Give your AI agent real design

How it works

UXKIN pairs an agent skill with a remote MCP server. The no-ui-slop skill sets how your agent approaches UI work, and the server gives it two tools: find_ui_references and find_ui_materials. One setup prompt works for Claude Code, Codex, Cursor and any MCP client that connects over HTTP.

1. 01Connect

   Copy the setup prompt from the Agent SDK tab and paste it into your agent. It installs the no-ui-slop skill and adds https://uxkin.com/mcp with Authorization: Bearer your-token. Keep the token in your agent's local settings or an environment variable. Never put it in code you commit, a link, a screenshot or a log. If it ever leaks, reset it and the old one stops working.

2. 02Understand

   Before changing any UI, the agent reads what's already there: the components, shared styles and the screens around the change. It works out the one thing the person using the screen is trying to do, and keeps good existing work. For a new web project with no design yet, it can suggest up to three Web references for you to choose from. Your pick becomes the project's DESIGN.md; an existing one is never replaced without asking.

3. 03Reference

   When a real example can settle a specific decision, like how to lay out a checkout or an empty state, find_ui_references searches real apps, screens, journeys and websites. One or two strong matches beat a long list. References guide decisions; they aren't templates to copy.

4. 04Materials

   When the design needs a concrete piece it doesn't have yet, find_ui_materials returns colors, typography, components and DESIGN.md content from real web design systems. The agent works from one consistent set of choices instead of inventing new ones for every screen, and it never stacks a second set of rules on top of your existing design system.

5. 05Build

   The agent builds something that feels made for this product, with the main task obvious and easy. Your requirements, your existing design system, accessibility, platform conventions and anything you ask for directly always come first.

6. 06Review

   Once the UI runs, the agent looks at the real result, not just the code. It fixes broken layouts, regressions, contrast and focus problems, anything private showing on screen, and anything that gets in the way of the main task. It checks what people will actually see.