AI assistants (MCP)

Let your AI assistant check and fix your layout

AI assistants write UI fast but never see it on a phone. Connect UICheck through MCP and they can: check the page at real screen sizes, read each problem with the CSS fix UICheck already tested on the live page, change the code and check again until it passes. Works in Claude, Cursor, VS Code, Windsurf and ChatGPT, and on localhost.

Start free Setup guide

Included in Every plan · Free plan to try UICheck, no card needed

13 tools: check, read, fix, re-check
Local or hosted: your terminal or uicheck.app/mcp
You choose: read, run or manage

01

The whole fix loop, done by your assistant

Ask in plain words. The assistant starts the check, reads each finding with its picture, edits your code and re-checks until the quality gate passes, then tells you what changed.

Works on localhost while you build
Progress while a check runs
The report link for your team

02

Every finding comes with its fix

Findings arrive with the element, what a visitor sees, how to fix it and often the exact CSS UICheck tried on the live page and confirmed works.

Grouped across screen sizes
WCAG criteria on accessibility findings
Cropped screenshot with the problem outlined

How it works

Set up in minutes

1
Connect it
In claude.ai or ChatGPT add uicheck.app/mcp and sign in. In Claude Code, Cursor or VS Code add one line: npx @uicheck-cli/cli mcp.
2
Ask
“Check localhost:3000 on phone and desktop and fix what you find.” The assistant starts the check and reads every finding with its screenshot.
3
It fixes and confirms
It changes your code, re-checks and tells you what is fixed, what is new and the new score, with the report link.

Set it up: the guide for Claude, Cursor, VS Code, Windsurf and ChatGPT

In every check

What it catches

Layout bugs the assistant just wrote: overflow, cut-off text, overlaps, tap targets
Accessibility problems with their WCAG criteria and a tested fix
Proof it worked: fixed / new / still there after the re-check
Localhost and staging, before anything is pushed
With your permission: findings marked fixed, tracker issues created, comments answered

Questions

What is MCP?

The Model Context Protocol: the standard way AI assistants use outside tools. UICheck’s MCP server gives an assistant tools to run checks, read findings, re-check and, if you allow it, manage the To fix list.

Which assistants work?

Claude (claude.ai, Desktop and Claude Code), Cursor, VS Code with Copilot, Windsurf and ChatGPT. Anything that speaks MCP over stdio or HTTP works.

Can it check localhost?

Yes, with the local server (npx @uicheck-cli/cli mcp without a token): checks then run on your own computer. The hosted address checks public and staging sites.

What can the assistant change?

Only what you allow when connecting: read checks, also run them, or also mark findings, create tracker issues and reply to comments. Disconnect it any time in Integrations.

Does it use my daily checks?

Yes, checks an assistant starts count like any other; Prism has no daily cap. Checks on your own computer without a token are free and unlimited.