Show your AI coding agent the bug — not a 4K movie of it
Some bugs only exist in motion: a flicker, a race condition, a layout that breaks halfway through an animation. Maxisnap records a region or the full screen on Windows and exports a one-click AI-optimized MP4 — short, right-sized, and ready to drop into Claude Code, Codex CLI, or Gemini CLI without uploading a needlessly huge file.
Windows 10 (1903+) / 11 · free tier · macOS not yet available
A screenshot freezes a moment; a recording shows the sequence. When you are pairing with an AI coding assistant in the terminal, some problems are impossible to describe in a still: a hover state that jumps, a form that submits twice, a video player that stutters for two seconds and then recovers, a scroll that jitters on one monitor but not the other. For those, a five-second clip explains more than three paragraphs of prose — and far more than a single frame.
The catch is that a raw screen recording is the wrong file to hand to a model. A minute of full-resolution capture can be tens of megabytes, and the tools that accept video are picky about size and slow to ingest a large upload. Maxisnap fixes the file, not the plumbing: it records what you point it at, lets you trim to the moment that matters, and exports a compact copy engineered for AI tools — then hands you that file to copy, drag, or reference by path.
When should I send a recording instead of a screenshot?
Send a screenshot when the problem is visible in one frame — a stack trace, a misaligned element, a wrong colour. Reach for a recording when the bug is about change over time: something that only appears while an animation runs, an interaction that has to happen in a certain order, or a timing issue you cannot reproduce with a single click. A recording also carries the exact reproduction steps for free, because the agent watches you perform them.
| The problem | Best format | Why |
|---|---|---|
| A stack trace or error dialog | Screenshot | One frame holds all the text the model needs |
| A layout that renders wrong | Screenshot | Static; a still is cheaper and just as clear |
| An animation or transition glitch | Recording | The fault only exists across frames |
| A hard-to-reproduce interaction bug | Recording | The clip is the reproduction steps |
| Stutter, lag, or a race condition | Recording | Timing is the whole point |
Why not just upload the raw recording?
Because a raw capture is far bigger than it needs to be. Full-resolution video at a high frame rate carries detail a model does not use — it reads a UI clearly at a modest width and a smooth-enough frame rate. An oversized clip is slower to upload, more likely to hit a per-file size limit, and, on tools that sample video into frames, no more informative than a right-sized one. Trimming the clip to the moment that matters and encoding it for legibility rather than fidelity is the single highest-leverage thing you can do before sending video to an AI tool.
Maxisnap's AI-optimized MP4 is that right-sized copy: roughly 1280 px wide, 24 fps, with mono audio, written as aioptimized_*.mp4. There is also an ultra-light option (a few frames per second, no audio) for when a tool accepts video but you want the smallest possible file.
| What you send | Typical size for a short clip | Result in the tool |
|---|---|---|
| Raw full-screen recording | Tens of MB | Slow upload, may exceed a size limit |
| Maxisnap AI-optimized MP4 | A fraction of the size | Uploads fast, stays under limits, reads clearly |
Sizes depend on clip length, motion, and content; the point is the ratio, not a fixed figure. Maxisnap does not automate any AI tool — it only produces the file.
How do I record a bug and hand it to Claude Code?
Four steps, one app, no re-encoding in a second tool:
- Record it. Press
Ctrl+Alt+3to record a region orCtrl+Alt+4for the full screen (the monitor under your cursor). Reproduce the bug, then press the hotkey again or click Stop. Both hotkeys are remappable. - Trim the dead air. The result window opens with an inline player and start/end trim handles — cut to the two seconds that actually show the fault so the model isn't watching you find it.
- Export the AI-optimized copy. Choose the AI-optimized format and Maxisnap writes a compact
aioptimized_*.mp4. In v2.4.9.2 an unchanged clip is delivered instantly and converted outputs are cached, so the file is ready the moment you ask for it. - Hand it over. Use the result window's Copy or Drag to move the file into your terminal, or Save it and reference the path in your prompt.
Because the export is already small and the right shape, you skip the "why is my upload taking so long / getting rejected" round-trip entirely.
Which AI coding tools accept a video, and how?
Support for video in terminal coding agents is newer and less uniform than support for images, and it changes quickly — so the most portable habit is to keep the file on disk and reference it, rather than relying on clipboard paste. Here is the current lay of the land; check each tool's own docs for today's exact behaviour.
| Tool | How to give it a file | Note |
|---|---|---|
| Claude Code | Reference the saved file by path in your prompt, or drag the file into the terminal. | A file path is the most reliable route on Windows and WSL, where clipboard paste can silently fail. |
| Codex CLI | Drag the file in, or pass it explicitly with the image/file flag your version supports. | Passing an explicit path avoids depending on a flaky clipboard. |
| Gemini CLI | Reference the file by path, or drag it into the prompt. | Multimodal support varies by version; the saved-file route works regardless. |
| ChatGPT / Claude / Gemini apps | Attach the aioptimized_*.mp4 in the chat's upload control. | The small file uploads fast and is less likely to hit a size limit. |
Whether a given assistant can read video, and how, is set by that tool and changes often. Maxisnap's job is to produce a small, standard MP4 you can hand over by whichever route works.
$ diff raw-capture ai-optimized.mp4
[RAW] full-screen, full-res, full frame rate -- tens of MB, slow upload, may exceed the limit [MAXISNAP] trim + AI-optimized export ++ ~1280px, 24fps, mono, a fraction of the size [RAW] the whole two-minute session -- the model watches you hunt for the bug [MAXISNAP] start/end trim to the fault ++ five seconds that show exactly the problem [RAW] re-encode in a second tool, find the file -- extra steps between you and the answer [MAXISNAP] Copy / Drag / Save from the result window ++ file in the terminal in one move
Can the recording include audio?
Yes. Maxisnap records system audio and your microphone together by default, driver-free, so you can narrate what you're doing while you reproduce a bug — “watch the counter here, it skips from 3 to 5” — which gives the model your intent as well as the visuals. The AI-optimized export keeps a single mono track to stay small, and the ultra-light option drops audio entirely when you only need the picture. If you'd rather send a silent clip, that's a setting, not a workaround.
Screen recording for AI coding — FAQ
Can I share a screen recording with Claude Code, Codex CLI, or Gemini CLI?
Yes. Maxisnap records a region or the full screen on Windows and exports a compact, AI-optimized MP4 that you hand to the tool by the route it supports — dragging the file in, or referencing its saved path in your prompt. A file path is the most reliable method on Windows and WSL. Maxisnap produces the file; it does not automate the assistant.
What are the AI-optimized video settings?
The AI-optimized MP4 is approximately 1280 px wide at 24 fps with mono audio, saved as aioptimized_*.mp4. An ultra-light option records at a few frames per second with no audio for the smallest possible file. You can also export standard MP4, GIF, or WebM.
How do I record only part of the screen?
Press Ctrl+Alt+3 and drag a box around the region you want; Ctrl+Alt+4 records the full screen under your cursor instead. Both hotkeys are remappable in Settings. After recording, the result window lets you trim the start and end before you export.
Why is my raw screen recording too big to upload to an AI tool?
Full-resolution video at a high frame rate is much larger than a model needs. Trimming to the relevant moment and exporting the AI-optimized MP4 produces a far smaller file that uploads quickly and is less likely to hit a per-file size limit, while staying clear enough for the tool to read.
Does it run on Windows?
Yes — Maxisnap is Windows-native (Windows 10 1903+ and Windows 11). A macOS build is in development with no public download yet. It is a 113 MB installer and the recorder, screenshot tool, annotation, and OCR are all in the one app.
Record the bug. Trim it. Hand it over.
Capture a region or the full screen, export an AI-optimized MP4, and drop it into your coding agent — free on Windows.
Related: screenshots for AI coding · paste a screenshot into Claude Code · screen recorder + screenshot tool · record a bug report