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.
Last verified:
What is Peekaboo?
Peekaboo is a macOS automation tool that gives AI agents the ability to see the screen and perform GUI interactions. It combines high-fidelity screen capture with AI analysis to enable end-to-end graphical user interface automation on macOS 15+. The tool works as both a native command-line interface and an MCP (Model Context Protocol) server, allowing AI assistants like Claude Desktop, Cursor, and Codex to drive desktop applications using the same primitives humans use.
Key features include pixel-accurate screen capture with optional Retina 2× scaling, click-by-intent automation that resolves UI elements and performs actions like clicking, typing, scrolling, dragging, hotkeys, and menu navigation, and an agent-native natural language interface that chains tools together. The tool captures screens/windows/menu bar, creates structured UI maps with stable IDs, and provides receipts for logging automation actions.
Peekaboo is designed for developers building AI agents, automation engineers creating GUI workflows, and anyone who wants to automate macOS applications using natural language commands. It supports multiple AI providers including OpenAI, Claude, Grok, Gemini, and Ollama, making it flexible for different AI infrastructure setups.
The tool requires macOS Screen Recording and Accessibility permissions to function, operates in a tight capture-interpret-act loop until tasks are complete, and provides deterministic outputs with typed JSON for composable automation workflows.
Peekaboo pricing
Pricing model: Freemium
The website does not display explicit pricing information. Peekaboo is installed via Homebrew (brew install steipete/tap/peekaboo) or npm for the MCP server (npx @steipete/peekaboo mcp), suggesting it may be free/open-source. The tool is available as a native CLI app and MCP server with no mentioned paid tiers on the official website.
Peekaboo pros
- Pixel-accurate screen capture with Retina 2× scaling option
- Natural language commands for GUI automation
- Works as both CLI and MCP server
- Supports multiple AI providers (OpenAI, Claude, Grok, Gemini, Ollama)
- Click-by-intent resolves UI elements automatically
- Handles clicks, typing, scrolling, dragging, hotkeys, menus, and Spaces
- Structured UI maps with stable IDs for reliable automation
- Deterministic outputs with typed JSON
- Composable automation that chains tools together
- Natively built for macOS using Swift 6.2
- Easy Homebrew installation
- MCP server works with Claude Desktop and Cursor
- Visualizer feedback for agent sessions
- Resumable agent sessions
- Fast enough for feedback loops
- Menu-bar visualizer in Mac app
- Receipts for logging automation actions
Peekaboo cons
- Only supports macOS 15 and later
- Requires Screen Recording permission
- Requires Accessibility permission
- No Windows support yet (only experimental community port)
- Node 22+ required for MCP server
- Swift 6.2 required for native usage
- Homebrew tap installation required
- Limited to macOS ecosystem only
- May struggle with highly custom/drawn UI elements
Frequently asked questions about Peekaboo
What is Peekaboo?
Peekaboo is a macOS automation toolkit that sees the screen and does clicks. It brings high-fidelity screen capture, AI analysis, and end-to-end GUI automation to macOS as a native CLI and MCP server, giving AI agents eyes to interact with desktop applications.
How do I install Peekaboo?
You can install Peekaboo two ways: via Homebrew for the native app/CLI (brew install steipete/tap/peekaboo) or via npm for the MCP server. Both expose the same toolset.
What permissions does Peekaboo require?
Peekaboo requires macOS Screen Recording and Accessibility permissions to function. You need to grant these permissions after installation for the tool to capture the screen and interact with UI elements.
Which AI providers does Peekaboo support?
Peekaboo supports OpenAI, Claude, Grok, Gemini, and Ollama as AI providers for its agent functionality.
What macOS version is required?
Peekaboo requires macOS 15 or later to run.
How does Peekaboo work?
Peekaboo operates in a tight loop: See (capture screen/window and get structured UI map with stable IDs), Decide (pick a target like button label, menu path, window, or Space), and Do (click, type, scroll, drag, or run full agent plan with loggable receipts).
What CLI commands does Peekaboo offer?
Key CLI commands include: peekaboo image/see (capture), peekaboo click/type/menu (automation actions), peekaboo agent (natural language agent), and peekaboo mcp (MCP server).
Can I use Peekaboo with Claude Desktop?
Yes, Peekaboo ships with an MCP server so Claude Desktop and Cursor can drive it using the same primitives you'd use from the shell.
What surfaces/interfaces does Peekaboo provide?
Peekaboo provides four surfaces: CLI for scripts/ad-hoc captures/CI, MCP server for Codex/Claude Code/Cursor, Mac app for menu-bar visualizer and permission prompts, and Library (Swift Package) for embedding in Swift apps.
Is Peekaboo open source?
The source code is available at github.com/openclaw/Peekaboo, and issues can be filed at github.com/openclaw/Peekaboo/issues. The author is @steipete.