Magic Hour AI

What you can ask

• Create a 5-second vertical video. Here’s the scene:Copy

  Example path get_video_project · Create with magic hour skill

• Create a square product photo on a clean studio background. The product is:
• Animate my attached photo. The movement should be:

MCP connector

Server URL

https://mcp.magichour.ai/Copy

Login

OAuth

Protocol

MCP 2025-11-25 · Interactive UI

Instructions

Create and edit images and video, generate music, and retrieve existing audio projects. Tool calls require authentication. When exposed, use a matching estimate with the exact create body; resolve blocked/unavailable. For a complete request, default to one eligible result without a budget question. Respect explicit settings and limits; never invent a quote. Estimates never authorize spending. get_workflow_guide gives static inputs/docs/review, not account eligibility. After creation, use matching wait_for_*_project with the returned id. For *_file_path inputs, upload local bytes with get_upload_urls before using file_path. Direct public media URLs must be stable and fetchable as raw bytes; hotlinked URLs can fail. Preserve exact signed download URLs, including query parameters; never append expiration metadata. For native ChatGPT attachments, use upload_chatgpt_image for images or upload_chatgpt_media for images, video and audio within its supported formats and 20 MiB limit. Set media_type to the attachment's actual kind, not the type of generation requested. Keep each returned path paired with its input role; never pass a bare file ID, filename or chat-local path as a Magic Hour asset. Omit unrequested options only when defaults are eligible. Image Editor defaults to paid-tier nano-banana. On Free, use Image Editor model=flux-2-klein, resolution=640px, image_count=1 only if suitable. Balance does not unlock restricted settings. On tier restrictions, explain the block; never retry unchanged or blindly remove fields. Ask before changing explicit model/resolution requirements.

Choose video from inputs:

Approved image -> create_video_from_image directly; edit only for a requested visual change or a source problem that prevents the shot.

Still portrait + audio -> create_talking_photo; text alone is not audio.

Existing video + replacement audio -> create_lip_sync_video.

Preserve an explicitly requested alternative provider.

Choose models from the current schema to match the requested inputs and settings, with verified account eligibility.

Preserve an explicitly requested model; explain unsupported settings before proposing a change.

Show all

Tools

Skills

1 in the ChatGPT listing

• Create with magic hourUse Magic Hour MCP tools in ChatGPT to generate or edit images, videos, and audio, upload source media, monitor renders, and return finished results.