项目概述：打造智能网页操作助手
在本教程中，我们将构建一个完整的、能够实际交互网页的AI助手。这个助手不仅能理解自然语言指令，还能通过 Playwright MCP 执行复杂的网页操作。我们将从零开始，搭建一个功能完备的系统，涵盖从环境配置到实际部署的全流程。
项目目标
- 构建一个能够执行以下任务的AI助手：
- 自动登录网站并处理认证
- 填写复杂表单和交互元素
- 提取、分析和结构化网页数据
- 处理多步骤工作流程
- 应对网页异常和动态内容
一、项目架构设计
技术栈选择
- 后端框架: Node.js + Express
- 浏览器自动化: Playwright
- AI 模型集成: Anthropic Claude API
- 协议层: 自定义 MCP (Model Context Protocol) Server
- 前端界面: React + Tailwind CSS
- 数据库: SQLite (用于会话存储)
- 任务队列: Bull (用于异步任务处理)
系统架构
用户界面 (React)
↓ (HTTP/REST API)
后端服务器 (Express + AI 路由)
↓ (MCP 协议)
Playwright MCP Server
↓ (浏览器控制)
Chromium/Firefox 实例
二、环境准备与项目初始化
步骤1：创建项目结构
mkdir ai-web-assistant
cd ai-web-assistant
mkdir -p src/{mcp,ai,routes,models,utils} public/{css,js} tests
touch package.json server.js .env.example README.md
步骤2：定义项目依赖
创建 package.json：
{
"name": "ai-web-assistant",
"version": "1.0.0",
"type": "module",
"scripts": {
"start": "node server.js",
"dev": "nodemon server.js",
"test": "jest",
"mcp:dev": "node src/mcp/server.js"
},
"dependencies": {
"express": "^4.18.2",
"cors": "^2.8.5",
"dotenv": "^16.3.0",
"playwright": "^1.40.0",
"@anthropic-ai/sdk": "^0.7.0",
"sqlite3": "^5.1.6",
"bull": "^4.11.0",
"express-rate-limit": "^7.1.0",
"helmet": "^7.0.0"
},
"devDependencies": {
"nodemon": "^3.0.0",
"jest": "^29.6.0"
}
}
运行 npm install 安装依赖。
步骤3：环境配置
创建 .env 文件：
API 配置
ANTHROPIC_API_KEY=your_anthropic_api_key_here
PORT=3000
NODE_ENV=development浏览器配置
BROWSER_TYPE=chromium
HEADLESS_MODE=false
BROWSER_TIMEOUT=30000数据库配置
DB_PATH=./data/sessions.db
安全配置
SESSION_SECRET=your_session_secret_here
RATE_LIMIT_WINDOW=900000
RATE_LIMIT_MAX=100
三、核心模块实现
1. Playwright MCP Server 实现
创建 src/mcp/server.js：
import { chromium, firefox, webkit } from'playwright';
import { EventEmitter } from'events';
class PlaywrightMCPServer extends EventEmitter {
constructor(config = {}) {
super();
this.config = {
browserType: config.browserType || 'chromium',
headless: config.headless !== false,
timeout: config.timeout || 30000,
...config
};
this.browser = null;
this.context = null;
this.page = null;
this.isInitialized = false;
this.sessionId = null;
}
// 初始化浏览器实例
async initialize(sessionId = null) {
try {
this.sessionId = sessionId ||session_${Date.now()};
const browserMap = { chromium, firefox, webkit };
const BrowserClass = browserMap[this.config.browserType] || chromium;
this.browser = await BrowserClass.launch({
headless: this.config.headless,
timeout: this.config.timeout,
args: ['--no-sandbox', '--disable-dev-shm-usage']
});
this.context = awaitthis.browser.newContext({
viewport: { width: 1280, height: 720 },
userAgent: 'AI-Web-Assistant/1.0',
acceptDownloads: true,
ignoreHTTPSErrors: true
});
// 添加页面错误处理
this.context.on('page', page => {
page.on('pageerror', error => {
this.emit('pageError', { sessionId: this.sessionId, error });
});
});
this.page = awaitthis.context.newPage();
// 设置默认超时
this.page.setDefaultTimeout(this.config.timeout);
this.page.setDefaultNavigationTimeout(this.config.timeout * 2);
this.isInitialized = true;
this.emit('initialized', { sessionId: this.sessionId });
return {
success: true,
message: 'Playwright MCP Server initialized successfully',
sessionId: this.sessionId
};
} catch (error) {
console.error('Failed to initialize Playwright:', error);
this.emit('error', error);
return { success: false, error: error.message };
}
}
// 工具定义 - MCP 协议核心
getTools() {
return {
navigate: {
name: 'navigate',
description: 'Navigate to a specific URL',
parameters: {
url: {
type: 'string',
description: 'The URL to navigate to'
},
waitUntil: {
type: 'string',
description: 'When to consider navigation successful',
enum: ['load', 'domcontentloaded', 'networkidle'],
default: 'networkidle'
}
}
},
click: {
name: 'click',
description: 'Click on an element using CSS selector, XPath, or text',
parameters: {
selector: {
type: 'string',
description: 'CSS selector, XPath, or text to identify the element'
},
selectorType: {
type: 'string',
description: 'Type of selector: css, xpath, or text',
enum: ['css', 'xpath', 'text'],
default: 'css'
},
waitForNavigation: {
type: 'boolean',
description: 'Whether to wait for navigation after click',
default: false
}
}
},
fill_form: {
name: 'fill_form',
description: 'Fill a form with multiple fields',
parameters: {
fields: {
type: 'object',
description: 'Object mapping selectors to values'
}
}
},
extract_data: {
name: 'extract_data',
description: 'Extract structured data from the page',
parameters: {
schema: {
type: 'object',
description: 'Schema defining what data to extract'
}
}
},
wait_for_element: {
name: 'wait_for_element',
description: 'Wait for an element to appear',
parameters: {
selector: {
type: 'string',
description: 'CSS selector for the element'
},
state: {
type: 'string',
description: 'Element state to wait for',
enum: ['attached', 'detached', 'visible', 'hidden'],
default: 'visible'
},
timeout: {
type: 'number',
description: 'Timeout in milliseconds',
default: 10000
}
}
},
screenshot: {
name: 'screenshot',
description: 'Take a screenshot for debugging',
parameters: {
fullPage: {
type: 'boolean',
description: 'Whether to capture full page',
default: false
}
}
},
get_page_info: {
name: 'get_page_info',
description: 'Get comprehensive information about the current page'
}
};
}
// 工具执行引擎
async executeTool(toolName, parameters = {}) {
if (!this.isInitialized) {
thrownewError('Playwright not initialized. Call initialize() first.');
}
try {
let result;
switch (toolName) {
case'navigate':
result = awaitthis.navigateToUrl(parameters.url, parameters.waitUntil);
break;
case'click':
result = awaitthis.clickElement(parameters.selector, parameters.selectorType, parameters.waitForNavigation);
break;
case'fill_form':
result = awaitthis.fillForm(parameters.fields);
break;
case'extract_data':
result = awaitthis.extractData(parameters.schema);
break;
case'wait_for_element':
result = awaitthis.waitForElement(parameters.selector, parameters.state, parameters.timeout);
break;
case'screenshot':
result = awaitthis.takeScreenshot(parameters.fullPage);
break;
case'get_page_info':
result = awaitthis.getPageInfo();
break;
default:
thrownewError(Unknown tool: ${toolName});
}
this.emit('toolExecuted', {
sessionId: this.sessionId,
toolName,
parameters,
result
});
return { success: true, data: result };
} catch (error) {
console.error(Tool execution failed: ${toolName}, error);
this.emit('toolError', {
sessionId: this.sessionId,
toolName,
parameters,
error: error.message
});
return {
success: false,
error: error.message,
suggestion: this.getErrorSuggestion(error.message)
};
}
}
// 具体的工具实现方法
async navigateToUrl(url, waitUntil = 'networkidle') {
if (!url.startsWith('http')) {
url = 'https://' + url;
}
const response = awaitthis.page.goto(url, {
waitUntil,
timeout: this.config.timeout
});
return {
url: this.page.url(),
status: response?.status(),
title: awaitthis.page.title(),
finalUrl: this.page.url()
};
}
async clickElement(selector, selectorType = 'css', waitForNavigation = false) {
let element;
switch (selectorType) {
case'css':
element = this.page.locator(selector);
break;
case'xpath':
element = this.page.locator(xpath=${selector});
break;
case'text':
element = this.page.getByText(selector, { exact: false });
break;
default:
thrownewError(Unsupported selector type: ${selectorType});
}
await element.waitFor({ state: 'visible' });
if (waitForNavigation) {
awaitPromise.all([
this.page.waitForNavigation({ waitUntil: 'networkidle' }),
element.click()
]);
} else {
await element.click();
}
return {
success: true,
element: awaitthis.getElementInfo(element)
};
}
async fillForm(fields) {
const results = {};
for (const [selector, value] ofObject.entries(fields)) {
try {
const element = this.page.locator(selector);
await element.waitFor({ state: 'visible' });
await element.fill(value);
results[selector] = { success: true, value };
} catch (error) {
results[selector] = { success: false, error: error.message };
}
}
return results;
}
async extractData(schema) {
const data = {};
for (const [key, config] ofObject.entries(schema)) {
try {
const { selector, type = 'text', attribute } = config;
const element = this.page.locator(selector);
switch (type) {
case'text':
data[key] = await element.textContent();
break;
case'attribute':
data[key] = await element.getAttribute(attribute);
break;
case'multiple':
data[key] = await element.allTextContents();
break;
default:
data[key] = await element.textContent();
}
} catch (error) {
data[key] = null;
}
}
return data;
}
async getElementInfo(element) {
try {
const boundingBox = await element.boundingBox();
const isVisible = await element.isVisible();
return {
visible: isVisible,
boundingBox,
tagName: await element.evaluate(el => el.tagName.toLowerCase())
};
} catch (error) {
return { error: error.message };
}
}
async takeScreenshot(fullPage = false) {
const screenshot = awaitthis.page.screenshot({
fullPage,
type: 'png'
});
return {
screenshot: screenshot.toString('base64'),
type: 'png',
fullPage
};
}
async getPageInfo() {
return {
url: this.page.url(),
title: awaitthis.page.title(),
content: awaitthis.page.content(),
viewport: this.page.viewportSize()
};
}
// 错误处理和建议
getErrorSuggestion(errorMessage) {
const suggestions = {
'timeout': '尝试增加等待时间或检查网络连接',
'element not found': '检查选择器是否正确，或等待元素加载',
'navigation failed': '检查URL是否正确，或网站是否可访问',
'target closed': '浏览器页面已关闭，需要重新初始化'
};
for (const [key, suggestion] ofObject.entries(suggestions)) {
if (errorMessage.toLowerCase().includes(key)) {
return suggestion;
}
}
return'请检查网络连接和页面状态后重试';
}
// 清理资源
async cleanup() {
try {
if (this.page) {
awaitthis.page.close();
}
if (this.context) {
awaitthis.context.close();
}
if (this.browser) {
awaitthis.browser.close();
}
this.isInitialized = false;
this.emit('cleanedUp', { sessionId: this.sessionId });
return { success: true, message: 'Resources cleaned up successfully' };
} catch (error) {
console.error('Cleanup failed:', error);
return { success: false, error: error.message };
}
}
}
exportdefault PlaywrightMCPServer;
2. AI 处理模块
创建 src/ai/handler.js：
import Anthropic from'@anthropic-ai/sdk';
import PlaywrightMCPServer from'../mcp/server.js';
class AIHandler {
constructor(apiKey) {
this.anthropic = new Anthropic({ apiKey });
this.mcpServer = new PlaywrightMCPServer();
this.conversationHistory = newMap();
}
// 初始化会话
async initializeSession(sessionId) {
const result = awaitthis.mcpServer.initialize(sessionId);
if (!this.conversationHistory.has(sessionId)) {
this.conversationHistory.set(sessionId, []);
}
return result;
}
// 处理用户指令
async processInstruction(sessionId, instruction, context = {}) {
try {
const history = this.conversationHistory.get(sessionId) || [];
// 构建系统提示词
const systemPrompt = this.buildSystemPrompt(context);
// 获取可用工具
const availableTools = this.mcpServer.getTools();
// 调用 Claude 模型
const message = awaitthis.anthropic.messages.create({
model: "claude-3-sonnet-20240229",
max_tokens: 4096,
system: systemPrompt,
messages: [
...history,
{ role: "user", content: instruction }
],
tools: Object.values(availableTools)
});
let finalResponse = '';
let currentMessage = message;
// 处理工具调用
while (currentMessage.content.some(item => item.type === 'tool_use')) {
const toolResults = [];
for (const contentItem of currentMessage.content) {
if (contentItem.type === 'tool_use') {
const toolName = contentItem.name;
const parameters = contentItem.input;
// 执行工具
const toolResult = awaitthis.mcpServer.executeTool(toolName, parameters);
toolResults.push({
type: 'tool_result',
tool_use_id: contentItem.id,
content: JSON.stringify(toolResult)
});
}
}
// 继续对话
currentMessage = awaitthis.anthropic.messages.create({
model: "claude-3-sonnet-20240229",
max_tokens: 4096,
messages: [
...history,
{ role: "user", content: instruction },
{ role: "assistant", content: currentMessage.content },
{ role: "user", content: toolResults }
],
tools: Object.values(availableTools)
});
}
// 提取最终响应
const textContent = currentMessage.content.find(item => item.type === 'text');
finalResponse = textContent ? textContent.text : '操作完成';
// 更新对话历史
history.push(
{ role: "user", content: instruction },
{ role: "assistant", content: currentMessage.content }
);
// 保持最近10轮对话
if (history.length > 20) {
history.splice(0, 4);
}
return {
success: true,
response: finalResponse,
sessionId
};
} catch (error) {
console.error('AI processing failed:', error);
return {
success: false,
error: error.message,
sessionId
};
}
}
// 构建系统提示词
buildSystemPrompt(context) {
return`你是一个专业的网页操作助手，可以通过浏览器自动化工具执行各种网页任务。
你的能力包括：
- 导航到指定网址
- 点击按钮和链接
- 填写表单和输入框
- 提取网页数据
- 等待页面加载
- 处理复杂交互
重要指导原则：
- 在执行操作前先分析页面结构
- 使用合适的选择器定位元素
- 处理可能出现的错误和异常
- 提供清晰的操作反馈
- 对于复杂任务，分解为多个步骤执行
当前上下文：${JSON.stringify(context)}
请谨慎操作，确保每一步都正确执行。如果遇到错误，请分析原因并提供解决方案。`;
}
// 获取会话历史
getSessionHistory(sessionId) {
returnthis.conversationHistory.get(sessionId) || [];
}
// 清理会话
async cleanupSession(sessionId) {
this.conversationHistory.delete(sessionId);
returnawaitthis.mcpServer.cleanup();
}
}
exportdefault AIHandler;
3. Express 服务器和路由
创建 server.js：
import express from'express';
import cors from'cors';
import helmet from'helmet';
import rateLimit from'express-rate-limit';
import dotenv from'dotenv';
import AIHandler from'./src/ai/handler.js';
// 加载环境变量
dotenv.config();
const app = express();
const PORT = process.env.PORT || 3000;
// 初始化 AI 处理器
const aiHandler = new AIHandler(process.env.ANTHROPIC_API_KEY);
// 中间件配置
app.use(helmet());
app.use(cors());
app.use(express.json({ limit: '10mb' }));
// 速率限制
const limiter = rateLimit({
windowMs: parseInt(process.env.RATE_LIMIT_WINDOW) || 15 * 60 * 1000,
max: parseInt(process.env.RATE_LIMIT_MAX) || 100,
message: '请求过于频繁，请稍后再试'
});
app.use(limiter);
// 会话存储
const sessions = newMap();
// API 路由
// 健康检查
app.get('/health', (req, res) => {
res.json({ status: 'ok', timestamp: newDate().toISOString() });
});
// 初始化会话
app.post('/api/session/init', async (req, res) => {
try {
const sessionId = req.body.sessionId ||session_${Date.now()}_${Math.random().toString(36).substr(2, 9)};
const result = await aiHandler.initializeSession(sessionId);
if (result.success) {
sessions.set(sessionId, {
createdAt: newDate(),
lastActivity: newDate()
});
res.json({
success: true,
sessionId,
message: '会话初始化成功'
});
} else {
res.status(500).json({
success: false,
error: result.error
});
}
} catch (error) {
console.error('Session init error:', error);
res.status(500).json({
success: false,
error: error.message
});
}
});
// 处理用户指令
app.post('/api/instruction', async (req, res) => {
try {
const { sessionId, instruction, context = {} } = req.body;
if (!sessionId || !instruction) {
return res.status(400).json({
success: false,
error: '缺少必要参数：sessionId 和 instruction'
});
}
// 更新会话活动时间
const session = sessions.get(sessionId);
if (session) {
session.lastActivity = newDate();
}
const result = await aiHandler.processInstruction(sessionId, instruction, context);
res.json(result);
} catch (error) {
console.error('Instruction processing error:', error);
res.status(500).json({
success: false,
error: error.message
});
}
});
// 获取会话历史
app.get('/api/session/:sessionId/history', (req, res) => {
const { sessionId } = req.params;
const history = aiHandler.getSessionHistory(sessionId);
res.json({
success: true,
sessionId,
history
});
});
// 清理会话
app.delete('/api/session/:sessionId', async (req, res) => {
try {
const { sessionId } = req.params;
const result = await aiHandler.cleanupSession(sessionId);
sessions.delete(sessionId);
res.json({
success: true,
sessionId,
message: '会话清理成功'
});
} catch (error) {
console.error('Session cleanup error:', error);
res.status(500).json({
success: false,
error: error.message
});
}
});
// 会话清理任务（定期清理过期会话）
setInterval(() => {
const now = newDate();
const SESSION_TIMEOUT = 30 * 60 * 1000; // 30分钟
for (const [sessionId, session] of sessions.entries()) {
if (now - session.lastActivity > SESSION_TIMEOUT) {
console.log(清理过期会话: ${sessionId});
aiHandler.cleanupSession(sessionId);
sessions.delete(sessionId);
}
}
}, 5 * 60 * 1000); // 每5分钟检查一次
// 错误处理中间件
app.use((error, req, res, next) => {
console.error('Unhandled error:', error);
res.status(500).json({
success: false,
error: '服务器内部错误'
});
});
// 404 处理
app.use('*', (req, res) => {
res.status(404).json({
success: false,
error: '接口不存在'
});
});
// 启动服务器
app.listen(PORT, () => {
console.log(AI Web Assistant 服务器运行在端口 ${PORT});
console.log(环境: ${process.env.NODE_ENV});
});
exportdefault app;
四、前端界面实现
创建 public/index.html：
<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>AI 网页操作助手</title>
<script src="https://cdn.tailwindcss.com"></script>
<style>
.message-user { background-color: #3b82f6; color: white; }
.message-assistant { background-color: #e5e7eb; color: #374151; }
.typing-indicator { display: inline-block; }
.typing-dot {
display: inline-block;
width: 8px; height: 8px;
background-color: #9ca3af;
border-radius: 50%;
margin: 02px;
animation: typing 1.4s infinite ease-in-out;
}
.typing-dot:nth-child(1) { animation-delay: -0.32s; }
.typing-dot:nth-child(2) { animation-delay: -0.16s; }
@keyframes typing {
0%, 80%, 100% { transform: scale(0); }
40% { transform: scale(1); }
}
</style>
</head>
<body class="bg-gray-100 min-h-screen">
<div class="container mx-auto px-4 py-8 max-w-4xl">
<header class="text-center mb-8">
<h1 class="text-3xl font-bold text-gray-800 mb-2">AI 网页操作助手</h1>
<p class="text-gray-600">使用自然语言指令自动化网页操作</p>
</header>
<div class="bg-white rounded-lg shadow-lg overflow-hidden">
<div class="bg-gray-800 text-white p-4 flex justify-between items-center">
<div>
<span id="sessionStatus" class="text-sm">未连接</span>
</div>
<div class="space-x-2">
<button id="initSession" class="bg-green-600 hover:bg-green-700 px-4 py-2 rounded text-sm">
开始新会话
</button>
<button id="clearSession" class="bg-red-600 hover:bg-red-700 px-4 py-2 rounded text-sm" disabled>
结束会话
</button>
</div>
</div>
<!-- 聊天区域 -->
<div class="h-96 overflow-y-auto p-4 space-y-4" id="chatMessages">
<div class="text-center text-gray-500 py-8">
发送指令开始与AI助手对话
</div>
</div>
<!-- 输入区域 -->
<div class="border-t p-4">
<div class="flex space-x-2">
<input
type="text"
id="instructionInput"
placeholder="输入你的指令，例如：打开百度并搜索AI最新进展..."
class="flex-1 border rounded-lg px-4 py-2 focus:outline-none focus:ring-2 focus:ring-blue-500"
disabled>
<button
id="sendButton"
class="bg-blue-600 hover:bg-blue-700 text-white px-6 py-2 rounded-lg disabled:bg-gray-400 disabled:cursor-not-allowed"
disabled>
发送
</button>
</div>
<div class="mt-2 text-sm text-gray-500">
<p>示例指令：</p>
<div class="flex flex-wrap gap-2 mt-1">
<button class="example-instruction text-xs bg-gray-200 hover:bg-gray-300 px-2 py-1 rounded" data-instruction="打开百度首页">打开百度</button>
<button class="example-instruction text-xs bg-gray-200 hover:bg-gray-300 px-2 py-1 rounded" data-instruction="搜索今天的热门新闻">搜索新闻</button>
<button class="example-instruction text-xs bg-gray-200 hover:bg-gray-300 px-2 py-1 rounded" data-instruction="提取当前页面的所有标题">提取标题</button>
</div>
</div>
</div>
</div>
<!-- 会话信息 -->
<div class="mt-4 bg-white rounded-lg shadow p-4">
<h3 class="font-semibold mb-2">会话信息</h3>
<div class="text-sm space-y-1">
<div>会话ID: <span id="sessionIdDisplay" class="font-mono">-</span></div>
<div>状态: <span id="connectionStatus">未连接</span></div>
<div>消息数: <span id="messageCount">0</span></div>
</div>
</div>
</div>
<script>
class AIAssistant {
constructor() {
this.sessionId = null;
this.isConnected = false;
this.messageCount = 0;
this.initializeElements();
this.attachEventListeners();
}
initializeElements() {
this.sessionStatus = document.getElementById('sessionStatus');
this.sessionIdDisplay = document.getElementById('sessionIdDisplay');
this.connectionStatus = document.getElementById('connectionStatus');
this.messageCountDisplay = document.getElementById('messageCount');
this.chatMessages = document.getElementById('chatMessages');
this.instructionInput = document.getElementById('instructionInput');
this.sendButton = document.getElementById('sendButton');
this.initSessionBtn = document.getElementById('initSession');
this.clearSessionBtn = document.getElementById('clearSession');
}
attachEventListeners() {
this.initSessionBtn.addEventListener('click', () => this.initializeSession());
this.clearSessionBtn.addEventListener('click', () => this.clearSession());
this.sendButton.addEventListener('click', () => this.sendInstruction());
this.instructionInput.addEventListener('keypress', (e) => {
if (e.key === 'Enter') this.sendInstruction();
});
// 示例指令点击事件
document.querySelectorAll('.example-instruction').forEach(btn => {
btn.addEventListener('click', (e) => {
this.instructionInput.value = e.target.dataset.instruction;
this.sendInstruction();
});
});
}
async initializeSession() {
try {
this.showLoading('正在初始化会话...');
const response = await fetch('/api/session/init', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({})
});
const data = await response.json();
if (data.success) {
this.sessionId = data.sessionId;
this.isConnected = true;
this.messageCount = 0;
this.updateUI();
this.addMessage('system', '会话已初始化，你可以开始发送指令了。');
} else {
thrownewError(data.error);
}
} catch (error) {
this.addMessage('error', `初始化失败: ${error.message}`);
} finally {
this.hideLoading();
}
}
async sendInstruction() {
const instruction = this.instructionInput.value.trim();
if (!instruction || !this.isConnected) return;
// 添加用户消息
this.addMessage('user', instruction);
this.instructionInput.value = '';
// 显示输入状态
const thinkingMessage = this.addMessage('assistant', '');
this.showTypingIndicator(thinkingMessage);
try {
const response = await fetch('/api/instruction', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({
sessionId: this.sessionId,
instruction: instruction
})
});
const data = await response.json();
// 移除输入状态
this.removeTypingIndicator(thinkingMessage);
if (data.success) {
this.addMessage('assistant', data.response);
} else {
this.addMessage('error', `操作失败: ${data.error}`);
}
} catch (error) {
this.removeTypingIndicator(thinkingMessage);
this.addMessage('error', `网络错误: ${error.message}`);
}
}
async clearSession() {
if (!this.sessionId) return;
try {
await fetch(`/api/session/${this.sessionId}`, {
method: 'DELETE'
});
} catch (error) {
console.error('清理会话失败:', error);
}
this.sessionId = null;
this.isConnected = false;
this.messageCount = 0;
this.updateUI();
this.clearMessages();
this.addMessage('system', '会话已结束。点击"开始新会话"重新开始。');
}
addMessage(role, content) {
this.messageCount++;
this.messageCountDisplay.textContent = this.messageCount;
const messageDiv = document.createElement('div');
messageDiv.className = `p-3 rounded-lg max-w-3/4 ${
role === 'user' ? 'message-user ml-auto' :
role === 'error' ? 'bg-red-100 text-red-800 border border-red-200' :
'message-assistant'
}`;
if (role === 'thinking') {
messageDiv.innerHTML = '<div class="typing-indicator"><span class="typing-dot"></span><span class="typing-dot"></span><span class="typing-dot"></span></div>';
} else {
messageDiv.textContent = content;
}
this.chatMessages.appendChild(messageDiv);
this.chatMessages.scrollTop = this.chatMessages.scrollHeight;
return messageDiv;
}
showTypingIndicator(messageElement) {
messageElement.innerHTML = '<div class="typing-indicator"><span class="typing-dot"></span><span class="typing-dot"></span><span class="typing-dot"></span></div>';
}
removeTypingIndicator(messageElement) {
messageElement.innerHTML = '';
}
clearMessages() {
this.chatMessages.innerHTML = '<div class="text-center text-gray-500 py-8">发送指令开始与AI助手对话</div>';
}
showLoading(message) {
this.initSessionBtn.disabled = true;
this.initSessionBtn.textContent = message;
}
hideLoading() {
this.initSessionBtn.disabled = false;
this.initSessionBtn.textContent = '开始新会话';
}
updateUI() {
this.sessionStatus.textContent = this.isConnected ? '已连接' : '未连接';
this.sessionIdDisplay.textContent = this.sessionId || '-';
this.connectionStatus.textContent = this.isConnected ? '活跃' : '未连接';
this.connectionStatus.className = this.isConnected ? 'text-green-600' : 'text-red-600';
this.instructionInput.disabled = !this.isConnected;
this.sendButton.disabled = !this.isConnected;
this.clearSessionBtn.disabled = !this.isConnected;
}
}
// 初始化应用
document.addEventListener('DOMContentLoaded', () => {
new AIAssistant();
});
</script>
</body>
</html>
五、测试与验证
1. 创建测试脚本
创建 tests/integration.test.js：
import { test, expect } from'@playwright/test';
import AIHandler from'../src/ai/handler.js';
import dotenv from'dotenv';
dotenv.config();
test.describe('AI Web Assistant Integration Tests', () => {
let aiHandler;
let sessionId;
test.beforeEach(async () => {
aiHandler = new AIHandler(process.env.ANTHROPIC_API_KEY);
const initResult = await aiHandler.initializeSession();
sessionId = initResult.sessionId;
});
test.afterEach(async () => {
await aiHandler.cleanupSession(sessionId);
});
test('should initialize session successfully', async () => {
expect(sessionId).toBeDefined();
expect(typeof sessionId).toBe('string');
});
test('should process simple navigation instruction', async () => {
const result = await aiHandler.processInstruction(
sessionId,
'请打开百度首页 https://www.baidu.com'
);
expect(result.success).toBe(true);
expect(result.response).toBeDefined();
});
test('should handle invalid instruction gracefully', async () => {
const result = await aiHandler.processInstruction(
sessionId,
'执行一个不存在的操作'
);
// 即使指令有问题，也应该有合理的响应
expect(result.response).toBeDefined();
});
});
2. 运行测试
npm test
六、部署与运行
1. 生产环境配置
创建 ecosystem.config.js：
module.exports = {
apps: [{
name: 'ai-web-assistant',
script: 'server.js',
instances: 'max',
exec_mode: 'cluster',
env: {
NODE_ENV: 'production',
PORT: 3000
},
env_production: {
NODE_ENV: 'production'
}
}]
};
2. Docker 配置
创建 Dockerfile：
FROM node:18-alpine
WORKDIR /app安装 Playwright 依赖
RUN apk add --no-cache
chromium
nss
freetype
freetype-dev
harfbuzz
ca-certificates
ttf-freefont设置环境变量
ENV PUPPETEER_SKIP_CHROMIUM_DOWNLOAD=true
ENV PUPPETEER_EXECUTABLE_PATH=/usr/bin/chromium-browser复制 package.json 并安装依赖
COPY package*.json ./
RUN npm ci --only=production复制源代码
COPY . .
创建非root用户
RUN addgroup -g 1001 -S nodejs
RUN adduser -S nextjs -u 1001
USER nextjs
EXPOSE3000
CMD ["npm", "start"]
3. 启动应用
开发模式
npm run dev
生产模式
npm start
七、实际应用场景
场景1：自动化数据收集
// 指令：收集 GitHub 趋势项目
const instruction =请访问 GitHub Trending 页面 (https://github.com/trending)， 收集今天最流行的 JavaScript 项目的前5名， 包括项目名称、星标数和描述， 并整理成 JSON 格式返回。;
场景2：自动化表单填写
// 指令：注册测试用户
const instruction = `
请打开我们的测试注册页面 http://localhost:3000/register，
填写以下信息：
- 用户名: testuser_${Date.now()}
- 邮箱: test${Date.now()}@example.com
- 密码: TestPassword123
然后点击注册按钮，并确认注册成功。
`;
场景3：复杂工作流程
// 指令：完整的电商流程测试
const instruction = `
请执行以下电商购物流程：
- 登录到测试电商网站
- 搜索"笔记本电脑"
- 选择第一个商品
- 添加到购物车
- 进入结算流程
- 填写测试配送信息
- 确认订单
请在每个步骤完成后报告状态。
`;
总结
通过本教程，我们成功构建了一个功能完整的 AI 网页操作助手，具备以下特点：
- 完整的架构：从前端界面到后端服务，再到浏览器自动化层
- 灵活的 MCP 协议：支持多种网页操作工具
- 智能的 AI 集成：利用 Claude 模型理解自然语言指令
- 健壮的错误处理：能够应对各种网页异常情况
- 可扩展的设计：易于添加新的工具和功能
这个项目展示了如何将现代 AI 技术与浏览器自动化相结合，创造出能够理解并执行复杂网页操作的智能助手。你可以在此基础上继续扩展，比如添加视觉识别、多浏览器支持、分布式任务处理等功能，打造更强大的自动化解决方案。
立即开始构建你自己的 AI 网页助手，释放自动化的无限可能！
