@lando-labs/lando-ds-mcp
MCP server for the Lando Labs Design System, enabling AI agents to introspect and generate code for React components, hooks, icons, design tokens, and theme presets via 15 tools.
README
@lando-labs/lando-ds-mcp
MCP server for the Lando Labs Design System. Gives AI agents authoritative introspection and code-generation for 127 React components, 19 utility hooks, 70 brand icons, ~1,500 Lucide icons, six token categories, and multiple theme presets — all sourced from the DS's own emitted metadata (@lando-labs/lando-ds-meta), always current with each DS release.
Built on the Model Context Protocol. Works with Claude Code, Claude Desktop, Cursor, Windsurf, Zed, and any MCP-compatible client.
What it does
Exposes 15 tools to any MCP-connected AI client:
| Tool | Purpose |
|---|---|
list_components |
Enumerate DS components, filter by category / capability / deprecation |
get_component |
Generate ready-to-use JSX for a component with props, children, and correct imports |
get_component_props |
TypeScript-style props interface for any component |
get_component_reference |
Full reference: description + examples + composes + props in one call |
get_component_capabilities |
Capability matrix — RSC-safe, polymorphic, forwardRef, deprecated |
get_composition_hints |
Which components officially compose with a given one (from meta) |
compose_components |
Merge multiple components into one snippet with unified imports |
get_design_tokens |
Design tokens (colors, spacing, typography, radius, shadows, motion) in CSS, hex, or both |
get_animation_examples |
Curated animation patterns with CSS keyframes and reduced-motion handling |
get_theme_presets |
Available theme presets (default brand-neutral; opt-in lando, forest, midnight, rose, slate, sunset) |
list_icons |
Browse icons: ds-brand (~70), lucide (full ~1,500-icon catalog; ~80 with rich metadata, the rest by name), or all |
search_icons |
Semantic search across icon catalogs with confidence-ranked results |
list_hooks |
Enumerate DS utility hooks (state, dom, browser, a11y, …), filter by category / RSC safety |
get_hook |
Full detail for a single hook — signature, returns, RSC safety, subpath, description |
get_ds_metadata |
DS package version, meta schema version, component/token/icon/hook summary |
Every tool reads from @lando-labs/lando-ds-meta — the DS's self-describing metadata, shipped as a zero-dependency JSON npm package. When the DS ships new components, deprecates old ones, or renames tokens, the MCP surfaces the change on the next content refresh with zero code changes.
Install + connect
You need Node.js 18 or later.
Claude Code
npm install -g @lando-labs/lando-ds-mcp
claude mcp add lando-ds lando-ds-mcp
Or via ~/.claude.json:
{
"mcpServers": {
"lando-ds": {
"command": "lando-ds-mcp"
}
}
}
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on Windows/Linux:
{
"mcpServers": {
"lando-ds": {
"command": "npx",
"args": ["-y", "@lando-labs/lando-ds-mcp"]
}
}
}
Restart Claude Desktop after saving.
Cursor
Edit ~/.cursor/mcp.json:
{
"mcpServers": {
"lando-ds": {
"command": "npx",
"args": ["-y", "@lando-labs/lando-ds-mcp"]
}
}
}
Windsurf
Edit ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"lando-ds": {
"command": "npx",
"args": ["-y", "@lando-labs/lando-ds-mcp"]
}
}
}
Zed
Zed uses context servers configured via ~/.config/zed/settings.json:
{
"context_servers": {
"lando-ds": {
"command": {
"path": "npx",
"args": ["-y", "@lando-labs/lando-ds-mcp"]
}
}
}
}
Generic MCP client
The server speaks MCP over stdio. Invoke lando-ds-mcp (after npm install -g) or npx @lando-labs/lando-ds-mcp.
Verify
After connecting, ask your agent to call get_ds_metadata. It should reply with the DS version, component count, and tool summary. If it does, you're wired up.
What it does not do
- Not a UI library. The MCP tells agents about DS components — the actual components live in
@lando-labs/lando-dsand must be installed in your project separately. - Not a code generator standalone. Runs as an MCP server; needs an MCP client to invoke it.
- Not a project scanner. Every tool reads from DS meta; nothing scans your project source. A future release may add DS-adoption auditing tools; not yet.
- Not Vue/Svelte/Angular compatible. The DS is React; the MCP's code-generation output is JSX.
- Not React Native compatible. The DS is web-only as of v0.57.0.
How it works
The DS publishes @lando-labs/lando-ds-meta — a zero-dependency npm package with meta.json (light) and meta.verbose.json (with descriptions, examples, and composition hints). The MCP takes a semver dep on it and exposes the data through the 15 tools above. npm update @lando-labs/lando-ds-meta is the whole content-refresh workflow.
Because meta is the source of truth, the MCP is a thin protocol adapter over emitted metadata — no synthesis of information the DS didn't emit.
Contributing
Contributions welcome — file an issue or open a PR. Some context up front:
- Development setup:
git clone, thennpm install && npm run build && npm test. Nothing else — DS meta arrives via npm. - Meta compatibility: MCP consumes DS meta schema
^1.x. Schema major bumps require an MCP major bump. - Testing:
npm testuses a custom runner insrc/test.ts. - Version discipline: version strings live in
package.json,src/index.ts, andCHANGELOG.md. The version-sync tests fail CI if these drift.
Issues: github.com/lando-labs/lando-ds-mcp/issues Security disclosures: see MAINTAINERS.md. Maintainers: MAINTAINERS.md.
License
Apache License 2.0. See LICENSE for full terms and NOTICE for attribution.
Copyright 2026 Lando Labs.
Related
- @lando-labs/lando-ds — The design system itself. React web components; the source of truth the MCP reflects.
- @lando-labs/lando-ds-meta — Zero-dependency JSON metadata subpackage the MCP consumes at runtime. Updating the MCP's DS content is
npm update @lando-labs/lando-ds-meta.
First published to public npm and public GitHub as @lando-labs/lando-ds-mcp@4.0.0 on 2026-07-16, alongside @lando-labs/lando-ds@0.57.0 and @lando-labs/lando-ds-meta@0.57.0.
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。