a11y-mcp Agentic Loop

Most accessibility tools are one-shot. You run them, they print a report, you read the report, you maybe fix some things, you run the tool again. That cycle is fine for humans but it's a strange waste of an AI coding agent, which would happily run the loop autonomously if you let it. a11y-mcp by priyankark is built for exactly that case.

It's an axe-core MCP server designed for "agentic loops" — workflows where an AI coding tool like Amp, Cline, Cursor, or GitHub Copilot can scan a page, get a structured list of violations, write the fix, save the file, re-scan, and repeat until the report is clean. The server is intentionally lightweight and fast so the loop doesn't bog down. The output is structured the way coding agents expect, not the way humans expect.

The audience is developers who use AI coding agents seriously — not as autocomplete but as something closer to a junior engineer who can run tools, read output, and decide what to do next. For those workflows, accessibility has historically been a manual gate. With this server installed, it can become a closed-loop pass: "make this page accessible" becomes a single instruction, and the loop runs until axe-core has nothing left to flag.

This is not the only axe-core MCP. There's also ronantakizawa/a11ymcp, which is already in this catalog and which serves a different audience — humans driving a conversation. What makes priyankark's a11y-mcp distinct is the agentic-loop framing: it's tuned for the case where the AI is making the calls, not the human.

Made by priyankark. The <span class="whitespace-nowrap">a-gnt</span> Community catalogs it because the rise of agent-native coding tools means accessibility tooling needs to fit into autonomous loops, not just one-shot human workflows.

Pair with The Design Handoff Assistant for the spec side, or The Accessibility Auditor for the human review at the end of the loop.