Peekaboo MCP Server

Content

Peekaboo is a macOS automation CLI and MCP server for agents that need to see and operate desktop applications. It can capture screens or windows, inspect UI elements, click by label or element ID, type text, send hotkeys, manage windows and Spaces, interact with menus and dialogs, and run optional visual analysis through configured model providers.

The npm package exposes the MCP wrapper as @steipete/peekaboo, while the Homebrew path installs the native macOS app and CLI.

Source Review

• https://github.com/openclaw/Peekaboo
• https://github.com/openclaw/Peekaboo/blob/main/README.md
• https://github.com/openclaw/Peekaboo/blob/main/docs/commands/mcp.md
• https://github.com/openclaw/Peekaboo/blob/main/docs/permissions.md
• https://github.com/openclaw/Peekaboo/blob/main/package.json

Verified on 2026-06-18. The README documents the MCP server for Codex, Claude Code, and Cursor, plus the npx -y @steipete/peekaboo startup path. The MCP command docs identify stdio as the default supported transport and note that HTTP/SSE transports are stubbed.

Features

• macOS screen, window, and menu-bar capture with optional Retina scaling.
• UI inspection through see, snapshots, element IDs, and labels.
• Click, type, press, hotkey, paste, scroll, swipe, drag, move, menu, menubar, dock, app, window, space, and dialog automation.
• Direct accessibility actions such as set-value and perform-action.
• Natural-language automation through the agent command.
• MCP server and CLI backed by the same tool catalog.
• Optional model-provider integration through configured AI providers.
• Permission tools for Screen Recording, Accessibility, and Event Synthesizing.

Installation

Run as an MCP server through npm:

{
  "mcpServers": {
    "peekaboo": {
      "command": "npx",
      "args": ["-y", "@steipete/peekaboo"]
    }
  }
}

For the full native CLI/app path:

brew install steipete/tap/peekaboo

Then verify permissions:

peekaboo permissions status

Use Cases

• Let a coding agent inspect a macOS app window while debugging UI state.
• Capture screenshots and UI element maps for local app QA.
• Automate repeatable desktop workflows in a dedicated test account.
• Run browser or app interactions when a browser-only MCP is not enough.
• Use local or hosted visual models to answer questions about a captured screen.

Safety and Privacy

Peekaboo operates at the desktop boundary, so it should be treated as sensitive automation. Start read-only, grant the narrowest practical macOS permissions, and use explicit human approval before any agent action that changes data, sends communication, submits a form, or manipulates a logged-in app.

When optional AI analysis is enabled, confirm whether screenshots or extracted UI context are sent to a hosted model provider. Use local models or isolated test environments for confidential screens.

Duplicate Check

Checked current content/mcp/, content/tools/, content/skills/, open pull requests, and repository-wide content for openclaw/Peekaboo, Peekaboo MCP, @steipete/peekaboo, macOS automation MCP, screenshot MCP server, and OpenClaw Peekaboo. Existing browser and desktop automation entries cover BrowserMCP, Chrome MCP, Playwright MCP, Windows MCP, and bb-browser, but no dedicated Peekaboo MCP entry, exact source URL duplicate, target file, or open duplicate PR was found.