vscode-playwright-mcp

vscode-playwright-mcp

Gives LLMs full control over VS Code — click UI elements, type code, run commands, take screenshots, read editor state, and record GIFs.

Category
访问服务器

README

vscode-playwright-mcp

A VS Code-aware layer on top of @playwright/mcp. Gives LLMs full control over VS Code — click UI elements, type code, run commands, take screenshots, read editor state, and record GIFs.

Use cases:

  • E2E testing of VS Code extensions
  • Automated VS Code workflows and tasks
  • AI-driven code editing and refactoring
  • Interactive demos and tutorials
  • Testing themes, keybindings, and settings

How It Works

Built on top of @playwright/mcp, which provides generic Electron automation (click, type, screenshot, etc.). We add a VS Code-specific layer: isolated launch with temp user-data-dir, command palette automation, editor state scraping, and Monaco-aware tools.

Generic tools from @playwright/mcp are aliased to vscode_* names for a unified LLM experience. Before vscode_launch, only 1 tool is visible (lazy discovery saves ~10k context tokens). After launch, all 26 tools appear.

Requirements

  • Node.js >= 22
  • VS Code installed locally (auto-detected on macOS and Linux)

Quick Start

{
  "mcpServers": {
    "vscode": {
      "command": "npx",
      "args": ["-y", "vscode-playwright-mcp"]
    }
  }
}

Custom VS Code binary:

{
  "mcpServers": {
    "vscode": {
      "command": "npx",
      "args": ["-y", "vscode-playwright-mcp", "--vscode-path", "/path/to/code"]
    }
  }
}

Tools

Lifecycle

Tool Description
vscode_launch Launch an isolated VS Code instance. Options: workspace, extensions, viewport size, settings.
vscode_close Close VS Code and clean up temp files/processes.

Vision (aliased from @playwright/mcp)

Tool Description
vscode_screenshot Full-window screenshot.
vscode_snapshot Accessibility tree with [ref=eN] annotations on interactive elements.
vscode_zoom Cropped screenshot of a specific region (native — Monaco-aware).
vscode_find_element Search a11y tree by role/name, returns refs (native).

Mouse & Keyboard (aliased from @playwright/mcp)

Tool Description
vscode_click Click by ref, element description, or coordinates.
vscode_type Type text into focused element.
vscode_press_key Press key combinations (e.g., Control+Shift+P).
vscode_hover Hover to reveal tooltips.
vscode_drag Drag from one point to another.
vscode_click_xy Click at exact pixel coordinates.
vscode_hover_xy Hover at exact pixel coordinates.
vscode_drag_xy Drag between exact pixel coordinates.
vscode_select_option Select from dropdown/combobox.
vscode_fill_form Fill form inputs.

VS Code Commands (native)

Tool Description
vscode_run_command Execute a VS Code command via Command Palette automation.
vscode_ensure_file Open and activate a file by path with verification.

State (native)

Tool Description
vscode_get_state Read editor state: active file, cursor, diagnostics, completions, peek widget.
vscode_get_hover Read hover tooltip content as text.

Advanced (mixed)

Tool Description
vscode_evaluate Run JavaScript in the VS Code renderer process (aliased).
vscode_wait_for Wait for selector state, text, or delay (aliased).
vscode_console Retrieve console messages from renderer (aliased).
vscode_resize Resize the VS Code window and viewport (native).
vscode_scroll Scroll up/down by amount (native).
vscode_gif Record actions as animated GIF (native).

Interaction Strategies

  1. Ref-based (preferred for UI): vscode_snapshot → see [ref=eN] annotations → vscode_click(ref="e5"). Deterministic, no coordinate guessing.

  2. Keyboard-first (preferred for navigation): vscode_snapshot → see shortcuts → vscode_press_key. Fastest for commands.

  3. Visual (required for editor content): Monaco editor is opaque to accessibility APIs. vscode_screenshot or vscode_zoom → read the image → vscode_click_xy(x, y).

How is this different from Playwright MCP?

This project is built on top of @playwright/mcp, not a replacement. Playwright MCP provides generic Electron automation. We add:

  • Isolated VS Code launch with temp user-data-dir, suppressed Welcome tab, settings injection
  • Command palette automation (vscode_run_command) — execute commands by name
  • Editor state scraping (vscode_get_state) — cursor, diagnostics, completions, peek widget
  • Monaco-aware tools (vscode_zoom, vscode_find_element) — the editor is invisible to generic a11y tools
  • Lazy tool discovery — 1 tool visible before launch, all ~26 after (saves context tokens)
  • GIF recording of VS Code sessions

Development

npm run build          # Build with tsup
npm run dev            # Build in watch mode
npm test               # Run unit tests (vitest)
npm run test:integration  # Run integration tests (requires VS Code)
npm run test:e2e       # Run E2E tests (full MCP protocol loop)
npm run typecheck      # Type-check with tsc
npm run lint           # Lint with ESLint

License

MIT

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选