Setup Guide
The PreTestAds MCP turns Claude into an ad-testing copilot. Connect it once, and from then on you can say "score my ad" in any chat — Claude uploads the file, waits for the neural model, reads back a 0–100 attention score benchmarked against 76 top-performing TikTok ads, and can even watch your video frame by frame to tell you why it scored that way. To see the whole thing in action, read the six-round iteration case study — a real session, unedited.
https://pretestads.com/mcpi want to score an adAn upload box appears right in the conversation. Drop in your video and Claude takes it from there.score_ad.You never call these directly — you just talk to Claude — but knowing what's under the hood helps you ask for the right things.
score_ad1 creditRuns a video, audio file, or text through the TRIBE neural model. Returns an analysis ID; scoring is async and typically takes 3–5 minutes for a video.
create_uploadfreeCreates the in-chat upload box for local files. After you drop a file in, scoring starts automatically.
get_ad_scorefreeThe full report for a run: score, label, verdict, hook strength, purchase signal, peak moments, and the second-by-second engagement curve.
get_ad_framesfreeStill frames from a scored video — one per second, or specific timestamps — so Claude can watch the ad and map the curve to what's actually on screen.
list_ad_scoresfreeYour recent scoring runs, newest first. Handy for comparing versions: "list my last five scores."
check_creditsfreeRemaining credits and plan on your account.
mark_ad_convertedfreeTag a scored ad as a real-world winner. This feedback sharpens the benchmark over time.
Score this ad, then watch it frame by frame and tell me exactly which second to fix.Score all three of these cuts and tell me which one to run.My ad scored 47 — re-cut it to fix the weak opening and score the new version.Compare my last two scores and chart both attention curves.Score this hook copy as text before I film it.The pattern that makes the MCP more than a scoreboard: score → watch → diagnose → edit → re-score. Claude can run the whole loop in one chat — and because the score is the referee, it will tell you when its own edit made things worse. (It happens. See the case study.)
The scoring loop gets sharper when Claude can also make footage. With a video-generation connector like the Higgsfield MCP in the same chat, Claude can extract a reference frame from your scored ad, generate a matching new shot, splice it in, and score the hybrid — a full make → test → fix loop where every iteration gets a number before it costs you media budget.
The MCP is for humans chatting with Claude on account credits. If you're building an autonomous agent, there are two other doors: the x402 pay-per-score API ($5 USDC per run, no account needed) and the open-source agent toolkit with ready-made clients for OpenAI, Gemini, LangChain, CrewAI, and more — plus a Claude Code plugin: /plugin marketplace add krecicki/pretestads-claude-plugin
Add the connector, drop in a video, and get a number before you spend a dollar running it.
Create Your Free AccountThen connect https://pretestads.com/mcp in Claude's Settings → Connectors.
In Claude (desktop app or claude.ai), go to Settings → Connectors → Add custom connector, name it PreTestAds, and paste https://pretestads.com/mcp. Click Connect and sign in with your PreTestAds account. Scoring runs bill the credits on that account.
Just ask Claude to score your ad. The MCP renders an upload box directly in the chat — drop your file in (MP4, MOV, or AVI up to 500MB) and scoring starts automatically. Claude polls for the result and reads the full report back to you, usually in about three to five minutes.
One account credit per scored ad — the same as scoring on the website. Checking status, listing past scores, and extracting frames from an already-scored video are free. If you'd rather pay per run without an account, agents can use the x402 API at $5 USDC per score instead.
Both. The get_ad_frames tool returns still frames from your scored video — one per second — so Claude can literally watch the ad and connect what's on screen to the exact seconds where attention rises or drops. That's what turns the score into specific, scene-level edit advice.