thmh
MCP server providing component catalog tools (search_components, get_component_detail) for coding agents to discover and reuse UI components.
README
Tehon-Mihon (thmh)
Story-less UI component catalog for the AI era — auto-generated from your code, served by a Vite plugin, exposed to agents via MCP
Overview
thmh reads your components directly: static analysis of their types and variant definitions generates every combination of their variants, so you never write a Story file.
It runs inside your app's own Vite dev server — one line in vite.config.ts — and inherits the aliases, Tailwind setup, and environment variables you already have, so there is no second build config to keep in sync.
Everything it knows is written to catalog.json, the single source of truth from which the human-facing UI, the agent-facing MCP server, and CI checks are all derived.
That MCP server exposes tools such as search_components and get_component_detail, letting coding agents find and reuse the components you already have instead of inventing near-duplicates.
The catalog UI is served at /__thmh/; repositories with no host app, such as standalone design systems, can use the thmh CLI instead.
The name
Tehon-Mihon is two Japanese words that both end in 本 (hon) — a book, and the original a thing is copied from.
手本 (てほん, [tehoɴ]) is a model to copy: the sheet of calligraphy a student sets beside their own paper and reproduces stroke for stroke. This is what an agent needs — the exact shape of a component, its props and its variants, ready to be followed.
見本 (みほん, [mihoɴ]) is a sample to look at: the swatch, the specimen, the one you examine before you choose. This is what a person needs — the component rendered, every variant on screen.
From one manifest, two books are born at once.
Getting Started (Prototype)
This section walks through the current prototype. The thmh CLI only implements build; dev, init, and mcp print a "not implemented" message for now.
Prerequisites: Node.js 24+ and pnpm 11 (see packageManager in package.json).
git clone <this repo> && cd thmh
pnpm install
pnpm build
pnpm dev:example
dev:example builds every package, then starts the Vite dev server for examples/react-app. Open:
http://localhost:5173/— the example app itselfhttp://localhost:5173/__thmh/— the catalog, with every variant of the exampleButtonrenderedhttp://localhost:5173/__thmh/api/catalog.json— the same data as a manifest
Edit examples/react-app/src/components/ui/button.tsx and save — the catalog re-analyzes and refreshes within a couple of seconds, no restart needed.
To generate a manifest without a dev server, run the CLI against a built app:
node packages/thmh/dist/cli.js build --root examples/react-app --out catalog.json
To run the test suite:
pnpm test
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。