MCP server

extract_brandExtract a brand from a website

Reads a public website and saves its brand (colours, fonts, logo, tone) as a new brand kit in the workspace. Starts an asynchronous job and returns it: call wait_for_job with the job id; when it succeeds, result.brand_kit_id is the new kit to pass to create_video. Usually takes 20-60 seconds. Fails if the plan's brand kit limit is reached.

• urlrequired · The brand's website, e.g. https://example.com (https is added when missing).
• idempotency_key · Optional. Any unique string (a UUID works). Retrying the call with the same key and the same arguments returns the first result instead of starting (and paying for) the work again.

list_brand_kitsList brand kits

Lists the workspace's brand kits (default first) with their colours and fonts. Use an id as brand_kit_id in create_video.

list_storiesList story templates

Lists the built-in stories (video templates such as product launch, feature drop, stat sequence) with their editable fields and default values. Pass a story id as story_id to create_video: with fields the video is built instantly without AI; without fields the AI writes it in that story's structure.

create_videoCreate an on-brand video or token film

Creates a video (a film) on the workspace's brand, or on a token's brand when you pass token (a contract address or token link: "make a launch film for my token 0x…"). Two modes, exactly as the API: 1) AI generation: give a brief (and optionally story_id, duration_sec). The video starts in status "generating"; the answer includes the generation job: call wait_for_job with it (typically 1-3 minutes, call again if it is still running), then get_video. 2) Instant: give story_id plus fields (values for that story's field keys, see list_stories) and a brand kit. The video is ready immediately. Brand: brand_kit_id, or url (the brand is read from that site during generation), or token (the token's website and Dexscreener image), or none of them to use the default kit. Credits: an AI generation (a brief, or a story without fields) counts against the plan's monthly generations, exactly as in the app; past that quota one credit buys 4 more generations and is taken automatically. Building a story with fields uses no AI and costs nothing. Rendering MP4s is a separate step (render_video).

• brand_kit_id · Brand kit to use (from list_brand_kits). Omit it and url to use the workspace's default kit.
• url · A website to take the brand from when there is no brand kit yet. Ignored when brand_kit_id is given.
• token · For a token (crypto coin) film: its contract address (0x…, or a Solana address) or a link to it (Dexscreener, pump.fun, Torch, a block explorer). Its live market figures go into the brief and its image becomes the coin's face in the film. Alone, the brand comes from the token's website; it also works with brand_kit_id or url.
• brief · What the video should say: the news, offer or message, in plain words. Moderated before generation.
• story_id · A story id from list_stories. Omit it to let the AI pick the structure.
• fields · Field values for the story, keyed by the story's field keys (see list_stories): text, number, colour (#rrggbb), toggle, image ({url}) or list values. Needs story_id and a brand kit; with fields the video is built instantly, without AI.
• format · Main format: 16:9, 9:16, 1:1 or 4:5 (default 16:9). Every format can be rendered later with render_video.
• duration_sec · Target length in seconds (6 to 90). Default: the story's length, or 20.
• locale · Language of the on-screen text, e.g. en, de, pt-BR. Default en.
• title · Title shown in the video library.
• idempotency_key · Optional. Any unique string (a UUID works). Retrying the call with the same key and the same arguments returns the first result instead of starting (and paying for) the work again.

get_jobGet a job

Status, stage, progress (0-1) and result of an asynchronous job (brand extraction or video generation). Prefer wait_for_job, which waits server-side instead of polling.

• job_idrequired · The job id (a UUID).

wait_for_jobWait for a job

Waits server-side (up to timeout_sec, at most 25 s) until a job succeeds, fails or is canceled, then returns it like get_job. If it is still running when the time is up, call wait_for_job again; do not poll in a tight loop.

• job_idrequired · The job id (a UUID).
• timeout_sec · How long to wait, in seconds (1 to 25, default 20).

get_videoGet a video

A video's status, title, scenes (id, label, duration, fields) and field values of its current version, the running generation job if any, and editor_url (the player and editor in the loudscene app, where share links are made).

• video_idrequired · The video id (a UUID).
• include_doc · Also return the full VideoDoc (composition code included, often 20-200 KB). Default false.

update_videoChange a video's text and values

Changes field values (headline, prices, dates, colours, images: whatever get_video lists under fields) and/or the title of a video, without AI and at no cost. Field changes are saved as a new version, which becomes the current one; the next render_video renders it (a new version is a new render charge). Read get_video first to see the field keys and their current values.

• video_idrequired · The video id (a UUID).
• fields · New values keyed by the video's field keys (see get_video): text, number, colour (#rrggbb), toggle, image ({url}) or list values. Only the fields you send change.
• title · A new title for the video library.
• note · Shown in the version history, e.g. what changed and why.

create_stillsPreview frames as PNGs

Captures still frames (PNG) of the video's current version at the given moments, to check the look before rendering. Starts an asynchronous job: call wait_for_job with its id; when it succeeds, result.stills lists { t, url } with signed PNG URLs. Free, no credits; a few seconds to a minute. The same request within ten minutes returns the same job.

• video_idrequired · The video id (a UUID).
• timesrequired · Moments to capture, in seconds into the video (1 to 24 of them), e.g. [1, 5, 12].
• format · Aspect ratio of the frames: 16:9, 9:16, 1:1 or 4:5. Default: the video's main format.

render_videoRender MP4s

Renders the video's current version as MP4 in one or more formats (16:9, 9:16, 1:1, 4:5) and returns one render per format. Renders take about a minute each: call wait_for_render (or get_render) with each render id to get the download URL. An identical finished render is returned instead of rendering again.

• video_idrequired · The video id (a UUID).
• formatsrequired · Formats to render: any of 16:9, 9:16, 1:1, 4:5.
• resolution · Short side in pixels: 720, 1080 (default) or 2160 (4K, Pro plan and up).
• idempotency_key · Optional. Any unique string (a UUID works). Retrying the call with the same key and the same arguments returns the first result instead of starting (and paying for) the work again.

get_renderGet a render

A render's status and progress (0-1); once succeeded, url is a signed MP4 download URL and poster_url a still image, both valid until url_expires_at (4 hours); call again for fresh ones.

• render_idrequired · The render id (a UUID).

wait_for_renderWait for a render

Waits server-side (up to timeout_sec, at most 25 s) until a render succeeds or fails, then returns it like get_render, with the MP4 URL when ready. If it is still running, call it again.

• render_idrequired · The render id (a UUID).
• timeout_sec · How long to wait, in seconds (1 to 25, default 20).