AdRetina™ · The vision layer of the PreTestAds MCP

Your AI can finally watch your ad.

Claude can't open a video file — no AI chatbot can. AdRetina™ streams your scored ad into Claude's vision one frame per second, pinned to the attention curve, so "your score dropped at 0:07" becomes "the logo card covering her face at 0:07 is killing you — cut it."

Add the MCP to Claude →See a live sample report first

Free on every scored ad · No credits · Only on the PreTestAds MCP

One still per second, timestamped, mapped against your AdCortex™ attention curve.

Every AI chat is blind to your video

Attach an mp4 to Claude, ChatGPT, or Gemini's chat and ask what's wrong with it. Most can't open it at all; none can tie what they see to a measured attention signal. You end up describing your own ad to an AI that's guessing.

How AdRetina™ works

AdCortex™ scores the ad

Upload from the chat (or the website). The fMRI-trained model grades it against 76 top TikTok ads and produces the second-by-second attention curve. One credit.

You say “watch it”

AdRetina decodes the stored video server-side and streams stills into Claude's vision — the full ad at one frame per second, or just the seconds you ask about. Free, instant, no credits.

Claude coaches scene by scene

With frames and curve in the same context, Claude points at the exact shot behind every dip and peak: what to cut, what to move, which frame is your thumbnail, what to reshoot.

You: Watch my ad frame by frame and tell me exactly which second to fix.

Claude: I'm looking at all 34 frames now. Your dip at 0:07 lines up with
the cut to the static logo card — attention falls 41% the moment her face
leaves the screen. Move the logo to the corner and hold the demo shot.
Your peak at 0:14 (the before/after reveal) is your thumbnail.

Why only PreTestAds has this

Not a Claude feature

Video input doesn't exist in Claude, ChatGPT, or Gemini chat. AdRetina is our own pipeline — storage, decoding, frame budgeting, timestamping — exposed through our MCP server. Remove the connector and the capability disappears.

Sized to actually fit

Naively dumping frames blows past an AI's context window. AdRetina budgets every still (about 15KB each) so a full 60-second ad fits comfortably in one conversation — with room left for the report, the curve, and the coaching.

Pinned to a real signal

Frames alone are a slideshow. AdRetina's stills carry timestamps that line up one-to-one with the AdCortex™ per-second attention curve — so the AI's advice is anchored to measured engagement, not vibes.

Built for the fix loop

Watch → diagnose → re-cut → rescore. Pair the MCP with a generation connector like Higgsfield and Claude can pull a reference frame, generate replacement footage, splice it, and score the new version — in one chat.

AdCortex™ is the brain. AdRetina™ is the eyes. Together they're the only ad-testing stack an AI agent can both read and see.

Setup

Connect the MCP in 3 steps

Paste one URL into Claude's connectors. No API key.

Case study

Six re-cuts, one surprise winner

Claude watched, re-cut, and rescored in one chat.

Guide

How to read an attention curve

The signal AdRetina pins every frame to.

Give your AI eyes on your ad

Create a free account, connect the MCP, score an ad, then just say "watch it."

Create Your Free Account

Then connect https://pretestads.com/mcp in Claude's Settings → Connectors.

AdRetina™ and AdCortex™ are trademarks of PreTestAds.

Frequently asked questions

What is AdRetina?

AdRetina™ is the vision layer of the PreTestAds MCP. After AdCortex™ scores your video ad, AdRetina extracts still frames server-side — one per second for the whole ad, or the exact seconds you ask about — and streams them into Claude's vision as images. Claude then connects what's on screen to the second-by-second attention curve and gives scene-level edit advice.

Can't Claude just watch a video I upload to the chat?

No. Claude (like every major AI chatbot) has image vision but no native video input — attach an mp4 to a chat and it can't open it. AdRetina is what bridges that gap: it turns your scored video into an ordered, timestamped frame sequence that Claude genuinely sees, which is the closest thing to watching that current AI models can do.

How does AdRetina work technically?

Your scored video already lives on our servers. When you say 'watch it,' our backend decodes it with ffmpeg at one frame per second (up to 60 frames), downscales each still so a full 60-second ad fits comfortably in the conversation, and returns them to Claude as MCP image content blocks — the protocol's format for 'here are pictures, look at them.' Each frame carries its timestamp, so Claude can line the footage up against the attention curve.

What does AdRetina cost?

Nothing. Watching a scored ad is free and burns no credits — you only pay the one credit it took to score the ad in the first place. You can ask Claude to re-watch, zoom into specific seconds, or compare versions as many times as you like.

Is AdRetina unique to PreTestAds?

Yes. AdRetina is our own engineering, exposed only through the PreTestAds MCP server — it isn't a Claude feature, and no mainstream connector we know of turns a video into a watchable, curve-pinned frame sequence. Without it, an AI chat can only read numbers about your ad; with it, the AI has eyes on the footage.

Does AdRetina work outside Claude?

Any MCP-compatible client whose model has image vision can use it — the frames arrive as standard MCP image content. In practice Claude (claude.ai, the desktop app, and Claude Code) is where we test and support it. Connect https://pretestads.com/mcp as a custom connector and say 'watch my ad.'

By Chris Krecicki · Published July 19, 2026