195 lines
7.0 KiB
Markdown
195 lines
7.0 KiB
Markdown
# MCP UI Kit
|
|
|
|
|
|
Build interactive React UIs for MCP tools.
|
|
|
|
The [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) enables AI assistants to interact with external tools and data. The [mcp-ui](https://github.com/anthropics/mcp-ui) project extends this with rich, interactive UI components that can be rendered directly in the AI chat.
|
|
|
|
**mcp-ui-kit** simplifies building these UI components by:
|
|
|
|
- **Bundling on-demand** — Write React components, they're bundled automatically when the tool is called
|
|
- **Zero config** — No webpack/vite setup needed, just `createUI()` and point to your `.tsx` file
|
|
- **Props from server** — Pass data from your MCP tool directly to React via `useProps()`
|
|
- **Two-way communication** — Components can `sendPrompt()` to the AI or `callTool()` to invoke other MCP tools
|
|
|
|
Built on top of [@mcp-ui/server](https://www.npmjs.com/package/@mcp-ui/server).
|
|
|
|
## Installation
|
|
|
|
```bash
|
|
npm install mcp-ui-kit @modelcontextprotocol/sdk
|
|
```
|
|
|
|
## Server Usage
|
|
|
|
Create UI components that bundle on-demand within your MCP server:
|
|
|
|
```typescript
|
|
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
|
|
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
|
|
import { createUI } from 'mcp-ui-kit/server'; // 👈 mcp-ui-kit
|
|
|
|
// Create the MCP server
|
|
const server = new McpServer({
|
|
name: 'my-server',
|
|
version: '1.0.0'
|
|
});
|
|
|
|
// Create a UI component // 👈 mcp-ui-kit
|
|
const dashboardUI = createUI( // 👈 mcp-ui-kit
|
|
'my-dashboard', // 👈 mcp-ui-kit
|
|
import.meta.resolve('./MyComponent.tsx') // 👈 mcp-ui-kit
|
|
); // 👈 mcp-ui-kit
|
|
|
|
// Register a tool with the UI
|
|
server.registerTool(
|
|
'dashboard',
|
|
{
|
|
description: 'Interactive dashboard',
|
|
_meta: dashboardUI.meta // 👈 mcp-ui-kit
|
|
},
|
|
async () => ({
|
|
content: [
|
|
{ type: 'text', text: 'Dashboard loaded' },
|
|
await dashboardUI.component({ // 👈 mcp-ui-kit
|
|
props: { title: 'Hello' }, // 👈 mcp-ui-kit
|
|
frameSize: ['700px', '500px'] // 👈 mcp-ui-kit
|
|
}) // 👈 mcp-ui-kit
|
|
]
|
|
})
|
|
);
|
|
|
|
// Start the server
|
|
const transport = new StdioServerTransport();
|
|
await server.connect(transport);
|
|
```
|
|
|
|
### With Input Schema
|
|
|
|
Tools can accept parameters via `inputSchema` and pass them to your UI component:
|
|
|
|
```typescript
|
|
import { z } from 'zod';
|
|
import { createUI } from 'mcp-ui-kit/server'; // 👈 mcp-ui-kit
|
|
|
|
const stockUI = createUI('stocks', import.meta.resolve('./StockDashboard.tsx')); // 👈 mcp-ui-kit
|
|
|
|
server.registerTool(
|
|
'stock_portfolio',
|
|
{
|
|
description: 'View stock portfolio with charts',
|
|
_meta: stockUI.meta, // 👈 mcp-ui-kit
|
|
inputSchema: {
|
|
symbols: z.array(z.string()).default(['AAPL', 'GOOGL']),
|
|
timeframe: z.enum(['1D', '1W', '1M', '1Y']).default('1M'),
|
|
},
|
|
},
|
|
async ({ symbols, timeframe }) => ({
|
|
content: [
|
|
{ type: 'text', text: `Showing ${symbols.join(', ')} for ${timeframe}` },
|
|
await stockUI.component({ // 👈 mcp-ui-kit
|
|
props: { symbols, timeframe }, // 👈 mcp-ui-kit (params → props)
|
|
})
|
|
]
|
|
})
|
|
);
|
|
```
|
|
|
|
## Client Usage
|
|
|
|
Helper functions for your React components (the ones you pass to `createUI`):
|
|
|
|
```tsx
|
|
// StockDashboard.tsx
|
|
import { useState } from 'react';
|
|
import { sendPrompt, callTool, useProps } from 'mcp-ui-kit/ui'; // 👈 mcp-ui-kit
|
|
|
|
function StockDashboard() {
|
|
const [analysis, setAnalysis] = useState<string | null>(null);
|
|
|
|
// Get props passed from server (matches inputSchema params) // 👈 mcp-ui-kit
|
|
const { symbols, timeframe } = useProps({ // 👈 mcp-ui-kit
|
|
symbols: ['AAPL'], // 👈 mcp-ui-kit
|
|
timeframe: '1M' // 👈 mcp-ui-kit
|
|
}); // 👈 mcp-ui-kit
|
|
|
|
const handleAnalyze = (symbol: string) => {
|
|
// Send a message to the AI chat // 👈 mcp-ui-kit
|
|
sendPrompt(`Analyze ${symbol} stock performance over ${timeframe}`); // 👈 mcp-ui-kit
|
|
};
|
|
|
|
const handleRefresh = async (symbol: string) => {
|
|
// Call another MCP tool // 👈 mcp-ui-kit
|
|
callTool('get_stock_price', { symbol }); // 👈 mcp-ui-kit
|
|
};
|
|
|
|
return (
|
|
<div>
|
|
<h1>Stock Portfolio ({timeframe})</h1>
|
|
{symbols.map(symbol => (
|
|
<div key={symbol}>
|
|
<span>{symbol}</span>
|
|
<button onClick={() => handleAnalyze(symbol)}>
|
|
Ask AI to Analyze
|
|
</button>
|
|
<button onClick={() => handleRefresh(symbol)}>
|
|
Refresh Price
|
|
</button>
|
|
</div>
|
|
))}
|
|
</div>
|
|
);
|
|
}
|
|
```
|
|
|
|
## MCP UI Inspector
|
|
|
|
The MCP UI Inspector is a developer tool for testing MCP servers with UI capabilities. It provides a visual interface to:
|
|
|
|
- **Connect to MCP servers** — Enter your server URL and establish a connection
|
|
- **Browse available tools** — See all registered tools and their descriptions
|
|
- **Configure parameters** — Fill in tool input schemas with a form UI or raw JSON
|
|
- **Execute tools** — Run tools and see both text and UI responses
|
|
- **Preview UI components** — View your React components rendered in real-time with full interactivity
|
|
|
|

|
|
|
|
The inspector makes it easy to iterate on your UI components without needing a full AI chat client. You can test `sendPrompt()` and `callTool()` interactions, verify props are passed correctly, and debug your components in isolation.
|
|
|
|
## API
|
|
|
|
### Server (`mcp-ui-kit/server`)
|
|
|
|
**`createUI(name, entryUrl)`** - Creates a UI component
|
|
- `name`: Component identifier (used in the `ui://` URI)
|
|
- `entryUrl`: Path to the component entry file
|
|
- Returns: `{ meta, component(opts?) }`
|
|
- `meta`: Object to spread into tool's `_meta` for UI resource linking
|
|
- `component(opts?)`: Async function returning the UI resource
|
|
- `opts.props`: Data to pass to your React component
|
|
- `opts.frameSize`: `[width, height]` e.g. `['700px', '500px']`
|
|
|
|
The `entryUrl` parameter accepts both formats:
|
|
|
|
```typescript
|
|
// ESM (recommended) - using import.meta.resolve()
|
|
createUI('dashboard', import.meta.resolve('./MyComponent.tsx'));
|
|
|
|
// CommonJS - using require.resolve() or absolute paths
|
|
createUI('dashboard', require.resolve('./MyComponent.tsx'));
|
|
createUI('dashboard', path.join(__dirname, './MyComponent.tsx'));
|
|
```
|
|
|
|
### UI (`mcp-ui-kit/ui`)
|
|
|
|
- **`useProps(defaults)`** - Get props passed from the server via `component({ props })`
|
|
- **`sendPrompt(message)`** - Send a message to the AI chat
|
|
- **`callTool(name, params)`** - Invoke an MCP tool
|
|
|
|
## Development
|
|
|
|
```bash
|
|
npm install
|
|
npm run dev:all # Start demo server + inspector
|
|
```
|