AI App through MCP
AI App uses the canonical interactive_lesson type.
Pass a focused request to kp_create_creation:
{ "type": "interactive_lesson", "prompt": "Build an interactive lesson that lets students explore how binary search narrows the range", "language": "en", "instructions": "Include a step control and a reset action"}Usually omit video-specific fields such as voice_id, duration_seconds, and scene_count. After completion, present the final launch URL from kp_get_creation.
When to choose AI App
Use AI App when the learner or user should interact with the concept rather than only watch it. Good requests name the user action, the learning or decision goal, the required controls, and the success state. A binary-search lesson, for example, should specify what a learner can change, what feedback they receive, and how they reset the simulation.
Keep the surface focused
interactive_lesson is not a video type. Avoid video-only fields and write the instructions field as product constraints: controls, steps, feedback, accessibility needs, or a reset action. Read kp_get_catalog first for the current supported surface, then create the result and poll its job ID just like every other MCP creation.
When the job succeeds, retrieve main_url from kp_get_creation and present that URL as the primary result. If the interaction needs a material change, create a revised result with clearer prompt and instruction text; do not assume scene-edit actions apply to an app.
Continue from here
Quickstartfor MCP setup and asynchronous jobsAvailable Toolsfor the supported tool contractVector Animationfor a non-interactive visual explanation