Rime

AI agents

Every page is rendered from one typed registry, and the same content is served as markdown, so an agent reads exactly what you see.

What an agent reads

  • /llms.txt: the index, the rules, the shadcn name map and install lines. Start here.
  • /llms-full.txt: every component, pattern and rule in one file.
  • The markdown twin of any page: add .md (/components/button.md), or use the Copy page menu at the top of every component page, which also opens the page in Claude.
  • https://rime.mjsons.net/r/registry.json: the registry, for the shadcn CLI and MCP.

In your project

The agents item adds docs/ui (llms.txt and llms-full.txt), a Rime skill for Claude Code (.claude/skills/rime) and AGENTS.rime.md. Add one line to your AGENTS.md: "UI: follow AGENTS.rime.md."

npx shadcn@latest add KezLahd/rime/agents

The MCP server

With the @rime namespace in components.json (see Installation), the shadcn MCP server lists, searches and installs Rime items and returns their examples:

{ "mcpServers": { "shadcn": { "command": "npx", "args": ["shadcn@latest", "mcp"] } } }

A good prompt

Read docs/ui/llms.txt (or https://rime.mjsons.net/llms.txt) first and follow its rules.
Build only with Rime components and tokens. If a component is missing from
components/ui, install it with the shadcn CLI rather than writing one.