MCP Server - Hooked API Documentation

Overview

Hooked’s MCP server lets AI apps (Claude, ChatGPT, Claude Code, Cursor, Windsurf, VS Code and any other client of the Model Context Protocol) make videos, images, clips and voices for your team, straight from the conversation. It is one URL:

https://api.hooked.so/v1/mcp

Nothing to install and no API key: the first time, the app opens a Hooked sign-in page, you pick the team, and it connects with OAuth. Everything it makes uses that team’s credits, and you can disconnect it at any time in Settings → Connected apps. Reconnecting an app replaces its previous connection, so you won’t end up with duplicates.

The quickest way is the Connect buttons on hooked.so/mcp or in Settings → Connected apps: they open each app with Hooked already filled in, or copy its config.

Connect

In Claude (web or desktop), open Settings → Connectors → Add custom connector.
Name it Hooked and paste https://api.hooked.so/v1/mcp.
Click Connect and sign in with your Hooked account.

In ChatGPT, open Settings → Apps → Advanced settings and turn on Developer mode (Plus, Pro, Business, Enterprise and Edu plans).
Create an app named Hooked, paste https://api.hooked.so/v1/mcp and choose OAuth.
Sign in with your Hooked account, then pick Hooked from the tools menu in any chat.

claude mcp add --transport http hooked https://api.hooked.so/v1/mcp

Then run /mcp in Claude Code and choose Authenticate.

Add this to ~/.cursor/mcp.json (the Cursor CLI reads the same file):

{
  "mcpServers": {
    "hooked": { "url": "https://api.hooked.so/v1/mcp" }
  }
}

Add this to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "hooked": { "serverUrl": "https://api.hooked.so/v1/mcp" }
  }
}

Add this to .vscode/mcp.json:

{
  "servers": {
    "hooked": { "type": "http", "url": "https://api.hooked.so/v1/mcp" }
  }
}

With an API key instead

For scripts and agents that cannot open a browser, send an API key in the x-api-key header (or as Authorization: Bearer):

claude mcp add --transport http hooked https://api.hooked.so/v1/mcp --header "x-api-key: YOUR_API_KEY"

Approving tool calls

Clients ask before a tool spends credits or changes something, and let you say “always allow” per tool. The right setting is the same everywhere and follows what the tool does:

Tools	Setting
Reads: `list_`, `get_`, `search`, `fetch`, `search_formats`	Always allow
Writes: `create_`, `generate_`, `edit_project`, `render_project`, `import_media`, `schedule_post`…	Ask
Irreversible: `delete_*`, `update_character`, `cancel_scheduled_post`	Ask, and leave “always allow” off
Research that costs credits: `list_niches`, `get_niche_*`, `search_social_videos`, `analyze_social_account`	Ask

The paid research tools are not marked read-only on purpose, so the client asks before they spend. What a saved search reads back (list_social_searches, get_social_search) is free and counts as a read.

Clients that understand the read-only hint stop asking for the reads on their own. Where you set it:

Claude (web and desktop): the first call offers Allow once / Always allow / Deny. Allow the reads; keep the writes and the delete_* tools asking.
Claude Code: reads need no approval; pin the rest with /permissions (or the permissions block in settings.json, entries mcp__hooked__<tool>).
Cursor: Settings → MCP lists Hooked’s tools; allow the reads and keep the rest on ask. Reads may not prompt at all.
Windsurf: Cascade → MCP, the per-tool allow list; same rule.
VS Code: the chat asks per tool (Copilot Chat tool settings); auto-approve only the reads.
ChatGPT: there is no per-tool prompt — adding the app is the approval. Hooked still confirms the important steps in the conversation: before spending on something large, before publishing, and before deleting.

What it can do

The server has 76 tools. You never call them yourself: you ask in plain words and the AI picks them. It also hands the AI instructions on how to work with you: it replies in your language, asks before spending credits on something large, quotes the price first when you ask, and waits for the result instead of asking you to check.

In-chat previews. In apps that show MCP Apps (Claude, ChatGPT), the video or image appears in the chat with a progress bar while it is made, avatars and voices show as a picker, and a file from your device goes up through a picker too.
Formats. The dashboard’s ~90 formats (Lego look, talking fruit, parkour gameplay, anime, Reddit stories…) are found with search_formats: “make it in a Lego style” just works.
Slash commands. /ugc, /slideshow, /brainrot, /faceless, /avatar, /cinematic, /music-video, /clone and /style start the main kinds of video.
Your files. Images and clips you give it go into your media library first: pass a public https link (works in any client), attach the file to the chat in ChatGPT, or — in an app that shows the widgets — ask to upload and pick a file from your device. From there they can be used in any video.
Research first. It can look at what is trending in a niche, search TikTok and YouTube, and analyze an account before writing your script or ad.
Prices first. The create tools for script, prompt, talking avatar, UGC, slideshow, clone, Reddit story, quiz and podcast take dryRun: true, which answers the exact credits without creating anything.

Video Creation

Tool	Description
`create_script_to_video`	A voice reads your script over AI images, AI video clips, your library media, gameplay or motion graphics
`create_talking_avatar`	An AI avatar speaks your script to camera, in its own voice
`create_prompt_to_video`	AI writes the script from your prompt, then a voice reads it over generated visuals
`create_tiktok_slideshow`	A TikTok-style slideshow from images or clips in your library, with text on each slide
`create_ugc_studio_video`	A UGC Studio ad in one of five formats: product in hand, app demo, fashion, before / after or voiceover
`create_reddit_story`	A viral Reddit, Facebook or Instagram post and its comments, narrated over gameplay or AI visuals
`create_quiz_video`	A narrated multiple-choice quiz over gameplay or AI visuals
`create_article_to_video`	A web article read, summarised and narrated over AI images, AI video clips, your media or gameplay
`create_pdf_to_video`	The same from a PDF that is online (a public `https` link)
`create_music_to_video`	A music video for your song: AI images, AI video clips, your media or gameplay following the lyrics, cut to the beat
`create_clone_video`	A TikTok, Instagram or YouTube video remade with your voice and new AI visuals
`create_cinematic`	Cinematic Studio: a short film with speaking characters from a story brief
`create_podcast`	A two-person podcast clip from a dialogue: cut between two people, or both in one shot (dualcast)
`create_hook_demo`	An avatar reaction with your hook text, followed by your product demo
`create_scrolling_video`	A website recorded scrolling, with background music
`create_pdf_to_brainrot`	A PDF that is online, narrated over gameplay
`add_captions`	One of your library videos, transcribed and rendered again with captions
`extend_video`	One of your library videos, continued from its last frame by an AI video model
`remove_video_background`	One of your library videos with the background behind the person removed

create_article_to_video, create_pdf_to_video and create_cinematic answer in a few seconds with the project in processing; the source is read (or the storyboard written) afterwards, and a failure there shows up as a failed project with the reason in message. Create tools are not idempotent: if one fails with a timeout, check list_projects (with source: "mcp") before calling it again.

Formats

Tool	Description
`search_formats`	The formats that fit what you ask for, each with the create tool and the arguments that set it up
`list_formats`	The menu of formats by family: brainrot gameplay, stories, UGC, slideshows, faceless, animated styles, music

Search

Tool	Description
`search`	Your projects and library media whose name matches, with a link to each in Hooked (what ChatGPT’s search and deep research use)
`fetch`	One `search` result in full, as readable text

Resources

Tool	Description
`list_avatars`	The AI avatars you can use
`get_avatar`	One of your own avatars, its status and its Hook + Demo reactions
`create_avatar_from_photo`	Make one of your own avatars from a photo (public URL or library image)
`generate_avatar`	Generate one of your own avatars from a description; poll `get_avatar`
`create_reaction`	A Hook + Demo reaction clip of one of your own avatars, usable as hook demo `avatarId` once done
`delete_avatar`	Irreversible. Delete one of your own avatars and its reactions
`list_voices`	The voices you can use, library and your own
`clone_voice`	Clone a voice from a recording at a public `https` URL (5-90 s); its id works as `voiceId` everywhere. Needs `consent: true`
`generate_speech`	Text to speech with any voice, up to 1,200 characters; answers a 10-minute MP3 link
`speech_to_speech`	Re-voice a recording (up to 5 minutes) with another voice; answers a 10-minute MP3 link
`delete_voice`	Irreversible. Delete one of your own voices. Refused while a project is still generated with it
`list_music`	Background music tracks
`list_catalog`	The values a create field accepts, by `name`: `caption-presets`, `text-styles`, `visual-styles`, `motion-styles`, `games`, `reactions`, `video-models`, `image-models`, `cinematic`, `camera-moves`, `gestures`, `voice-presets`, `accents` (`GET /v1/catalog/{name}`)
`list_characters`	Your Cinematic characters; their ids go in `create_cinematic` `characterIds`
`create_character`	Save a Cinematic character with a photo from your library, a URL or a generated portrait
`update_character`	Change a character’s name, descriptions, style or photo
`delete_character`	Irreversible. Delete a Cinematic character; projects that cast it keep working
`get_account`	Plan, billing mode, credit balance, storage and (byok) which provider keys work. Never a key value

Media Library

Tool	Description
`list_media`	Images and videos of your library, newest first; their ids go in the `media` fields of the create tools
`get_media`	One library media and its `status` (`PENDING`, `PROCESSING`, `COMPLETED`, `FAILED`)
`import_media`	Import an image or video from a public `https` URL (any client) or, in ChatGPT, a file attached to the chat
`upload_media`	Open a file picker in the chat (in apps that show MCP Apps) to add an image or video from your device; the file uploads by itself
`create_media_upload`	The uploader’s first step: a library row and a presigned URL to PUT the file to (called by the widget, not by you)
`complete_media_upload`	The uploader’s second step: confirm the file is in storage and mark the media `COMPLETED` (called by the widget)
`generate_image`	The Media Generator: one AI image from a prompt into your library; poll `get_media` until `COMPLETED`
`generate_video`	The Media Generator: one AI clip, from a prompt or a library image; poll `get_media` until `COMPLETED`
`expand_image`	Grow an image to another aspect ratio; AI paints the new area into a new library image
`remove_image_background`	Cut an image out of its background (PNG with transparency)
`delete_media`	Irreversible. Delete a library image or video and its file. Refused while a project uses it

An AI app cannot send a file’s bytes through a tool call, so in a plain client give a public https link — a create tool’s photo field takes one directly, or import_media saves it to your library. In ChatGPT you can also attach the file to the chat. In an app that shows the widgets (Claude, ChatGPT) ask to upload and a picker appears: the file goes straight to storage and its id lands in the conversation. Otherwise, upload the file in the dashboard and ask the AI to find it with list_media.

Projects and Videos

Tool	Description
`get_project`	Where a project stands (`draft`, `processing`, `completed`, `failed`), with its latest video once there is one
`list_projects`	Your team’s projects, newest first, each as `get_project` shows it; filter by `status`, `type`, `source`
`get_captions`	A project’s caption lines (text, words, times), numbered as `edit_project` takes them
`edit_project`	Change a finished project’s caption style, position or text, its background music or its logo; anything else is edited in the dashboard
`render_project`	Render a project again (after `edit_project`, or after a failed render); free
`get_video`	A rendered video: status (`STARTED`, `COMPLETED`, `FAILED`) and download `url`
`list_videos`	Your team’s rendered videos, newest first
`delete_project`	Irreversible. Delete a project with its videos and their files
`delete_video`	Irreversible. Delete one rendered video and its file; its project stays

Research

Tool	Description
`list_niches`	The niches Hooked follows. 1 credit
`get_niche_trending_videos`	The videos trending in a niche on TikTok and YouTube. 1 credit
`get_niche_creators`	The creators behind a niche’s trending videos. 1 credit
`get_niche_metrics`	A niche’s views, engagement and top hashtags over the last days. 1 credit
`search_social_videos`	Search TikTok or YouTube for a term and save the results. 16 credits
`list_social_searches`	The searches your team has saved. Free
`get_social_search`	The videos one saved search found, with their views, likes and hooks. Free
`analyze_social_account`	Analyze a public TikTok or YouTube account and save it. 8 credits
`list_analyzed_accounts`	The accounts your team has analyzed. Free
`get_analyzed_account`	One analyzed account: stats, best videos, hashtags, best posting times. Free

The niche tools and the social search tools (search_social_videos, list_social_searches, get_social_search) are part of HookTrend: they appear only for teams that have it. The paid research tools are not marked read-only, so the AI asks before using them. Credits are refunded when a read fails.

Publishing

Tool	Description
`list_social_accounts`	The YouTube, TikTok and Instagram accounts your team connected (Studio → Channels)
`schedule_post`	Publish a finished video on one of those accounts at a set time. Free
`list_scheduled_posts`	Your scheduled posts and their status
`update_scheduled_post`	Change the time or the text of a post not published yet
`cancel_scheduled_post`	Cancel a post not published yet

The AI confirms the account, the time and the text with you before scheduling: a post goes out publicly.

The delete_* tools cannot be undone. Tell your agent to call them only when you ask it to delete a specific project, video, media, character or voice.

Every tool returns the API’s JSON response, as text and as structured content. A create tool returns { "success": true, "data": { "projectId", "status" }, "message" }. The video is made in the background: get_project, get_video, get_media and get_avatar take waitSeconds (up to 45), so one call waits for it instead of the AI asking again and again. list_avatars, list_voices and list_characters are paged like the API: each answer carries total, limit and offset, and the agent asks for the next page with offset until it has seen total items. list_music returns every track.

Parameter	Tools	Description
`limit`	both	Page size, 1 to 100 (default 50)
`offset`	both	Rows to skip (default 0)
`type`	both	`all` (default), `library` or `custom` (your team’s own avatars or cloned voices)
`language`	`list_voices`	Filter by language, e.g. `English`, `Spanish`
`gender`	`list_voices`	Filter by gender, e.g. `Male`, `Female`

Example Conversations

Once connected, you can interact naturally: Create a video:

“Make a 30-second video about productivity tips for remote workers, narrated by an energetic female voice.”

Avatar ad:

“Have a young female avatar read this script as a UGC Studio app demo for my fitness app, showing https://myfitnessapp.com.”

From a source:

“Turn this blog post into a 45-second vertical video: https://example.com/blog/remote-work-tips”

Browse resources:

“What AI avatars do you have? Show me the female ones.”

Check status:

“Is the video from that project ready? Give me the download link.”

A format:

“Make a brainrot video over Minecraft parkour about the history of Rome.”

Your product photo:

“Make a product-in-hand UGC ad for this serum: https://…/serum.jpg” (In an app that shows the widgets you can also say “let me upload the photo” and pick it from your device; in ChatGPT you can attach the photo to the chat instead of pasting the link.)

Publish:

“Schedule it on my TikTok for tomorrow at 6 pm.”

Tool Details

create_script_to_video

{
  "script": "Your video script here...",
  "voiceId": "voice_id_from_list_voices",
  "mediaType": "ai-images",
  "preset": "realistic",
  "aspectRatio": "ratio_9_16",
  "musicId": "music_id_from_list_music",
  "captionPreset": "wrap1",
  "webhook": "https://your-app.com/webhooks/hooked?token=YOUR_SECRET"
}

Parameter	Type	Required	Description
`script`	string	Yes	The narration, up to 10,000 characters
`voiceId`	string	Yes	Voice ID from `list_voices`
`mediaType`	string	No	`ai-images` (default), `ai-videos`, `media`, `gameplay` or `motion-graphics`
`preset`	string	No	Visual style for AI media, e.g. `realistic`, `anime`
`media`	string[]	No	Library media IDs, for `media` (required there) or `motion-graphics`
`quality`	string	No	AI media quality: `base` (default), `pro` or `ultra`
`characterMode`	string	No	`speaking`: the AI characters say the lines themselves (talking formats)
`aiModel`	string	No	Model for the AI media (`video-models` / `image-models` catalog)
`game`, `gameClip`	string	No	For `gameplay`: the game and one of its clips (`games` catalog)
`aspectRatio`	string	No	`ratio_9_16` (default), `ratio_16_9` or `ratio_1_1`
`musicId`	string	No	Music ID from `list_music`
`captionPreset`	string	No	Caption style preset, e.g. `wrap1` (default)
`webhook`	string	No	Public HTTPS URL called when the video is rendered or the project fails
`dryRun`	boolean	No	`true`: only answer the exact price; nothing is created

search_formats answers these arguments ready for each format: a Lego-look video is mediaType: "ai-videos", preset: "lego", quality: "ultra".

create_talking_avatar

Parameter	Type	Required	Description
`script`	string	Yes	What the avatar says, up to 10,000 characters
`avatarId`	string	Yes	Avatar ID from `list_avatars`
`language`	string	No	Two-letter language code; detected from the script when omitted
`musicId`	string	No	Music ID from `list_music`
`webhook`	string	No	Public HTTPS URL

create_prompt_to_video

{
  "prompt": "A video about productivity tips for remote workers",
  "voiceId": "voice_id_from_list_voices",
  "targetDuration": 30,
  "mediaType": "ai-images",
  "webhook": "https://your-app.com/webhooks/hooked?token=YOUR_SECRET"
}

Parameter	Type	Required	Description
`prompt`	string	Yes	Describe the video, up to 4,000 characters
`voiceId`	string	Yes	Voice ID from `list_voices`
`targetDuration`	number	No	Length in seconds, 10 to 120 (default 60)
`mediaType`	string	No	`ai-images` (default), `ai-videos`, `media`, `gameplay` or `motion-graphics`
`media`	string[]	No	Library media IDs, for `media` (required there) or `motion-graphics`
`webhook`	string	No	Public HTTPS URL

create_tiktok_slideshow

{
  "name": "5 productivity tips",
  "slides": [
    { "media": "library_media_id_1", "text": "Tip 1: Plan tomorrow tonight", "duration": 3 },
    { "media": "library_media_id_2", "text": "Tip 2: One task at a time" }
  ]
}

Parameter	Type	Required	Description
`slides`	array	Yes	`{ media, text?, duration? }` in order. `media` is a library media ID (image or video); `duration` is in seconds (default 3)
`name`	string	No	Project name
`musicId`	string	No	Music ID from `list_music`
`webhook`	string	No	Public HTTPS URL

create_ugc_studio_video

Calls POST /v1/project/create/ugc-studio/{format}. The format’s own fields go at the top level, next to format; see each format’s page under UGC Studio.

Parameter	Type	Required	Description
`format`	string	Yes	`product_in_hand`, `app_demo`, `fashion`, `before_after` or `voiceover`
`script`	string	Yes	The ad script, up to 10,000 characters
`avatarId`	string	Yes	Avatar ID from `list_avatars`
`productImageKey`	string	`product_in_hand`	HTTPS URL of the product photo. Required for `product_in_hand`, optional for `voiceover`
`appUrl`, `scrollMode`, `showcaseAfterSeconds`	string, string, number	`app_demo`	Website to record (unless `media` has your recording), how it scrolls, and when the screencast enters (0-15 s)
`outfitImageKeys`	string[]	`fashion`	Up to 4 outfit photos, unless `media` has your outfit videos
`beforeImageKey`, `afterImageKey`	string	No	`before_after`: your own photos, both or neither
`voiceId`	string	`voiceover`	Narrator voice from `list_voices`
`media`	string[]	No	Library media IDs: B-roll (`product_in_hand`, `before_after`), the screen recording (`app_demo`) or outfit videos (`fashion`)
`language`	string	No	Two-letter language code; detected from the script when omitted
`webhook`	string	No	Public HTTPS URL

create_article_to_video, create_pdf_to_video

{
  "sourceUrl": "https://example.com/blog/remote-work-tips",
  "voiceId": "voice_id_from_list_voices",
  "summaryType": "summarize",
  "targetDuration": 45,
  "mediaType": "ai-images",
  "preset": "cinematic"
}

Parameter	Type	Required	Description
`sourceUrl`	string	`create_article_to_video`	Public `https` URL of the article
`pdfUrl`	string	`create_pdf_to_video`	Public `https` link to the PDF, up to 20 MB, with a text layer (a scanned PDF is refused). A file that is not online cannot be sent through the MCP server: use the API with `pdfBase64`
`voiceId`	string	Yes	Voice ID from `list_voices`
`summaryType`	string	No	`summarize` (default), `summarize_long` or `key_as_is`
`targetDuration`	number	No	Length in seconds, 10 to 600 (default 30); ignored with `key_as_is`
`customPrompt`	string	No	Extra instructions for the summary (`summarize` only)
`mediaType`	string	No	`ai-images` (default), `ai-videos`, `media` or `gameplay`
`preset`	string	No	Visual style for AI media, e.g. `realistic`, `cinematic`
`media`	string[]	No	Library media IDs, required for `media`
`aspectRatio`	string	No	`ratio_9_16` (default), `ratio_16_9` or `ratio_1_1`
`musicId`	string	No	Music ID from `list_music`
`webhook`	string	No	Public HTTPS URL

BYOK teams need a Firecrawl key for create_article_to_video (it reads the page with it), not for create_pdf_to_video.

create_music_to_video

{
  "audioUrl": "https://example.com/music/summer-nights.mp3",
  "mediaType": "ai-images",
  "preset": "anime",
  "visualGuidelines": "A road trip along the coast at sunset"
}

Parameter	Type	Required	Description
`audioUrl`	string	`audioUrl` or `musicId`	Direct public `https` link to the song file (MP3, M4A, AAC, WAV or OGG, up to 10 MB). A Spotify, YouTube or Suno song page is refused: link the file itself
`musicId`	string	`audioUrl` or `musicId`	A track from `list_music` or one of your team’s own audio files
`mediaType`	string	No	`ai-images` (default), `ai-videos`, `media` or `gameplay`
`preset`	string	No	Visual style for AI media, e.g. `anime`, `cinematic`
`quality`	string	No	AI media quality: `base` (default), `pro` or `ultra`
`visualGuidelines`	string	No	Extra direction for the AI visuals, up to 2,000 characters
`media`	string[]	No	Library media IDs, required for `media`
`addSoundWave`	boolean	No	Audio-reactive sound wave over the video (default `true`)
`aspectRatio`	string	No	`ratio_9_16` (default), `ratio_16_9` or `ratio_1_1`
`language`	string	No	Two-letter code of the lyrics; detected when omitted
`webhook`	string	No	Public HTTPS URL

The video follows the lyrics: a song without vocals fails, with the credits refunded. The song is copied into your media library and counts toward your storage. BYOK teams need OpenRouter (and Gemini for the Omni Flash model); no ElevenLabs.

create_clone_video

Parameter	Type	Required	Description
`sourceVideoUrl`	string	Yes	A TikTok, Instagram or YouTube video URL
`voiceId`	string	Yes	Voice ID from `list_voices`
`targetDurationSeconds`	number	No	Length of the new video in seconds, 1 to 600 (default 30)
`mustInclude`	string	No	Something the new video has to mention, up to 500 characters
`mediaType`	string	No	`ai-images` (default) or `ai-videos`
`aspectRatio`	string	No	`ratio_9_16` (default), `ratio_16_9` or `ratio_1_1`
`musicId`	string	No	Music ID from `list_music`
`webhook`	string	No	Public HTTPS URL

create_reddit_story, create_quiz_video

create_reddit_story takes prompt (what the post is about), voiceId, and optionally platform (reddit, facebook, instagram), targetDuration (10-600 s), language, narrationScope (post_and_comments or comments_only), showCard, cardTheme and name. create_quiz_video takes questions (2 to 10 { question, options, correctIndex }, written in the language they are read in), voiceId, and optionally title, answerSeconds, revealSeconds, showOutro, outroText and name. Both take the same background: mediaType (gameplay by default, ai-images, ai-videos or media), preset, quality, aiModel, game, gameClip, media, plus aspectRatio, musicId, webhook and dryRun. See Reddit Story and Quiz Video.

create_podcast, create_hook_demo, create_scrolling_video, create_pdf_to_brainrot

Tool	Parameters
`create_podcast`	`script` (each turn in `[A] … [/A]` or `[B] … [/B]`), `mode?` (`podcast` or `dualcast`); podcast: `avatarId` (host) and `guest` (`{ source: "generated", description }` or `{ source: "avatar", avatarId }`, `voice?`), `hostSet?`; dualcast: `dualcastImageKey`, `dualcastAvatarId` or `dualcastAvatars` (`{ left, right }`), `dualcastSet?`; `productImageKey?`, `hostVoice?`, `accent?`, `language?`, `aspectRatio?`, `musicId?`, `name?`, `webhook?`, `dryRun?`
`create_hook_demo`	`avatarId` (a reaction: `reactions` catalog), `text`, `media?` (library ids played after it), `textPreset?` (`text-styles` catalog), `textAlignment?`, `aspectRatio?`, `musicId?`, `name?`, `webhook?`
`create_scrolling_video`	`websiteUrl`, `scrollMode?` (`restart`, `loop`, `once`, `fixed`), `targetDuration?` (5-120 s), `aspectRatio?`, `musicId?`, `name?`, `webhook?`
`create_pdf_to_brainrot`	`pdfUrl`, `voiceId`, `summaryType?`, `targetDuration?`, `customPrompt?`, `game?`, `gameClip?`, `aspectRatio?`, `musicId?`, `language?`, `name?`, `webhook?`

See Podcast Interview, Hook Demo, Scrolling Video and PDF to Brainrot.

add_captions, extend_video, remove_video_background, expand_image, remove_image_background

Tool	Parameters
`add_captions`	`mediaId` (a library video), `language?`, `captionPreset?`, `captionAlignment?`, `musicId?`, `name?`, `webhook?`
`extend_video`	`mediaId` (a library video), `prompt`, `targetDuration?` (seconds to add), `model?`, `resolution?` (`720p`, `1080p`), `generateAudio?`, `name?`, `webhook?`
`remove_video_background`	`mediaId` (a library video up to 10 minutes), `aspectRatio?`, `captionPreset?` (captions only when set), `language?`, `name?`, `webhook?`
`expand_image`	`mediaId` or `imageUrl`, `aspectRatio` (the new ratio), `prompt?` (what the new area shows)
`remove_image_background`	`mediaId` or `imageUrl`, `waitSeconds?`

add_captions, extend_video and remove_video_background create projects, followed with get_project. expand_image answers a new library image to follow with get_media. remove_image_background charges once: calling it again while it runs, or after it is done, is free and answers its state.

schedule_post

videoId (a rendered video: get_project → video.id, or list_videos), integrationId (from list_social_accounts), scheduledDateTime (ISO 8601 in UTC with Z, in the future) and platformData?: title, description, tags and privacyStatus for YouTube; title and privacyStatus for TikTok; caption for Instagram. Accounts are connected in the dashboard, not through the MCP server. See Schedule Video.

create_cinematic

Cinematic Studio: a short film with speaking characters from a story brief. The call answers in 1-2 minutes, because the storyboard is written first. Credits for the whole run are checked up front and charged in stages. Parameters: storyBrief (max 4,000 characters), durationSeconds (10-120), model (a video model, e.g. grok_imagine_video), and optionally storyTone, presetId, aspectRatio, language (en or es), characterIds (from list_characters), musicId, webhook. The accepted models, tones and styles come from list_catalog with name: "cinematic". See Cinematic Studio.

Tool	Parameters
`get_project`	`projectId` (from a create tool or `list_projects`), `waitSeconds?` (up to 45)
`list_projects`	`status?` (`draft`, `processing`, `completed`, `failed`), `type?` (a create endpoint name such as `script-to-video` or `ugc-studio`; see List Projects), `source?` (`api`, `web`, `mcp`), `search?` (part of the name), `limit?` (1 to 100, default 50), `offset?`
`get_video`	`videoId` (from `get_project` or `list_videos`), `waitSeconds?`
`list_videos`	`status?` (`STARTED`, `COMPLETED`, `FAILED`), `limit?` (1 to 100, default 50), `offset?`

A project shows up in list_videos only once its render has started. list_projects lists it as soon as it is created.

get_captions, edit_project, render_project

Tool	Parameters
`get_captions`	`projectId`
`edit_project`	`projectId`, and at least one of: `caption?` (`preset?`, `alignment?` `top`/`middle`/`bottom`, `disabled?`), `captionSegments?` (`[{ index, text }]`, indexes from `get_captions`), `music?` (`musicId?` from `list_music` or `null` to remove, `volume?` 0 to 1), `branding?` (`logoMediaId?` a library image or `null` to remove, `position?` a corner, `size?` 20 to 200)
`render_project`	`projectId`, `webhook?`

edit_project saves the changes to the project’s timeline as the dashboard editor does, and renders nothing: call render_project afterwards and follow the new videoId with get_video. A new caption style re-cuts the lines, so call get_captions again before correcting text. Anything else (clips, voice, script, timing, text layers) is edited in the dashboard. Refused (409) while the project is processing or one of its videos renders. See Edit Project and Get Project Captions.

delete_project, delete_video, delete_media

All three are irreversible. Each takes one id: projectId, videoId or mediaId.

Tool	Deletes	Refused (409, nothing deleted) while
`delete_project`	The project, its videos and their files, posts recorded for them. Library media stays	The project is `processing`, one of its videos renders, or one has a pending scheduled post
`delete_video`	The video, its file and thumbnail, posts recorded for it. The project stays	The video renders, or it has a pending scheduled post
`delete_media`	The library image or video and its file	A project uses it, or a dashboard generation is still making it

A pending scheduled post comes back in data.scheduleIds; cancel it in the dashboard or with Cancel Scheduled Post first. See Delete Project, Delete Video and Delete Media.

create_character, update_character, delete_character

create_character takes name and optionally description, voiceProfile, presetId and one photo source: mediaId (a COMPLETED library image), imageUrl (imported after the answer) or generatePortrait: true (a portrait from the description, charged one image: 4 credits with the default imageModel, given back if it fails). With imageUrl or generatePortrait the character comes back with imageStatus: "PROCESSING"; it is ready when list_characters shows it COMPLETED. update_character takes characterId and only the fields to change (null clears description or voiceProfile; mediaId: null removes the photo). delete_character is irreversible and never refused: Cinematic projects that cast the character keep their own copy. See Create Character, Update Character and Delete Character.

list_media, get_media, import_media, upload_media, generate_image, generate_video

Tool	Parameters
`list_media`	`type?` (`image`, `video`), `search?` (part of the name), `limit?` (1 to 100, default 50), `offset?`
`get_media`	`mediaId` (from `list_media`, `import_media`, `generate_image` or `generate_video`), `waitSeconds?`
`import_media`	`url` (a public `https` link; any client) or, in ChatGPT, `file` (a chat attachment); JPEG, PNG, WEBP, GIF, MP4, MOV or WEBM up to 100 MB; `name?`
`upload_media`	No parameters: opens the in-chat file picker. In apps that show MCP Apps; a plain client has no picker, so use `import_media` there
`generate_image`	`prompt`, `model?` (`image-models` catalog), `aspectRatio?`, `style?` (`visual-styles` catalog), `referenceMediaIds?`, `name?`
`generate_video`	`prompt`, `model?` (`video-models` catalog), `imageMediaId?` or `imageUrl?`, `endImageMediaId?`, `referenceMediaIds?`, `durationSeconds?`, `aspectRatio?`, `resolution?`, `audio?`, `name?`

import_media, generate_image and generate_video answer { mediaId, status: "PROCESSING" } at once. A create tool refuses a media that is not COMPLETED yet, so the agent polls get_media first. upload_media is driven by the widget end to end: it asks create_media_upload for a presigned URL, PUTs the file to storage, then calls complete_media_upload; the widget writes the new id into the conversation for the agent to use. The generate tools cost what the dashboard’s Media Generator costs (usedCredits in the answer) and give the credits back when the generation fails; the video-models catalog says what each video model takes (durations, resolutions, ratios, start and end frames, references, sound). See Generate Image and Generate Video.

Credits and Your Own Keys

Videos made through the MCP server cost the same as in the dashboard. Teams on bring-your-own-keys need their own provider keys in Settings → AI keys; when one is missing, the create tools fail with a 402 missing_credentials error listing the providers to add. Create tools are not idempotent. If one times out there is no projectId to check, so look with list_projects (newest first, source: "mcp"; a project is listed as soon as it is created) before trying again, or you may get (and pay for) the same video twice. Projects made through the MCP server carry an AI badge in the dashboard.

Troubleshooting

The tools do not appear in the app

Restart the app after adding the config, and in ChatGPT pick Hooked from the tools menu of the chat. Viewers of a team cannot create content.

402 missing_credentials

Your team brings its own provider keys and one is missing: add it in Settings → AI keys. The error lists which.

​Overview

​Connect

​With an API key instead

​Approving tool calls

​What it can do

​Video Creation

​Formats

​Search

​Resources

​Media Library

​Projects and Videos

​Research

​Publishing

​Example Conversations

​Tool Details

​create_script_to_video

​create_talking_avatar

​create_prompt_to_video

​create_tiktok_slideshow

​create_ugc_studio_video

​create_article_to_video, create_pdf_to_video

​create_music_to_video

​create_clone_video

​create_reddit_story, create_quiz_video

​create_podcast, create_hook_demo, create_scrolling_video, create_pdf_to_brainrot

​add_captions, extend_video, remove_video_background, expand_image, remove_image_background

​schedule_post

​create_cinematic

​get_project, list_projects, get_video, list_videos

​get_captions, edit_project, render_project

​delete_project, delete_video, delete_media

​create_character, update_character, delete_character

​list_media, get_media, import_media, upload_media, generate_image, generate_video

​Credits and Your Own Keys

​Troubleshooting

​Next Steps

Skill File

Webhooks

Overview

Connect

With an API key instead

Approving tool calls

What it can do

Video Creation

Formats

Search

Resources

Media Library

Projects and Videos

Research

Publishing

Example Conversations

Tool Details

create_script_to_video

create_talking_avatar

create_prompt_to_video

create_tiktok_slideshow

create_ugc_studio_video

create_article_to_video, create_pdf_to_video

create_music_to_video

create_clone_video

create_reddit_story, create_quiz_video

create_podcast, create_hook_demo, create_scrolling_video, create_pdf_to_brainrot

add_captions, extend_video, remove_video_background, expand_image, remove_image_background

schedule_post

create_cinematic

get_project, list_projects, get_video, list_videos

get_captions, edit_project, render_project

delete_project, delete_video, delete_media

create_character, update_character, delete_character

list_media, get_media, import_media, upload_media, generate_image, generate_video

Credits and Your Own Keys

Troubleshooting

Next Steps