shadcn/ui

The comparison rates four of them and the research adds typed props; each link below was opened on November 8, 2026.

• Design tokens: Partly. The theme (opens in a new tab) is CSS variables such as --background and --primary. There is no file of design tokens in the W3C format and no tokens package, and the color JSON the site serves is not documented as one.
• Agent access: Yes. The official shadcn MCP server (opens in a new tab) runs with npx shadcn@latest mcp, and browses, searches and installs components from registries, in Claude Code, Cursor, VS Code with Copilot and Codex.
• Docs agents can read: Yes. There is an llms.txt (opens in a new tab), and every docs page opens as Markdown when you add .md. There is no llms-full.txt.
• Guidance for agents: Yes. The Skills page (opens in a new tab) installs a shadcn skill with npx skills add shadcn/ui; it reads your components.json and runs shadcn info --json, so the agent knows the project before it edits.
• Typed props: Yes. The components are TypeScript files, and each docs page, such as Button’s (opens in a new tab), has an API reference of its props.

shadcn/ui pairs strong agent tooling with design tokens kept in CSS, the same pattern Chakra UI and Shopify Polaris show in the table. The research could not settle one thing: whether the color JSON under /r/colors/ is meant as a public feed. No page says so, so the rating stays Partly until one does. The docs are moving to a split by base library, and some addresses already redirect, so a link that worked last month may land somewhere new. The code is under MIT.

Teams that want to own their component code outright and let agents install it, read it and edit it in place. Registry owners also get the MCP server for their own registry, so a team can serve its own components the same way.

• Who owns each copied component after the agent edits it, and how updates reach code you have changed.
• Where your design tokens will live. With only CSS variables, a design token file for agents is yours to write.
• That the skill’s rules match your conventions, since it reads your project but follows shadcn/ui’s defaults.

With shadcn/ui you already have the code; what it leaves to you is the design tokens and the written rules that keep an agent on your brand. That gap is what an AI-ready design system closes, and what Simbolik builds (opens in a new tab) when a team wants it done for them.