uisight MCP

The uisight MCP server opens mobile and desktop versions of a site and reports contrast and button sizes as exact numbers.

  • 130 ★ on the repository
  • Last commit
MCP server

Extended access

Why this level

  • The agent drives a browser: it opens addresses, including localhost, taps buttons, types text and reads console and network errors.
  • Frames and error messages carry the content of the opened pages.
All reasons and checks

sololabstr/uisight

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add mcp/uisight

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The agent drives a browser: it opens addresses, including localhost, taps buttons, types text and reads console and network errors.
  • Frames and error messages carry the content of the opened pages.

This is third-party code. Review the repository files before installing.

What it does

uisight runs two live sessions of one site side by side, mobile and desktop, and measures the interface. The `inspect` tool returns findings as text with an element selector: for example, invisible text with 1.04:1 contrast or a button smaller than 44 pixels. A human watches next to the agent. The panel at `localhost:5055` shows the same screens, and a pinned note with a frame reaches the agent through `marks`. The other tools are `see_screen`, `goto`, `tap`, `type_text`, `scroll`, `set_device` and `status`. Device profiles include iPhone, Pixel, iPad and a laptop, with light and dark themes. Checks cover WCAG AA contrast, tiny text, images without alt, clipped text, covered elements and fields hidden under the on-screen keyboard.

Who it is for. Front-end developers and designers who check responsive interfaces together with an agent.

Good fit when

  • An agent fixes layout and receives contrast numbers with selectors.
  • You need to compare the interface across several devices in light and dark themes in a single run.
  • A human spots a bug on screen and pins a note with a frame for the agent to read.
  • Before a release you check buttons under 44 pixels and controls under a bar or the keyboard.

Not a fit when

  • You test native iOS or Android apps: uisight works only with the web.
  • A single screenshot and a model's look are enough and no measurements are needed.
  • The machine has no room for a browser: Chromium takes about 700 MB.

Example request

Look at my app through uisight on iPhone and desktop in both themes and fix the contrast problems you find.

MCP

Transport
stdio
Authentication
not required

Environment variables

UISIGHT_NO_INSTALL
A value of 1 turns off the offer to download the browser.
UISIGHT_LANG
A value of tr switches tool names to Turkish.

Limitations

Playwright downloads browsers separately, and the first run needs Chromium of about 700 MB. In a terminal uisight offers to fetch it, while in CI and when launched from an agent it downloads nothing and names the command to run. The `UISIGHT_NO_INSTALL=1` variable turns the offer off everywhere. For the real iOS Safari engine you add webkit to Chromium with a Playwright command. Native apps are not covered by the checks.

How to disable. Remove the uisight entry from the MCP client configuration, for Claude Code run `claude mcp remove uisight`. Delete downloaded Playwright browsers separately.

FAQ

How do I connect uisight to Claude Code?

Run `claude mcp add --scope user uisight -- npx -y -p uisight@latest uisight-mcp`. For Cursor and other MCP clients the README has a JSON configuration with the same `npx`.

How do I install uisight in Claude Desktop?

Download `uisight-<version>.mcpb` from the releases page and open the file. Claude Desktop asks for the page, the tool set and whether to download the browser on first use.

What does inspect show?

Findings per device and theme: contrast, invisible text, touch targets under 44 pixels, clipped text and covered controls. Each finding comes with a selector.

How do I run a one-off audit without MCP?

Run `npx uisight https://yourapp.com --theme both`. The command saves PNGs, a gallery and a report for iPhone, Pixel and desktop in light and dark themes.

What does the panel at localhost:5055 show?

Mobile and desktop sessions side by side. A human clicks, scrolls and pins a note with a frame, and the agent reads it through `marks`.

Discussion

No questions or reviews yet.

An accessibility scanner for agents and CI: renders a page in Chromium with axe-core or lints React, Vue and Angular source.

MCP serverMinimal access117Repository stars

Console, network, screenshots and Lighthouse audits from your open Chrome session to the agent via a DevTools extension

MCP serverBroad access7.3KRepository stars

A skill and CLI for agents: UI without the templated look, 25 commands, rule-based anti-slop checks and a 0 to 100 quality score

SkillExtended access432Repository stars

An MCP server and skill for interfaces: 31 design systems for shadcn/ui, UI and UX principles and accessibility review for agent-written code

MCP serverExtended access556Repository stars

A UI design skill: three genuinely different directions side by side, one chosen and refined against real screenshots in agents like Claude Code

SkillExtended access2.8KRepository stars

Skills for design engineers: quick UI cleanup, accessibility, metadata and animation performance fixes

SkillMinimal access10.7KRepository stars

Comparison with similar tools

Cards that share the main search query, and how they differ.

ToolOfficialAuthenticationRepository starsMain difference
uisight (this card)nonot required130
Playwright MCPyesnot required39,203drives a browser through the accessibility tree; uisight measures contrast and sizes, runs mobile and desktop sessions side by side and shares the screen with a human
Accessibility Agentsno434agents and skills carrying WCAG 2.2 AA requirements; uisight computes values on a live page and returns them with selectors

README in short

The author notes that a screenshot and a measurement spend about the same tokens: a frame is roughly 460 and an inspect result roughly 570. The saving comes from a one-off report: the model reads it once, while stepping through screens with tools resends the whole conversation at every step.

README badge

Report a problem

Foxx AIuisight
Foxx AI

I am Foxx AI and I have already vetted this tool. Ask about install, setup or anything else, and I will keep it simple.