ui.sh is a toolkit designed to help AI coding agents get better at designing and building UIs. It is not positioned as a traditional component library or a visual website builder, but as a set of design and coding rules that agents such as Claude Code, Amp, Cursor, OpenCode, and Codex can reference. Created by the team behind Tailwind CSS and Refactoring UI, it focuses on common frontend issues that arise when AI writes UI code, including color, spacing, typography, responsive design, accessibility, and component structure.
According to the documentation, the main entry point after installation is using /ui in your prompt. The examples cover adding dark mode, improving mobile responsiveness, cleaning up Tailwind CSS utility classes, generating multiple design concepts, trying different heading fonts, and breaking large pages into reusable components. Under the hood, it is described as a database of UI best practices that agents can intelligently reference for relevant tasks. It is especially friendly to teams using Tailwind CSS, as the documentation explicitly says it optimizes class names according to the Tailwind team’s recommended approach.
Installation is done with npx @uidotsh/install your-token-here, so the overall barrier to entry is fairly low. Cursor requires enabling the uidotsh MCP connection in settings; Claude desktop requires adding a custom connector and configuring the remote MCP server URL. The documentation also notes that the token environment variable must be exposed in the shell, rather than only placed in a .env file. The official claim is that the rules work with any AI coding agent, but testing has mainly focused on Claude Code and Codex, with the best results coming from Claude Opus 4.6.
The current page only offers “Request an invite” and does not disclose pricing, free quotas, team plans, or enterprise support, so commercial predictability is limited. There is also no visible information about licensing, self-hosting, or data handling. The documentation only mentions a remote MCP server, so it is not yet possible to confirm whether private deployment is supported.
Its strengths are that it plugs directly into developers’ existing AI coding workflows and focuses on real frontend quality rather than simply generating disposable prototypes. The documentation examples are also concrete, with clear troubleshooting information. The downside is that it depends on high-quality models; the official docs also note that Codex performs poorly at UI design. Without access to models such as Claude Opus, results may be weaker. It is best suited for developers and small teams using AI coding agents to build Web UIs, especially Tailwind CSS projects.
The main content does not provide information about access from mainland China, payment methods, or localization. Since the product depends on external AI tools such as Claude, Cursor, and Codex, actual usability may be affected by network access, account availability, and payment conditions. Alternative options include v0.dev, Lovable, Bolt.new, Cursor’s built-in capabilities, or using Claude Code directly.
⚠ This review is compiled from public sources and does not constitute a purchase recommendation. Verify all facts on the vendor's official site. Verify on ui.sh official site.
ui.sh is an United States Dev Tools provider. TG4G tracks its product information, an overall rating of 8.0/10, and a China-accessibility score of China direct-connect friendly. Click "Visit Official Site" to reach ui.sh directly.