feat(mcp): MCP connector

2026-03-18 03:12:20 +08:00
parent 3063b2a06d
commit cef39d5090
8 changed files with 1479 additions and 17 deletions
--- a/packages/extension/src/entrypoints/background.ts
+++ b/packages/extension/src/entrypoints/background.ts
@@ -30,7 +30,34 @@ export default defineBackground(() => {
 		}
 	})

+	// external messages (from localhost launcher page via externally_connectable)
+
+	chrome.runtime.onMessageExternal.addListener((message, sender, sendResponse) => {
+		if (message.type === 'OPEN_HUB') {
+			openOrFocusHubTab(message.wsPort).then(() => {
+				if (sender.tab?.id) chrome.tabs.remove(sender.tab.id)
+				sendResponse({ ok: true })
+			})
+			return true
+		}
+	})
+
 	// setup

 	chrome.sidePanel.setPanelBehavior({ openPanelOnActionClick: true }).catch(() => {})
 })
+
+async function openOrFocusHubTab(wsPort: number) {
+	const hubUrl = chrome.runtime.getURL('hub.html')
+	const existing = await chrome.tabs.query({ url: `${hubUrl}*` })
+
+	if (existing.length > 0 && existing[0].id) {
+		await chrome.tabs.update(existing[0].id, {
+			active: true,
+			url: `${hubUrl}?ws=${wsPort}`,
+		})
+		return
+	}
+
+	await chrome.tabs.create({ url: `${hubUrl}?ws=${wsPort}`, pinned: true })
+}
--- a/packages/mcp/README.md
+++ b/packages/mcp/README.md
@@ -0,0 +1,86 @@
+# @page-agent/mcp
+
+MCP server that lets AI agent clients (Claude Desktop, Copilot, etc.) control your browser through the [Page Agent](https://github.com/alibaba/page-agent) extension.
+
+## Prerequisites
+
+- Node.js >= 20
+- [Page Agent Extension](https://chromewebstore.google.com/detail/page-agent-ext/akldabonmimlicnjlflnapfeklbfemhj) installed in Chrome
+- An LLM API key (OpenAI-compatible)
+
+## Installation
+
+### Claude Desktop
+
+Add to `~/Library/Application Support/Claude/claude_desktop_config.json`:
+
+```json
+{
+    "mcpServers": {
+        "page-agent": {
+            "command": "npx",
+            "args": ["-y", "@page-agent/mcp"],
+            "env": {
+                "LLM_BASE_URL": "https://dashscope.aliyuncs.com/compatible-mode/v1",
+                "LLM_API_KEY": "sk-xxx",
+                "LLM_MODEL_NAME": "qwen3.5-plus"
+            }
+        }
+    }
+}
+```
+
+### Cursor / Copilot
+
+Same format — add the config to the MCP settings of your client.
+
+## MCP Tools
+
+| Tool           | Input              | Description                                          |
+| -------------- | ------------------ | ---------------------------------------------------- |
+| `execute_task` | `{ task: string }` | Execute a browser task in natural language. Blocking. |
+| `get_status`   | —                  | Returns `{ connected, busy }`                        |
+| `stop_task`    | —                  | Stop the currently running task.                     |
+
+## Environment Variables
+
+| Variable         | Default | Description           |
+| ---------------- | ------- | --------------------- |
+| `LLM_BASE_URL`   | —       | LLM API base URL      |
+| `LLM_API_KEY`    | —       | LLM API key           |
+| `LLM_MODEL_NAME` | —       | Model name            |
+| `PORT`           | `38401` | HTTP + WebSocket port |
+
+## How It Works
+
+```
+┌──────────────┐  stdio   ┌──────────────────┐  WebSocket   ┌──────────────┐
+│ Claude /     │◄────────►│ @page-agent/mcp  │◄────────────►│ Hub tab      │
+│ Copilot      │  (MCP)   │ (Node.js)        │  (localhost) │ (extension)  │
+└──────────────┘          └──────────────────┘              └──────┬───────┘
+                                   │                               │
+                                   │ HTTP                          │ useAgent
+                                   ▼                               ▼
+                          ┌──────────────────┐              ┌──────────────┐
+                          │ Launcher page    │              │ MultiPage    │
+                          │ (localhost:PORT) │              │ Agent        │
+                          └──────────────────┘              └──────────────┘
+```
+
+1. Agent client starts the MCP server via stdio (`npx @page-agent/mcp`).
+2. Server starts HTTP + WS on `localhost:PORT`, opens the launcher page in browser.
+3. Launcher page triggers the extension to open a **hub tab** (`hub.html?ws=PORT`).
+4. Hub connects to the WS server. MCP tools now proxy tasks to the hub.
+
+The hub tab speaks a generic WebSocket protocol (defined in `hub-ws.ts` in the extension package) and has no knowledge of MCP. See the hub's protocol docs for message format details.
+
+## Architecture
+
+Pure JS ESM, no build step. Source files are the published artifacts.
+
+```
+src/
+├── index.js        # CLI entry: MCP server (stdio) + opens launcher
+├── hub-bridge.js   # HTTP server + WebSocket bridge to hub tab
+└── launcher.html   # Bootstrap page: detects extension, triggers hub open
+```
--- a/packages/mcp/package.json
+++ b/packages/mcp/package.json
@@ -0,0 +1,34 @@
+{
+    "name": "@page-agent/mcp",
+    "version": "1.5.8",
+    "type": "module",
+    "bin": {
+        "page-agent-mcp": "src/index.js"
+    },
+    "files": [
+        "src/"
+    ],
+    "description": "MCP server for controlling the browser via Page Agent extension",
+    "keywords": [
+        "page-agent",
+        "mcp",
+        "browser-automation",
+        "chrome-extension"
+    ],
+    "author": "Simon<gaomeng1900>",
+    "license": "MIT",
+    "repository": {
+        "type": "git",
+        "url": "https://github.com/alibaba/page-agent.git",
+        "directory": "packages/mcp"
+    },
+    "homepage": "https://alibaba.github.io/page-agent/",
+    "engines": {
+        "node": ">=20"
+    },
+    "dependencies": {
+        "@modelcontextprotocol/sdk": "^1.27.1",
+        "ws": "^8.19.0",
+        "zod": "^4.3.5"
+    }
+}
--- a/packages/mcp/src/hub-bridge.js
+++ b/packages/mcp/src/hub-bridge.js
@@ -0,0 +1,136 @@
+#!/usr/bin/env node
+import { readFileSync } from 'node:fs'
+import http from 'node:http'
+import { fileURLToPath } from 'node:url'
+import { WebSocketServer } from 'ws'
+
+const EXT_ID = 'akldabonmimlicnjlflnapfeklbfemhj'
+const STORE_URL = `https://chromewebstore.google.com/detail/page-agent-ext/${EXT_ID}`
+
+const launcherTemplate = readFileSync(
+	fileURLToPath(new URL('./launcher.html', import.meta.url)),
+	'utf-8'
+)
+
+/**
+ * HTTP + WebSocket bridge to the hub.html extension tab.
+ * - HTTP serves the launcher page (triggers extension to open hub)
+ * - WS carries execute/stop commands and result/error responses
+ */
+export class HubBridge {
+	/** @type {number} */
+	port
+
+	/** @type {http.Server} */
+	#httpServer
+
+	/** @type {WebSocketServer} */
+	#wss
+
+	/** @type {import('ws').WebSocket | null} */
+	#hub = null
+
+	/** @type {{ resolve: (r: {success: boolean, data: string}) => void, reject: (e: Error) => void } | null} */
+	#pendingTask = null
+
+	/** @param {number} port */
+	constructor(port) {
+		this.port = port
+		this.#httpServer = http.createServer((_req, res) => {
+			const html = launcherTemplate
+				.replaceAll('__EXT_ID__', EXT_ID)
+				.replaceAll('__STORE_URL__', STORE_URL)
+				.replaceAll('__WS_PORT__', String(port))
+			res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' })
+			res.end(html)
+		})
+		this.#wss = new WebSocketServer({ server: this.#httpServer })
+		this.#wss.on('connection', (ws) => this.#onConnection(ws))
+	}
+
+	/** @returns {Promise<void>} */
+	async start() {
+		return new Promise((resolve, reject) => {
+			this.#httpServer.on('error', (/** @type {NodeJS.ErrnoException} */ err) => {
+				if (err.code === 'EADDRINUSE') {
+					reject(
+						new Error(`Port ${this.port} is in use. Another Page Agent MCP server may be running.`)
+					)
+				} else {
+					reject(err)
+				}
+			})
+			this.#httpServer.listen(this.port, () => {
+				console.error(`[page-agent-mcp] HTTP + WS on http://localhost:${this.port}`)
+				resolve()
+			})
+		})
+	}
+
+	get connected() {
+		return this.#hub?.readyState === 1
+	}
+
+	get busy() {
+		return this.#pendingTask !== null
+	}
+
+	/**
+	 * @param {string} task
+	 * @param {Record<string, unknown>} [config]
+	 * @returns {Promise<{success: boolean, data: string}>}
+	 */
+	async executeTask(task, config) {
+		if (!this.connected) throw new Error('Hub is not connected. Is the extension running?')
+		if (this.#pendingTask) throw new Error('Agent is already running a task.')
+
+		return new Promise((resolve, reject) => {
+			this.#pendingTask = { resolve, reject }
+			this.#hub.send(JSON.stringify({ type: 'execute', task, config }))
+		})
+	}
+
+	stopTask() {
+		if (this.connected) {
+			this.#hub.send(JSON.stringify({ type: 'stop' }))
+		}
+	}
+
+	/** @param {import('ws').WebSocket} ws */
+	#onConnection(ws) {
+		if (this.#hub && this.#hub.readyState === 1) {
+			ws.close(4000, 'Another hub is already connected')
+			return
+		}
+
+		this.#hub = ws
+		console.error('[page-agent-mcp] Hub connected')
+
+		ws.on('message', (/** @type {Buffer} */ rawData) => {
+			/** @type {{ type: string, success?: boolean, data?: string, message?: string }} */
+			let msg
+			try {
+				msg = JSON.parse(rawData.toString('utf-8'))
+			} catch {
+				return
+			}
+
+			if (msg.type === 'result') {
+				this.#pendingTask?.resolve({ success: msg.success ?? false, data: msg.data ?? '' })
+				this.#pendingTask = null
+			} else if (msg.type === 'error') {
+				this.#pendingTask?.reject(new Error(msg.message ?? 'Unknown error from hub'))
+				this.#pendingTask = null
+			}
+		})
+
+		ws.on('close', () => {
+			console.error('[page-agent-mcp] Hub disconnected')
+			if (this.#hub === ws) this.#hub = null
+			if (this.#pendingTask) {
+				this.#pendingTask.reject(new Error('Hub disconnected while task was running'))
+				this.#pendingTask = null
+			}
+		})
+	}
+}
--- a/packages/mcp/src/index.js
+++ b/packages/mcp/src/index.js
@@ -0,0 +1,95 @@
+#!/usr/bin/env node
+import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js'
+import { exec } from 'node:child_process'
+import { platform } from 'node:os'
+import * as z from 'zod/v4'
+
+import { HubBridge } from './hub-bridge.js'
+
+const env = process.env
+const port = parseInt(env.PORT || '38401')
+
+/** @type {Record<string, string>} */
+const llmConfig = {}
+if (env.LLM_API_KEY) llmConfig.apiKey = env.LLM_API_KEY
+if (env.LLM_BASE_URL) llmConfig.baseURL = env.LLM_BASE_URL
+if (env.LLM_MODEL_NAME) llmConfig.model = env.LLM_MODEL_NAME
+
+// --- Hub bridge (HTTP + WebSocket) ---
+
+const hub = new HubBridge(port)
+await hub.start()
+
+// Open launcher in default browser
+const url = `http://localhost:${port}`
+const cmd = platform() === 'darwin' ? 'open' : platform() === 'win32' ? 'start ""' : 'xdg-open'
+exec(`${cmd} "${url}"`, (err) => {
+	if (err) console.error(`[page-agent-mcp] Could not open browser: ${err.message}`)
+})
+
+// --- MCP server (stdio) ---
+
+const mcpServer = new McpServer({ name: 'page-agent', version: '1.5.8' })
+
+mcpServer.registerTool(
+	'execute_task',
+	{
+		description:
+			'Execute a browser automation task described in natural language. ' +
+			'The Page Agent extension will control the browser to complete the task. ' +
+			'Blocks until the task is complete.',
+		inputSchema: { task: z.string().describe('Task description in natural language') },
+	},
+	async ({ task }) => {
+		try {
+			const config = Object.keys(llmConfig).length > 0 ? llmConfig : undefined
+			const result = await hub.executeTask(task, config)
+			return {
+				content: [
+					{
+						type: 'text',
+						text: result.success
+							? `Task completed successfully.\n\n${result.data}`
+							: `Task failed.\n\n${result.data}`,
+					},
+				],
+			}
+		} catch (err) {
+			return {
+				content: [{ type: 'text', text: `Error: ${err.message}` }],
+				isError: true,
+			}
+		}
+	}
+)
+
+mcpServer.registerTool(
+	'get_status',
+	{
+		description: 'Check the current status of the Page Agent hub connection and agent.',
+	},
+	async () => ({
+		content: [
+			{
+				type: 'text',
+				text: JSON.stringify({ connected: hub.connected, busy: hub.busy }, null, 2),
+			},
+		],
+	})
+)
+
+mcpServer.registerTool(
+	'stop_task',
+	{
+		description: 'Stop the currently running browser automation task.',
+	},
+	async () => {
+		hub.stopTask()
+		return { content: [{ type: 'text', text: 'Stop signal sent.' }] }
+	}
+)
+
+const transport = new StdioServerTransport()
+await mcpServer.connect(transport)
+console.error('[page-agent-mcp] MCP server ready (stdio)')
--- a/packages/mcp/src/launcher.html
+++ b/packages/mcp/src/launcher.html
@@ -0,0 +1,129 @@
+<!doctype html>
+<html lang="en">
+	<head>
+		<meta charset="UTF-8" />
+		<meta name="viewport" content="width=device-width, initial-scale=1.0" />
+		<title>Page Agent – Connecting</title>
+		<style>
+			* {
+				margin: 0;
+				padding: 0;
+				box-sizing: border-box;
+			}
+			body {
+				font-family:
+					system-ui,
+					-apple-system,
+					sans-serif;
+				background: #0a0a0a;
+				color: #e5e5e5;
+				display: flex;
+				align-items: center;
+				justify-content: center;
+				min-height: 100vh;
+			}
+			.card {
+				text-align: center;
+				max-width: 420px;
+				padding: 3rem 2rem;
+			}
+			h1 {
+				font-size: 1.25rem;
+				font-weight: 600;
+				margin-bottom: 0.5rem;
+			}
+			p {
+				font-size: 0.875rem;
+				color: #a3a3a3;
+				line-height: 1.6;
+			}
+			.spinner {
+				width: 32px;
+				height: 32px;
+				border: 3px solid #333;
+				border-top-color: #fff;
+				border-radius: 50%;
+				animation: spin 0.8s linear infinite;
+				margin: 0 auto 1.5rem;
+			}
+			@keyframes spin {
+				to {
+					transform: rotate(360deg);
+				}
+			}
+			a {
+				color: #60a5fa;
+				text-decoration: none;
+			}
+			a:hover {
+				text-decoration: underline;
+			}
+			.install {
+				display: none;
+			}
+			.install.show {
+				display: block;
+			}
+			.btn {
+				display: inline-block;
+				margin-top: 1rem;
+				padding: 0.5rem 1.25rem;
+				background: #2563eb;
+				color: #fff;
+				border-radius: 0.5rem;
+				font-size: 0.875rem;
+				font-weight: 500;
+			}
+			.btn:hover {
+				background: #1d4ed8;
+				text-decoration: none;
+			}
+		</style>
+	</head>
+	<body>
+		<div class="card">
+			<div id="connecting">
+				<div class="spinner"></div>
+				<h1>Connecting to Page Agent</h1>
+				<p>Opening the hub in your browser...</p>
+			</div>
+			<div id="install" class="install">
+				<h1>Extension Not Found</h1>
+				<p>The Page Agent browser extension is required.<br />Install it and try again.</p>
+				<a class="btn" href="__STORE_URL__" target="_blank">Install Extension</a>
+				<p style="margin-top: 1.5rem; font-size: 0.75rem">
+					After installing, restart the MCP server or refresh this page.
+				</p>
+			</div>
+		</div>
+		<script>
+			;(function () {
+				var EXT_ID = '__EXT_ID__'
+				var wsPort = __WS_PORT__
+
+				function showInstall() {
+					document.getElementById('connecting').style.display = 'none'
+					document.getElementById('install').classList.add('show')
+				}
+
+				try {
+					if (typeof chrome === 'undefined' || !chrome.runtime || !chrome.runtime.sendMessage) {
+						showInstall()
+						return
+					}
+					chrome.runtime.sendMessage(
+						EXT_ID,
+						{ type: 'OPEN_HUB', wsPort: wsPort },
+						function (response) {
+							if (chrome.runtime.lastError || !response || !response.ok) {
+								showInstall()
+							}
+						}
+					)
+				} catch (e) {
+					showInstall()
+				}
+			})()
+		</script>
+	</body>
+</html>