Screenshots, sized for your AI coding assistant

Coding with an AI assistant in the terminal is a visual job more often than it looks. A stack trace in a dialog, a layout that renders wrong, a Figma frame you want built, a chart the model should read — the fastest way to explain any of them is a picture. But handing a picture to a command-line agent is fiddlier than it should be, and a raw 4K capture is the wrong picture to hand over.

Maxisnap solves the picture, not the plumbing. It does not auto-drive the agent or paste for you — it gives you a clean, correctly-sized image (or a short recording) in one keystroke, so the paste, drag, or file-path step that follows is quick and reliable. Below is exactly how each of the three main CLIs takes an image today, and where a purpose-built export earns its place.

Because a full-resolution screenshot is expensive and often counter-productive. Vision models tile an image into patches and charge tokens per patch, so a 4K capture can cost well over a thousand image tokens — while a copy resized to a sensible width usually costs a few hundred and reads just as clearly, because the model downsamples large images internally anyway. Oversized images are also the ones that hit per-image size limits or get silently rejected. Right-sizing the image up front is the single highest-leverage thing you can do before sending a screenshot to a coding agent.

Three keystrokes, no re-encoding, no second app:

• Capture it. Press Ctrl+Alt+5 and drag a box around the error dialog, the broken layout, or the design frame.
• Redact anything private. If the shot shows an API key, a token, or a customer's data, drag the blur tool over it before it ever reaches the model. Add an arrow or a numbered pin if you want to point the agent at one spot.
• Export the AI-optimized copy. One click produces a copy sized to a default max width of 1600 px (configurable in Settings). Save it, copy it, or drag it straight into the terminal — then paste, drop, or reference its path in Claude Code, Codex, or Gemini.

Because the export is already the right size, you skip the "why did my paste get rejected / cost so much" round-trip entirely. And because blur is applied destructively, the secret you covered can't be recovered from the file you hand over.

You can absolutely snip and paste manually — for the occasional screenshot it's fine. A dedicated export earns its keep once you're doing it all day, care about token cost, and want to redact before sending. Here's the honest trade-off, including against a file-path snipper like Invoke: