Peekaboo is an open-source macOS command-line tool and MCP server that gives AI agents eyes: it captures screenshots of applications, windows or the entire screen, and can answer visual questions about what it sees using local or remote AI models. It is part of the OpenClaw ecosystem and has earned more than 5,300 stars on GitHub, with active development through 2026.
Agents that drive a Mac need reliable perception, and Peekaboo is built precisely for that. As a CLI it is fully scriptable โ capture a specific app, a single window, a display or a region, in the format and at the scale you need โ and as an MCP server it plugs into agent frameworks so models can request screenshots and visual analysis as tools. Its visual question answering can run against local models for privacy or cloud vision models for capability, letting an agent read dialogs, verify UI state, or find the element it needs to act on. Peekaboo is commonly used as the perception layer for macOS automation agents, GUI testing and accessibility workflows, pairing naturally with tools that perform the clicks and keystrokes after Peekaboo reports what is on screen.
Peekaboo is completely free and open source under the MIT license. Cloud vision models, if you choose them, are billed by their providers.
Peekaboo is for developers building Mac-using AI agents, GUI test automation or accessibility tooling who need a dependable way for software to see and understand a macOS screen.