// the find
Jpisnice/shadcn-ui-mcp-server
A mcp server to allow LLMS gain context about shadcn ui component structure,usage and installation,compaitable with react,svelte 5,vue & React Native
An MCP server that gives AI assistants direct access to shadcn/ui component source, demos, and metadata across React, Svelte, Vue, and React Native. The idea is that instead of an LLM hallucinating component APIs or copying outdated snippets, it fetches live source from the official repos via GitHub API. Squarely aimed at developers using AI-assisted coding tools who spend time correcting bad shadcn usage.
Covers four separate shadcn ports (React, Svelte, Vue, React Native) with framework-specific GitHub API targets rather than one generic blob, so the fetched code is actually from the right codebase. The caching layer with circuit breaker logic shows someone thought about GitHub rate limits rather than just hammering the API on every request. SSE transport support means you can run it as a shared server for a team rather than one process per developer. The `.mcpb` packaging for Claude Desktop one-click installs is a real usability win — most MCP servers make you hand-edit JSON configs.
The server proxies GitHub API calls at runtime, so it is entirely dependent on GitHub uptime and your token quota — if GitHub is slow, every AI request stalls waiting for component source. There are four separate axios utility files (axios.ts, axios-svelte.ts, axios-vue.ts, axios-react-native.ts) which is a code smell suggesting the framework switching is copy-paste rather than a clean abstraction. The tweakcn theme integration feels bolted on — it bundles 41 theme presets as a static JSON file alongside dynamic fetching tools, which will silently drift out of sync as tweakcn ships new themes. No test suite visible in the tree beyond test-package scripts, so breakage from upstream shadcn repo restructuring won't be caught automatically.