openclaw/Peekaboo

Peekaboo is a macOS CLI & optional MCP server that enables AI agents to capture screenshots of applications, or the entire system, with optional visual question answering through local or remote AI models.

What it solves

Peekaboo provides a way to automate macOS applications by allowing a system to "see" the screen and interact with the user interface. It bridges the gap between high-level intent (like "Open Safari and search for a project") and the low-level technical actions (clicks, typing, and accessibility inspection) required to control a Mac.

How it works

Peekaboo operates as a CLI and menu-bar application that uses macOS accessibility permissions and screen recording to inspect the UI. It can generate structured UI maps of running applications, take screenshots, and perform synthetic inputs like clicking, typing, and scrolling. It includes a built-in agent that can plan multi-step workflows using natural language, and it also supports the Model Context Protocol (MCP), allowing external AI clients like Claude Code or Cursor to use these same automation tools.

Who it’s for

It is designed for developers and power users who want to automate repetitive macOS tasks, as well as AI developers building agents that need native desktop control over Mac applications.

Highlights

  • Native UI Automation: Performs clicks, typing, and scrolling using accessibility inspection and coordinates.
  • AI Agent Integration: Includes a natural-language agent capable of planning and executing multi-step tasks.
  • MCP Support: Exposes its toolset to MCP-compatible clients for external AI orchestration.
  • Background Execution: Supports delivering input to apps in the background without requiring them to be frontmost in some cases.
  • Structured Inspection: Can output JSON maps of application UI elements for precise targeting.

Related

  • Project
  • Project
  • Project
  • Project
  • Project