appCN
MCP server for appCN - discover, search, install, and correctly use copy-paste React Native / Expo UI components (including an AI-native collection) from the live registry
README
<div align="center">
appCN
Copy-paste mobile components for React Native + Expo.
A shadcn-philosophy library: you own the source, NativeWind styles the surface, Reanimated drives the motion — and the featured AI-native collection ships streaming bubbles, prompt composers, reasoning traces, and a reactive 3D voice orb.
Browse components · CLI docs · Mobile app · Live showcase · GitHub
</div>
Quick start
The CLI handles NativeWind + Reanimated wiring and registers the @app-cn
shadcn namespace for you. One command to set up, one to add a component.
# 1. (Recommended) One-shot project setup
npx @app-cn/cli@latest init
# 2. Add any component by slug
npx @app-cn/cli@latest add voice-sphere
Prefer your own pipeline? appcn add is a thin wrapper around shadcn — you can
skip the CLI entirely:
# Copy a single component via shadcn (no setup, one-off)
npx shadcn@latest add https://appcn.vercel.app/r/voice-sphere.json
# Or pull the whole library as a managed dep
npm install @app-cn/ui
Building with an AI agent? Add the appCN MCP server so Claude Code, Cursor, or Windsurf can discover and install components for you — and write new code in appCN's motion/haptic style:
# Claude Code
claude mcp add appcn -- npx -y @app-cn/mcp
See MCP docs for Cursor / Windsurf / VS Code config.
Status: v0.1.0 live on npm. Five components shipped — the AI collection is the marketed wedge and is the first place to land new work.
What's in the box
New: appCN is live on real phones — scan a QR with Expo Go and the entire component gallery loads on Android or iPhone in seconds. No emulator, no developer account. Standalone Play Store APK is in review — shipping soon.
Base
| Component | Delight |
|---|---|
Button |
Spring-y press scale to 0.96, settles in 140 ms — feels weighted, never twitchy |
AI-native
| Component | Delight |
|---|---|
StreamBubble |
Three-phase thinking → token stream → settle |
PromptInput |
Send morphs into a stop with a spinning ring |
ReasoningTrace |
Auto-collapses the instant the answer lands |
VoiceSphere |
Reactive 3D orb that breathes with voice level |
Every component has an interactive web preview, a QR you can scan with Expo Go to run it live on your phone, and a video — same code path on every surface. → appcn.vercel.app/components
Why appCN
- Real-phone previews. Scan a QR with Expo Go and the component runs live on your device with native gestures. No emulator gymnastics.
- AI-native flagship. The AI collection is the wedge — components built for AI apps that nobody else ships well.
- Quadruple distribution. The
appcnCLI, the shadcn registry,@app-cn/uifrom npm — or the MCP server, so your AI coding agent installs components for you. Same primitives, your call. - Motion-first defaults. Reanimated + gesture-handler, shared motion
tokens (
duration/easing/spring), one delight detail per component. - Dark-mode first. Designed dark, then derived to light. Premium reads as depth, not rainbows.
- You own the source. Copy-paste philosophy means no lock-in — once a component lands in your repo, edit freely.
Documentation
| Surface | What's there |
|---|---|
| appcn.vercel.app | Live docs site — landing, every component, live previews |
| CLI docs | init, add, env vars, the idempotency contract |
| Mobile app | Play Store + Expo Go install paths, deep-link contract |
CLAUDE.md |
Architecture, monorepo map, commands, Metro/Reanimated gotchas |
DESIGN.md |
Taste layer — motion rules, the "one delight detail" rule, the 8-step SOP |
CONTRIBUTING.md |
Dev setup, the Changesets release flow, PR conventions |
AGENTS.md |
Cross-tool guide for AI coding agents (Claude / Cursor / Windsurf / Codex / Aider) |
Contributing
Every new component ships through the 8-step SOP in DESIGN.md → Component checklist. Don't skip steps — the PR template enforces them and a missing changeset blocks the release pipeline.
Working with an AI agent? Invoke the /new-component Claude skill — it
walks the checklist, drops a typed <slug>.meta.ts template ready to fill,
wires the registry + showcase entries, and verifies with typecheck + registry
build.
Full guide: CONTRIBUTING.md.
Releases
Releases are driven by Changesets and published to npm via OIDC Trusted Publishing — zero long-lived secrets.
pnpm changeset # describe what changed + pick a bump
git commit -am "..." # commit the changeset alongside your code
# Open a PR. After merge to main, a "Version Packages" PR opens automatically.
# Merge that PR → CI publishes via OIDC.
Every published version gets a verified provenance badge on npm.
Deploying
<details> <summary>The site is two Vercel projects from one monorepo — click for setup.</summary>
| Project | Vercel root directory | What it serves |
|---|---|---|
| Docs | apps/web |
Landing, /components/*, /docs/*, the shadcn registry at /r/*.json |
| Showcase | apps/showcase |
Expo web export — the iframe target for interactive previews |
Deploy order (matters):
- Deploy the showcase first (
apps/showcase) so you have a URL. - Deploy the docs, setting
NEXT_PUBLIC_SHOWCASE_WEB_URLto that URL. - Optional: custom domains (
appcn.vercel.app→ docs,appcn-showcase.vercel.app→ showcase).
Required env vars (docs project):
| Variable | Value |
|---|---|
NEXT_PUBLIC_SHOWCASE_WEB_URL |
https://<showcase-project>.vercel.app |
NEXT_PUBLIC_EXPO_URL |
(empty for v1 — needs EAS Update for the QR tab) |
NEXT_PUBLIC_REGISTRY_URL |
https://<docs-domain>/r |
The showcase project needs no env vars. Defaults live in
apps/web/.env.example.
</details>
Acknowledgments
Inspired by shadcn/ui's copy-paste philosophy and react-native-reusables's docs-site polish. Built on top of NativeWind, Reanimated, and Expo.
License
MIT — you own the code.
<div align="center"> <sub>Made with care by <a href="https://github.com/Salah-XD">Salah-XD</a>.</sub> <br /> <sub>Powered by <a href="https://www.shineup.digital"><b>Shineup</b></a> ↗</sub> </div>
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。