mac-pilot-mcp
Self-learning macOS automation MCP server that remembers successful patterns and avoids repeated failures, enabling AI agents to execute multi-step workflows via reusable recipes.
README
<p align="center"> <h1 align="center">mac-pilot-mcp</h1> <p align="center"> <strong>Self-learning macOS automation for AI agents</strong> </p> <p align="center"> The only macOS MCP server that <em>remembers</em>. Save what works, skip what doesn't. </p> <p align="center"> <a href="https://www.npmjs.com/package/mac-pilot-mcp"><img src="https://img.shields.io/npm/v/mac-pilot-mcp?color=cb3837&label=npm" alt="npm version"></a> <a href="https://www.npmjs.com/package/mac-pilot-mcp"><img src="https://img.shields.io/npm/dm/mac-pilot-mcp?color=cb3837" alt="npm downloads"></a> <a href="https://github.com/leesgit/mac-pilot-mcp"><img src="https://img.shields.io/github/stars/leesgit/mac-pilot-mcp?style=social" alt="GitHub stars"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue" alt="License"></a> <a href="#"><img src="https://img.shields.io/badge/platform-macOS-000?logo=apple&logoColor=white" alt="macOS"></a> <a href="#"><img src="https://img.shields.io/badge/node-%3E%3D18-339933?logo=node.js&logoColor=white" alt="Node.js"></a> </p> </p>
The Problem
There are 7+ macOS automation MCP servers. None of them remember anything. Every session starts from zero — same trial and error, same failures, same wasted tokens.
The Solution
Mac-Pilot is different. It learns from every interaction:
- Success? The pattern is auto-saved as app-specific knowledge
- Failure? The error is recorded so it won't repeat the same mistake
- Multi-step workflow? Save it as a recipe — replay it in one call next time
Ships with 21 built-in recipes so you're productive from the first run.
Quick Start
Install
npm install -g mac-pilot-mcp
Connect to your AI client
<details> <summary><strong>Claude Code</strong></summary>
claude mcp add mac-pilot -- mac-pilot-mcp
Or manually add to ~/.claude.json:
{
"mcpServers": {
"mac-pilot": {
"command": "mac-pilot-mcp"
}
}
}
</details>
<details> <summary><strong>Claude Desktop</strong></summary>
Add to claude_desktop_config.json:
{
"mcpServers": {
"mac-pilot": {
"command": "npx",
"args": ["-y", "mac-pilot-mcp"]
}
}
}
</details>
<details> <summary><strong>Cursor</strong></summary>
Add to your MCP settings (.cursor/mcp.json):
{
"mcpServers": {
"mac-pilot": {
"command": "npx",
"args": ["-y", "mac-pilot-mcp"]
}
}
}
</details>
<details> <summary><strong>Windsurf</strong></summary>
Add to your MCP config:
{
"mcpServers": {
"mac-pilot": {
"command": "npx",
"args": ["-y", "mac-pilot-mcp"]
}
}
}
</details>
Grant Permissions
Mac-Pilot needs Accessibility access for UI automation:
- System Settings → Privacy & Security → Accessibility
- Toggle ON for your terminal app (Terminal, iTerm2, VS Code, Cursor, etc.)
- Restart the terminal
How It Works
You: "Export the current Figma frame as PNG"
┌─ First time ─────────────────────────────────────────┐
│ │
│ 1. mac_recipe_search("Figma export") → No matches │
│ 2. mac_state() → Figma is frontmost │
│ 3. mac_run(applescript) → File > Export > PNG │
│ 4. mac_recipe_save("export-figma-png", steps=[...]) │
│ │
│ ✓ Worked! Pattern saved automatically. │
└───────────────────────────────────────────────────────┘
┌─ Next time ───────────────────────────────────────────┐
│ │
│ 1. mac_recipe_search("Figma export") │
│ → Found: export-figma-png (100% success rate) │
│ 2. mac_recipe_run("export-figma-png") │
│ → Done instantly │
│ │
│ ⚡ 4 steps → 2 steps. No trial and error. │
└───────────────────────────────────────────────────────┘
Tools
| Tool | What it does |
|---|---|
mac_run |
Execute AppleScript, JXA, shell commands, open apps/URLs, click, type, keypress |
mac_state |
Query system state — frontmost app, windows, clipboard, running apps |
mac_find_ui |
Find UI elements via Accessibility API (buttons, fields, menus) |
mac_screenshot |
Capture screen/window/region as base64 PNG |
mac_recipe_save |
Save a working action sequence as a reusable recipe |
mac_recipe_run |
Replay a saved recipe with parameter substitution |
mac_recipe_search |
Full-text search across recipes and action history |
Built-in Recipes
21 recipes ship out of the box — no setup needed:
| Category | Recipes | Example |
|---|---|---|
| System | toggle-dark-mode set-volume mute-toggle lock-screen show-desktop screenshot-desktop empty-trash get-dark-mode |
mac_recipe_run { name: "toggle-dark-mode" } |
| Finder | new-finder-window get-selected-files |
mac_recipe_run { name: "new-finder-window", params: { path: "/tmp" } } |
| Safari | safari-current-url safari-current-title |
mac_recipe_run { name: "safari-current-url" } |
| Clipboard | get-clipboard set-clipboard |
mac_recipe_run { name: "set-clipboard", params: { text: "Hello" } } |
| Notifications | notify |
mac_recipe_run { name: "notify", params: { title: "Done", message: "Build passed" } } |
| Terminal | open-terminal-at kill-process |
mac_recipe_run { name: "open-terminal-at", params: { path: "~/dev" } } |
| Windows | list-windows close-front-window |
mac_recipe_run { name: "close-front-window" } |
| Music | music-play-pause music-next-track |
mac_recipe_run { name: "music-play-pause" } |
Examples
AppleScript
{ "actionType": "applescript", "script": "tell application \"Finder\" to get name of every window" }
JXA (JavaScript for Automation)
{ "actionType": "jxa", "script": "Application('Safari').documents[0].url()" }
Shell command
{ "actionType": "shell", "command": "ls -la ~/Desktop" }
Open an app or URL
{ "actionType": "open", "target": "Safari" }
{ "actionType": "open", "target": "https://github.com" }
Type text
{ "actionType": "type", "text": "Hello World" }
Keyboard shortcut
{ "actionType": "keypress", "text": "cmd+c" }
{ "actionType": "keypress", "text": "cmd+shift+4" }
Find UI elements
mac_find_ui { "app": "Safari", "role": "AXButton" }
mac_find_ui { "app": "Finder", "searchText": "Downloads" }
Electron apps (VSCode / Cursor / Slack / Discord) — macOS Accessibility
exposes a thin tree for Electron, so Mac-Pilot can optionally fall back to
Chrome DevTools Protocol when the app is launched with
--remote-debugging-port=<PORT>. See docs/ELECTRON-SUPPORT.md.
mac_find_ui {
"app": "Visual Studio Code",
"searchText": "Run Test",
"useElectronFallback": "auto"
}
Take a screenshot
mac_screenshot { "target": "screen", "scale": 0.3 }
mac_screenshot { "target": "window", "windowName": "Safari" }
Save a custom recipe
mac_recipe_save {
"name": "open-project",
"description": "Open VS Code at project directory",
"steps": [
{ "actionType": "shell", "params": { "command": "code {{path}}" }, "description": "Open VS Code" }
],
"parameters": [
{ "name": "path", "description": "Project directory path" }
],
"tags": ["dev", "vscode"]
}
Security
Mac-Pilot blocks dangerous operations before they execute:
| Layer | Protection |
|---|---|
| Hard block | sudo, rm -rf /, curl|sh, dd if=, $() subshell injection, keychain access, csrutil disable, diskutil erase, and 20+ patterns |
| Risk classification | Every action is rated low / medium / high / blocked |
| Audit log | All actions (including blocked ones) are logged to SQLite |
| Dry run | Test any action with dryRun: true before executing |
| Auto-cleanup | Action logs older than 30 days are pruned automatically |
Architecture
~/.mac-pilot/pilot.db (SQLite, WAL mode)
├── action_log — Every action executed, with timing + success/failure
├── action_log_fts — Full-text search index over action history
├── recipes — Saved automation sequences
├── recipes_fts — Full-text search index over recipes
├── app_knowledge — Per-app quirks, selectors, workarounds (auto-learned)
└── security_log — Blocked command audit trail
Built-in recipes are auto-loaded on first run. Your custom recipes and learned knowledge persist across sessions.
Comparison
| Feature | mac-pilot-mcp | Other MCP servers |
|---|---|---|
| Self-learning (auto-saves knowledge) | Yes | No |
| Reusable recipes with parameters | Yes | No |
| Built-in recipe library | 21 | 0 |
| JXA + AppleScript | Both | Usually one |
| Full-text search (recipes + history) | Yes | No |
| Security audit log | Yes | Rare |
| Risk classification (4 levels) | Yes | No |
| Dry run mode | Yes | Rare |
| Action log auto-cleanup | Yes | No |
Troubleshooting
<details> <summary><strong>"Accessibility access not allowed"</strong></summary>
Your terminal needs Accessibility permission:
- System Settings → Privacy & Security → Accessibility
- Toggle ON for your terminal
- Restart the terminal app completely </details>
<details> <summary><strong>"Application not found or not running"</strong></summary>
The target app must be running. Open it first:
mac_run { "actionType": "open", "target": "AppName" }
</details>
<details> <summary><strong>Screenshots are too large / slow</strong></summary>
Reduce the scale (default is 0.5):
mac_screenshot { "target": "screen", "scale": 0.3 }
</details>
<details> <summary><strong>Recipe not found</strong></summary>
Recipe names are case-sensitive. Search first:
mac_recipe_search { "query": "your keyword" }
</details>
<details> <summary><strong>Command blocked unexpectedly</strong></summary>
Use dry run to check the risk classification:
mac_run { "actionType": "shell", "command": "your-command", "dryRun": true }
</details>
Requirements
- macOS (darwin only)
- Node.js >= 18
- Accessibility permission for UI automation
Contributing
Issues and PRs welcome at github.com/leesgit/mac-pilot-mcp.
git clone https://github.com/leesgit/mac-pilot-mcp.git
cd mac-pilot-mcp
npm install
npm run build
npm test # 144 tests
License
MIT - Byeongchang Lee
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.