macOS Companion MCP Server
Enables AI agents to directly access native macOS services, media, system health, and administration tools through a local MCP server.
README
<p align="center"> <img src="https://raw.githubusercontent.com/modelcontextprotocol/logo/main/mcp-logo-dark.svg" alt="MCP Logo" width="120" /> </p>
macOS Companion MCP Server 💻
The Local Bridge Between AI Agents and the macOS Substrate.
You don't need another generic electron wrapper. You need raw, direct access to the native data structures already running on your machine. macos-mcp is a bare-metal Model Context Protocol (MCP) server that hooks your agents (OpenClaw, Claude Desktop, Cursor) directly into native macOS services, media, and system administration utilities.
This is critical because context engineering is an ongoing exercise. A simple file-read tool won't help when you need to know if the CPU is throttling, what your latest podcast playback head is, or if Apple Notes is refusing to sync.
🎯 The Toolset
Highly granular tools exposed directly to your agent layer:
| Domain | Tools | Actionable Intel Provided |
|---|---|---|
| System Health | run_health_audit, get_battery_health, get_storage_scan, get_startup_items |
Machine Intel. Deep dive into thermal pressure, cycle counts, memory swaps, and local Time Machine snapshots. |
| Native Apps | list_calendars, get_calendar_events, create_reminder, get_note, send_imessage |
Productivity. Headless access to Reminders, Notes, and Calendar without UI bloat. |
| Media / Web | get_music_state, get_safari_tabs, get_recent_podcast_episodes |
Context. Directly reads MTLibrary.sqlite to find exact podcast listening playback heads. |
| Admin | run_disk_cleanup, kill_process, restart_service, run_shortcut |
Control. Triggers native /usr/bin/shortcuts and clears DerivedData or user caches safely. |
🚀 Getting Started
1. Installation
Clone it, build it, and point your MCP config at the dist/index.js file.
git clone https://github.com/surendranb/macos-mcp.git
cd macos-mcp
npm install
npm run build
2. Configuration (Claude Desktop / Cursor)
Add the server to your agent's MCP settings configuration file:
{
"mcpServers": {
"macos-companion-mcp": {
"command": "node",
"args": ["/absolute/path/to/macos-mcp/dist/index.js"]
}
}
}
Note: Replace /absolute/path/to/ with the actual path to where you cloned the repository.
🛠️ Project Philosophy (The "Anti-Slop" Rules)
This project focuses on bare-metal efficiency for local AI agents:
- Zero Sloppy Abstractions: We hit
/usr/bin/shortcutsexplicitly. No PATH collisions with python environments. - No Bulky SQLite Bindings: The server executes queries against local databases using macOS's built-in
sqlite3command-line utility. This prevents heavy native compilation failures ofnode-sqlitepackages on Apple Silicon. - Fire and Forget Warmups: Notes and Reminders have aggressive headless cold-start times. We background
osascriptcalls at startup to warm the process memory so the agent isn't left waiting. - Local Only: All processing stays on the machine.
🧪 Testing
A regular laptop combined with basic testing harnesses is enough. Run npm run test to spin up a headless stdio test suite that probes the endpoints and outputs a clean success dashboard.
License
MIT 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。