Give your AI eyes — without handing over your screen
Screenshots are a leak vector: your screen holds API keys, passwords, and private chats. Shotcrisp lets agents capture through it, blacking out sensitive text before the model ever sees a pixel.
The problem with AI and screenshots
The moment you hand a raw screenshot to an AI, everything visible goes with it — the bank tab behind your editor, the API key in your terminal, the private thread in Slack. Most "let AI see my screen" tools capture the pixels and ship them, no questions asked.
Redacted before it is shared
Shotcrisp scans each capture and covers known secret patterns — emails, API keys, passwords, credit-card numbers — before the image or text leaves your Mac. It's not optional and it's not a flag: agent captures are redacted by default. Detection is best-effort, so keep truly sensitive apps on the never-capture list, but the common footguns are handled for you.
You approve every client
Access is off until you turn it on, and each AI client is approved once — verified by its code signature so it can't impersonate a trusted app. Revoke any client from Settings, and check the activity log to see exactly what was captured and when.
Your private apps stay invisible
Add 1Password, Signal, Mail, or anything else to the never-capture list and its windows can't appear in any capture — human or agent — even open in the background.
Shotcrisp is a native macOS app; AI screen access works through its built-in MCP server with Claude Code, Claude Desktop, and other MCP clients.