way to make accessibility auditing an automatic part of the coding process rather than a separate manual review step. The agent can test a page immediately after changes are made, catch regressions early, and produce detailed reports that developers can act on directly within their existing coding environment.