Give your agents eyes.

macOS automation that sees the screen and does the clicks.

Peekaboo brings high‑fidelity screen capture, AI analysis, and end‑to‑end GUI automation to macOS — as a native CLI and an MCP server.

Install Download npm

macOS 15+ Swift 6.2 Node 22+ (MCP) OpenAI · Claude · Grok · Gemini · Ollama

peekaboo see —app Safari

Window

Menu bar

Buttons

Text fields

snapshot_id: … elements: 142 clickable: yes

Install

Two ways: native app/CLI via Homebrew, or MCP server via npm. Both expose the same toolset.

brew install steipete/tap/peekaboo

Tap Quick start

Requires macOS Screen Recording + Accessibility permissions. Grant guide.

One-liner

peekaboo "Open Safari, go to github.com, and search for Peekaboo"

Tools you can trust: deterministic outputs, typed JSON, and composable automation.

Screens, windows, menu bar. Optional Retina 2× scaling. Fast enough for feedback loops.

CLI: peekaboo image, peekaboo see

Resolve UI elements and act on them: click, type, scroll, drag, hotkeys, menus, Spaces.

CLI: peekaboo click, peekaboo type, peekaboo menu

Natural language that chains tools — plus an MCP server so Claude Desktop/Cursor can drive it.

CLI: peekaboo agent, peekaboo mcp

A tight loop: capture → interpret → act. Repeat until the task is done.

01

See

Capture a screen/window and get a structured UI map with stable IDs.
02

Decide

Pick a target: a button label, a menu path, a specific window, a Space.
03

Do

Click, type, scroll, drag, or run a full agent plan — with receipts you can log.

Start here, then dive deep.