eaight
An AI-native web browser that exposes a full MCP server, letting Claude Code, Codex CLI, and Gemini CLI see and control the browser in real time.
README
Eaight
An AI-native web browser that exposes a full Model Context Protocol (MCP) server, letting Claude Code, Codex CLI, and Gemini CLI see and control the browser in real time.
What it does
Eaight is a Chromium-based desktop browser built for AI-assisted workflows. It works like any browser for daily use, but runs a built-in MCP server that gives AI tools direct access to the page:
- 18 browser automation tools: navigate, click, type, scroll, screenshot, extract elements, execute JavaScript — all available as MCP tools
- 9 live resources: DOM, text content, accessibility tree, network logs, console output, cookies, screenshots — streamed to the AI in real time
- 5 interaction prompts: pre-built templates for page analysis, form filling, data extraction, and navigation
- Multi-transport: WebSocket and SSE transports run simultaneously for maximum compatibility
- Auto-discovery: Claude Code detects the browser automatically via a discovery file — no manual configuration needed
Beyond automation
- Tab management with groups
- Chrome extension support (install from Chrome Web Store)
- Prompt library with GitHub sync
- Custom new tab page with 3D animation
- Light/dark/system themes
- Security controls for MCP data sharing (screenshots, DOM, cookies — each toggleable)
Built with
- Desktop: Electron 28 + TypeScript
- Frontend: React 18 + Zustand + Tailwind CSS
- MCP: @modelcontextprotocol/sdk 1.0.4
- AI SDKs: @anthropic-ai/sdk 0.32.1
- Graphics: Three.js (new tab page animation)
- Build: Vite + electron-builder
Architecture
Claude Code / Codex / Gemini CLI
│
│ JSON-RPC (stdio or WebSocket)
▼
┌─────────────────────────────────┐
│ MCP Server (port 9222) │
│ │
│ ┌───────────┐ ┌────────────┐ │
│ │ 18 Tools │ │ 9 Resources│ │
│ │ navigate │ │ dom │ │
│ │ click │ │ text │ │
│ │ type │ │ screenshot │ │
│ │ scroll │ │ a11y tree │ │
│ │ screenshot│ │ network │ │
│ │ ... │ │ console │ │
│ └─────┬─────┘ └─────┬──────┘ │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────────────────┐ │
│ │ Tab Manager │ │
│ │ (Electron BrowserView) │ │
│ └─────────────────────────┘ │
└─────────────────────────────────┘
│
▼
┌─────────────────────────────────┐
│ React UI │
│ Tab Bar │ URL Bar │ AI Panel │
│ Bookmarks │ Settings │
└─────────────────────────────────┘
The MCP server runs inside the Electron main process with direct access to BrowserViews. A stdio bridge (mcp-stdio-bridge.ts) translates between Claude Code's stdio protocol and the internal WebSocket, so Claude Code can connect without network configuration.
Auto-discovery writes connection info to %APPDATA%/eaight/mcp-server.json, which Claude Code reads to find the browser.
Getting started
Prerequisites
- Node.js 18+
- Claude Code (to use the MCP integration)
Installation
npm install
npm run dev
The MCP server starts automatically on port 9222. Claude Code detects it via auto-discovery.
CLI
Control the browser from the command line:
node eaight-cli.js nav https://example.com
node eaight-cli.js text # extract page text
node eaight-cli.js screenshot # capture screenshot
node eaight-cli.js click "button.submit"
Configuration
Settings are available in the browser UI (gear icon):
- AI Settings: MCP server port, authentication mode
- Security: control what data is shared with AI (screenshots, DOM, cookies)
- Appearance: theme, AI panel position
License
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。