mcp-AI-playwright-template
A template for building a custom Playwright MCP server that enables AI-driven browser automation, translating high-level test intentions into dynamic Playwright actions.
README
This project is set up to debug with LOCAL AI
This is in sync with C:\MCP_BDC\mcpServer-Playwright-templateV8 on author's hard drive.
Custom MCP-Playwrght Server
Architecture Overview and Configuration
High-level test cases are processed by the AI and translated into structured JSON commands that are natively understood by the Model Context Protocol (MCP) server.
Underlying Language Model: The project leverages the Phi-3 Mini LLM for this translation layer.
Client Requirements: The MCP-aware AI client (in this architecture, the Playwright MCP server itself) requires a specific configuration file to establish communication.
Configuration File Path
Ensure the configuration file is present at the following location, replacing {userName} with your local system username:
C:\Users{userName}.config\mcp\clients\playwright-mcp-config.json This file tells the MCP server to expect you calls from:
{
"servers": {
"playwright-mcp": {
"command": "node",
"args": ["./mcp/server.js"],
"env": {
"NODE_ENV": "development"
}
}
}
}
Compiling TypeScript Files to Distinct Target Directories
To compile TypeScript (.ts) source files into JavaScript (.js) and distribute them to specific target directories, you can leverage dedicated TypeScript configuration files (e.g., tsconfig.*.json).
Run the standard project build script
npm run build
Compile using a specific TypeScript configuration target
npx tsc -p tsconfig.prompts.json
In another window verify if ollama is installed
DOS PROMPT>>>ollama list
possibly it is scheduled to start on boot
DOS PROMPT>>>tasklist | findstr ollama
verify if it is running
DOS PROMPT>>>>curl http://localhost:11434/api/tags
If not start Phi-3::
DOS PROMPT>>>ollama run phi3
It it is expected to fail in GitHub Action Workflow every time. To run other AI's you need to replace the code below
async function callOllama(prompt) {
const response = await axios.post("http://localhost:11434/api/generate", {
model: "phi3",
prompt: prompt,
stream: false
});
RUN RESULT:
TOOL RECEIVED: close_browser
ARGS: undefined
=======================
[close_browser] Closing browser
[close_browser] Page closed
[close_browser] Browser closed
========================
📊 TEST SUMMARY
========================
Test 1: ✅ PASS - BDC Entrepreneur support
Test 2: ✅ PASS - BDC reCAPTCHA Link must be present
Test 3: ✅ PASS - BDC Should Not Contain a string MakunaimaRA
Test 4: ❌ FAIL - BDC must contain string 'Warszwa'
Test 5: ✅ PASS - BDC must contain string 'Solutions'
============================================================
Total: 4/5 tests passed
TOTAL TIME: 116572ms
============================================================
## Test cases are re-defined by AI ( Phi-3-mini ) but they are not truly plain English.
For example:
[STDIN] {"method":"tools/call","params":{"name":"search_text","arguments":{"text":"Warszwa"}},"jsonrpc":"2.0","id":15}
is constructed from :
{
"id": 4,
"question": "Is the text 'Warszwa' present on the BDC website?",
"description": "BDC must contain string 'Warszwa'",
"baseUrl": "https://www.bdc.ca/fr",
"expectation": "should-contain",
"searchTerm": "Warszwa"
}
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。