MCP AI Video Guide

AI Video through MCP

AI Video uses ai_video. MCP exposes video settings as flat kp_create_creation arguments rather than the OpenAPI video_generation object.

Read kp_get_catalog before choosing a model, ratio, resolution, or duration.

Gemini Omni Flash

{  "type": "ai_video",  "prompt": "Move slowly through a futuristic city on a rainy night, with neon reflections and natural ambient audio",  "aspect_ratio": "16:9",  "video_model": "gemini-omni-flash",  "video_task": "text_to_video",  "video_resolution": "720p",  "video_duration_seconds": 8}

After completion, Omni supports kp_edit_creation with action="iterate" and a new prompt. The server manages the interaction ID.

Veo 3.1 Fast or Standard

Use video_model="veo-3.1-fast" or video_model="veo-3.1-standard" with the same task, resolution, and duration fields. Veo does not support conversational iteration.

The current public MCP accepts text-to-video only. First frames, last frames, and attachments remain unavailable until the catalog says otherwise. Native audio stays enabled by the service.

When to choose AI Video

Use AI Video for a short visual moment, product atmosphere, cinematic B-roll, or a motion-first concept. Describe the subject, action, setting, camera behavior, and sound intent in one coherent prompt. It is not the best choice for a long instructional sequence with tightly controlled narration; use Explainer Video or Vector Animation for that kind of explanation.

Choose a model deliberately

Read kp_get_catalog immediately before creation. It tells you which model supports the requested task, ratio, resolution, duration, native audio, and iteration. Gemini Omni Flash can accept a later iterate edit when the completed detail exposes that action. Veo results do not support conversational iteration, so revise them by creating a new result with a changed prompt.

Keep the returned job ID, poll kp_get_job_status, and read the completed URLs and artifacts through kp_get_creation. Do not send first-frame, last-frame, or attachment fields while the public catalog marks them unavailable.

Continue from here

Quickstart for the shared asynchronous workflow
AI Image for a single generated visual
Explainer Video for narration-led teaching or product explanation