Submit Log in

Peekaboo

by OpenClaw

v3.9.8

MIT

Repository npm

Developer Tools

macoscomputer-usescreenshotsguiautomation

Summary

macOS screen capture and GUI automation over MCP — see the screen, then click, type and navigate it, with local or hosted vision models.

Features

see: capture and annotate the screen before acting on it
Full input and navigation set: click, type, hotkey, scroll, drag, window, space, dock
Background input delivery that does not steal focus
Local vision via Ollama or LM Studio, or hosted providers
agent and run tools for multi-step and scripted workflows

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

Open your agent's MCP configuration file.
Merge the snippet below into it.
Restart the agent — the "Peekaboo" tools will be available.

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

Claude Code

Add the server with a single CLI command.

claude mcp add peekaboo -- npx -y @steipete/peekaboo

Claude Desktop

Add the server to claude_desktop_config.json (Settings → Developer → Edit Config).

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

Description

Peekaboo is the macOS counterpart to the desktop-automation servers that exist for Windows: a CLI and MCP server that lets an agent take a real look at the screen and then act on it. The see tool captures and annotates the UI so the model knows what it is pointing at; everything after that is ordinary interaction — click, type, set-value, press, hotkey, paste.

Navigation and discovery are covered properly rather than as an afterthought: scroll, swipe, drag, move, plus window, space, app and dock for arranging the environment, and menu, menubar, dialog, screen and list for finding what is available without guessing coordinates. An agent tool takes a natural-language instruction and runs the multi-step flow itself, and run executes saved workflow scripts.

A detail that matters in practice: input can be delivered to a target process in the background, so automation does not have to steal focus from whatever you are doing. Vision can come from hosted providers (OpenAI, Anthropic, xAI, Google, MiniMax, Kimi) or from a local model through Ollama or LM Studio, so screenshots need not leave the machine.

Requires macOS 15 Sequoia or later, with Screen Recording and Accessibility permissions granted; the npm MCP server needs Node 22+. Published as @steipete/peekaboo, MIT licensed.