The predictions come from neural models trained on Tobii infrared gaze-fixation data, which measures fixations rather than viewability proxies — so the numbers will not line up with tools that estimate attention from viewport time. Generation and prediction sit behind the same connection, which is what lets a creative be scored, refined and scored again.
An MCP server for Viomba, bringing human attention prediction, attention-optimised creative generation and video attention analysis into an MCP client, with a Meta Marketing API bridge.
- Attention KPIs predicted for a display ad — seen percentage and view time — plus a predicted gaze heatmap
- An ad creative sliced into timed PNG screenshots for frame-level inspection
- Creative generation as a job you start, poll, fetch with full HTML, iterate on, list and cancel
- An attention-weighted site map generated across a set of domains
- Video analysed for attention, audio and video events, and an AVS sync score, with visual attention heatmaps predicted for video ads
- A one-time browser upload URL for sharing a local video file, and a poll that returns a token usable as videoUrl
- Meta credentials connected in the dashboard, then ad accounts and ad creatives listed and predictions run against the real ad library
- Usage and per-tool breakdowns for your own account, and request history
A Viomba account. Meta features need a Meta credential connected in the Viomba dashboard before the ad-library tools will return anything. Video work goes through the upload flow rather than a direct file path.
One command — npx -y mcp-remote https://mcp.viomba.com/mcp