Chromate MCP
A Model Context Protocol server for Chrome tab automation using the Chrome DevTools Protocol, enabling AI agents to select tabs, capture screenshots, and interact with web pages via CSS viewport coordinates.
README
Chromate MCP
Chromate MCP is a Model Context Protocol server for Chrome tab automation over the Chrome DevTools Protocol (CDP). It is designed for AI agents that work from screenshots: select a tab, inspect a coordinate grid, click CSS viewport coordinates, then receive an after screenshot.
Capabilities
- Connect to Chrome remote debugging from a fixed endpoint, Chrome 144+ auto-connect metadata, or local CDP port discovery.
- List and select Chrome tabs by CDP target id.
- Capture viewport screenshots with an optional grid and crosshair overlay.
- Click, scroll, type text, press keys, wait, and read page/viewport info.
- Keep all coordinates in CSS viewport pixels, matching CDP mouse events.
Chromate controls the web page content area. It does not operate the Chrome address bar, native tab strip, extension popups, file pickers, or OS windows.
Install
npm install
npm run build
Start Chrome
Chromate does not launch Chrome automatically. It can connect to an already-running Chrome in two ways.
For Chrome 144+, open chrome://inspect/#remote-debugging in Chrome and enable remote debugging. Chromate reads Chrome's DevToolsActivePort metadata by default and connects to the running browser after Chrome shows and you approve the permission dialog.
For older Chrome versions or sandbox/VM setups, start Chrome with a remote debugging port:
google-chrome-stable \
--remote-debugging-address=127.0.0.1 \
--remote-debugging-port=9222 \
--user-data-dir=/tmp/chromate-profile
Chromate auto-discovers local Chrome on ports 9222, 9223, 9224, and 9333 when CHROMATE_CDP_ENDPOINT is not set. Any Chromium-based browser that exposes CDP can work if it supports /json/version and a browser-level WebSocket endpoint.
MCP Configuration
Example client configuration:
{
"mcpServers": {
"chromate": {
"command": "node",
"args": ["/data0/chromate/dist/index.js"],
"env": {}
}
}
}
Set CHROMATE_CDP_ENDPOINT only when you want to force a specific CDP HTTP or WebSocket endpoint.
During development, use:
npm run dev
Environment
CHROMATE_CDP_ENDPOINT: CDP HTTP or WebSocket endpoint. If omitted, Chromate auto-discovers local CDP.CHROMATE_AUTO_CONNECT: read Chrome 144+DevToolsActivePortmetadata before scanning ports. Default:trueCHROMATE_AUTO_CONNECT_CHANNEL: Chrome channel for default profile lookup:stable,beta,dev, orcanary. Default:stableCHROMATE_AUTO_CONNECT_USER_DATA_DIR: explicit Chrome user data directory containingDevToolsActivePortCHROMATE_CDP_DISCOVERY_PORTS: comma-separated local ports to scan. Default:9222,9223,9224,9333CHROMATE_DISCOVERY_TIMEOUT_MS: per-port discovery timeout. Default:350CHROMATE_CONNECT_TIMEOUT_MS: connection timeout. Default:10000CHROMATE_ACTION_TIMEOUT_MS: command timeout. Default:30000CHROMATE_SETTLE_DELAY_MS: wait after auto actions. Default:500CHROMATE_GRID_STEP: screenshot grid spacing. Default:100CHROMATE_LOG_LEVEL:silent,error,info, ordebug. Default:info
Workflow
- Call
list_tabs. - Call
select_tabwith the desiredtabId. - Call
screenshotand inspect the grid. - Call
clickwith CSS viewport coordinates. - Use the returned after screenshot to continue.
See docs/tool-contract.md for the full tool contract.
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。