Saasflare UI

AI & MCP

Wire your editor into the design system — an MCP server, llms.txt, and a machine-readable registry mean generated UI is correct, typed, and on-brand from the first prompt.

Most component libraries make your AI guess. @saasflare/ui hands it the source of truth, three ways: a live MCP server, a static llms.txt catalog, and a machine-readable registry.

MCP server

A Model Context Protocol server is served at /api/mcp (stateless, streamable HTTP). Point any MCP client at it and your assistant can query the catalog directly while it writes code.

Connect Claude / Cursor

mcp.json
{
  "mcpServers": {
    "saasflare-ui": {
      "url": "https://ui.saasflare.io/api/mcp"
    }
  }
}

For local development against your own instance, swap the URL for http://localhost:3001/api/mcp.

Tools it exposes

ToolWhat it does
list_componentsEnumerate all 117 components with titles + descriptions
get_componentFull props, variants, and usage examples for one component
search_componentsFuzzy search across the catalog
recommend_for_use_caseSuggest components for a described UI need
get_install_commandThe exact shadcn add command for a component
list_palettes / get_paletteThe 24 palettes and their tokens
list_hooksThe exported hooks (useDisclosure, useMeasure, …)

llms.txt

A complete, machine-readable catalog lives at two static endpoints — drop the whole design system into any model's context in one fetch:

Machine-readable registry

Every component is published as a shadcn-compatible registry item, so an agent can both read the spec and install the source:

npx shadcn@latest add "https://ui.saasflare.io/r/button.json"

The index is at /registry.json; each item at /r/{name}.json.

IntelliSense-grade types

Every export ships strict TypeScript with JSDoc, including runnable @example blocks. That means correct completions and hover docs in your editor — and a copilot that imports from @saasflare/ui with the right props the first time.

Next steps

On this page