Annotate

What is Annotate?

Annotate is a free, local-only macOS app that replaces written prompts for AI coding agents with a screen recording you narrate and annotate. Instead of typing a long brief, you record your screen, draw directly on the UI to point at the exact element, and speak the change you want. The session is then handed off to Cursor, Claude Code, or Codex, which reads the recording via MCP — frames and speech, not a giant video dump. It runs entirely on Apple Silicon, requires no login, and nothing ever leaves your device.

Who it's for

• Cursor users who want to skip drafting written prompts and show the agent exactly what to change on screen.
• Claude Code and Codex users who need a faster, more precise way to communicate UI fixes or multi-step changes without screenshots.
• Developers working across multiple displays who want to capture an app on one monitor and an editor on another in a single take, with voice and drawing synced to the timeline.

Key features

Screen recording with annotation tools

Capture one window or every display in a single session. While recording, you can draw with a pen, rectangle, or text tool directly over the UI — so the agent knows which button you mean, not which one you vaguely gestured at.

Voice-to-text synced with frames

Your speech is transcribed and aligned with what was on screen at the moment you said it. If you say “make this button match the sidebar colour,” the agent sees the exact frame where you pointed and hears the words in context — no guesswork.

MCP-based handoff to your agent

The session opens directly in Cursor, Claude Code, or Codex. The agent reads the recording through MCP, which means it processes frames and speech transcripts rather than a massive video file. This keeps token usage low and intent clear.

Local-only processing

Everything — frames, transcripts, and annotations — stays on your Mac. There’s no upload, no cloud dependency, and no account required. It’s free for macOS Apple Silicon and designed to respect your privacy by design.

What stands out

A screenshot can’t show what changed — Annotate keeps the motion so your agent sees the full story.

Static screenshots lose clicks, timing, and voice. Annotate preserves all three, syncing frames, gestures, and speech into a coherent walkthrough. That means the agent follows what you meant, not what one frozen frame happened to show. It’s a fundamentally different input method for AI coding agents — one that removes the friction of writing precise prompts and replaces it with showing, pointing, and speaking.