Usability Test MCP

What it does

Point it at a live site, a localhost build or a Figma flow and give it a task, like "find the monthly price and start a free trial". An AI test participant works through the product in a real browser, one step at a time. You get a short report: the three fixes that matter most, each with a screenshot of where it happened.

How it works

• A clean participant. Each test runs in a separate session with no access to your code, files or earlier findings, so it meets the product the way a first-time visitor would.
• A real browser. The participant uses a visible Chromium window and takes one action at a time, so you can watch what it does.
• A short report. At most three fixes, each backed by a screenshot. It also runs automated accessibility checks and keyboard journeys.
• No API keys. It runs on your existing Claude or ChatGPT subscription.

What it isn't

It's a quick check before real users, not a replacement for them. The participants are AI, so the findings are model judgements rather than user research, and the accessibility checks are not a WCAG audit. If you need to know what your actual customers struggle with, that's what a UX audit is for.

Try it

It installs with one command in Claude Code or Codex (Node 22 or later) — copy it from the box below.

Then run /usability-test followed by a URL, a localhost address or a Figma link. It's free for personal use; see the repo on GitHub for setup details and licensing.