KreatorFlow

Gere e edite vídeos com IA (Seedance 2.5, Kling, Wan) e imagens (Seedream), sincronize os lábios, crie vozes e planeje cenas em 3D pelo Claude, ChatGPT, Cursor ou VS Code. 33 ferramentas, OAuth.

Servidor MCP hospedado

npx add-mcp 'https://kreatorflow.ai/api/mcp'

Instala no Claude Code, Codex, Cursor e outros

Documentação

Seedance MCP server for AI video and images

Generate videos & images

See the setup steps

Connect Claude

  1. 1In Claude, open Customize, then Connectors, click + Add and choose Add custom connector (on Team and Enterprise plans an Owner adds it first under Organization settings, then you click Connect).
  2. 2Name it KreatorFlow, paste the server URL and click Continue.
  3. 3Sign in with your KreatorFlow account, press Allow, then click Add.
  4. 4In a chat, click + at the lower left, open Connectors, turn KreatorFlow on and ask for an image or a video.

https://kreatorflow.ai/api/mcp

Requires an active paid plan. Generations use your KreatorFlow plan credits. No API key needed: you sign in in the browser.

KreatorFlow is a remote MCP server (https://kreatorflow.ai/api/mcp, OAuth) that lets Claude, ChatGPT, Cursor and VS Code make Seedance 2.5/2.0 video, images, Lip Sync and audio on your plan credits; developers use /v1/mcp with an API key.

Your AI assistant, now with a video studio

The KreatorFlow MCP server connects Claude, ChatGPT, Cursor, VS Code and other Model Context Protocol clients to your KreatorFlow account. Describe a shot in the chat and the assistant generates it with Seedance for video and Seedream for images, then hands you the link. It can also edit images and video, swap faces, run Kitsune motion transfer and object swaps on a clip, and turn text into speech, make music and sound effects, clone voices, and transcribe or translate speech. Generations are saved to your History like anything you make in the studio.

You sign in once with OAuth, so there is no API key to paste, and each generation uses your plan credits at 10% more than the video studio and image studio price, to cover running the connector. KreatorFlow is an official partner of BytePlus, a ByteDance company.

Coming from another MCP video tool? Flashloop MCP vs KreatorFlow MCP puts the two connectors side by side: setup in each app, the tools and how billing works, and AI video MCP servers compared lists what each video platform's own MCP page says about models, sign-in and billing.

BytePlus, a ByteDance company

Tools your assistant gets

The server at https://kreatorflow.ai/api/mcp exposes 33 tools. Video, edits and Kitsune run as jobs: the assistant starts one, then checks it with get_job until it is ready. Audio tools reply with the result straight away. The Stage tools plan scenes with Kiryu for free and return a link that opens the scene in Stage, where Make it real turns it into a video.

Inputs marked * are required.

  • get_account

    Shows your KreatorFlow plan and the credits you have left. No inputs.
  • generate_image

    Makes an image with Seedream (Seedream 5.5, the default model, unless you pick another) and returns the image URL, or a job to poll. Spends plan credits.
    • prompt
      • aspect_ratio
      • resolution
      • model
      • template
      • style_template
      • use_pass
  • generate_video

    Starts a Seedance video (Seedance 2.5 unless you pick a model), from text or from a first frame, with optional voice clips the mouths follow, and returns a job id to poll with get_job. Spends plan credits, or uses your Unlimited pass for that model when you ask for it.
    • prompt
      • duration
      • aspect_ratio
      • resolution
      • model
      • audio
      • first_frame
      • audio_references
      • reference_videos
      • video_role
      • keep_clip_sound
      • shots
      • voices
      • character_voices
      • template
      • style_template
      • use_pass
  • generate_speech

    Text to speech: reads your text aloud in a built-in voice (an English one by default) or one of your cloned voices and returns the audio link and its uploadId in the same reply (the uploadId works as Lip Sync audio). Spends plan credits by the length of the text.
    • text
      • voice
      • speed
      • volume
      • pitch
      • emotion
  • generate_sound

    Makes music, sound effects or dialogue from a text prompt and returns the audio link in the same reply. Spends plan credits by the length of the audio.
    • prompt
      • tone
      • speech_rate
      • volume
  • kitsune

    KreatorFlow Kitsune: moves the motion of a source video onto your character, swaps an object or swaps a face. Inputs are workspace uploadIds. A face or a character photo needs the account owner's consent, confirmed once per image on kreatorflow.ai (the call returns the link until then). Returns a job to poll with get_job. Spends plan credits.
    • mode
      • video
      • source_duration_seconds
      • images
      • target
      • prompt
      • width
      • height
      • style_transform_id
      • casting
      • tier
      • orientation
  • lip_sync

    Makes the person in a photo speak a voice track (OmniHuman 2.0), 4 seconds to 5 minutes. The photo and the voice are workspace uploadIds (the uploadId generate_speech returns works as the voice). Paid plans only. Only use photos of people who agreed: calling it records your acceptance of the Terms and Acceptable Use Policy for the photo. Returns a job to poll with get_job. Spends plan credits by the second of audio.
    • image
      • audio
      • prompt
      • resolution
  • edit_image

    Edits an image you already have with a prompt (Seedream image edit). The source is a workspace uploadId. Returns a job to poll with get_job. Spends plan credits.
    • prompt
      • source
      • model
      • resolution
      • aspect_ratio
      • reference_uploads
  • face_swap

    Puts the face from one image onto another. Both images are workspace uploadIds. The account owner confirms once per new face, on kreatorflow.ai, that they are that person or have their permission: until then the call returns a confirmation link to show them. Returns a job to poll with get_job. Spends plan credits.
    • target
      • face
  • edit_video

    Transforms, restyles or extends a video you already have (Seedance). The source is a workspace uploadId. Returns a job to poll with get_job. Spends plan credits.
    • source_video
      • edit_mode
      • prompt
      • source_origin
      • resolution
      • ratio
  • clone_voice

    Creates a cloned voice from an audio sample (a public https URL or a workspace uploadId). The account owner must first confirm on kreatorflow.ai that they own the voice or have written permission: the first call returns a confirmation link to show them, then call again. Spends plan credits.
    • name
      • audio
      • language
      • description
  • transcribe

    Speech to text: turns spoken audio (a public https URL or a workspace uploadId) into a transcript in the same reply. Spends plan credits by the length of the audio.
    • audio
      • language
  • translate_speech

    Translates spoken audio into another language and speaks it back. Returns the translated text, an audio link and its uploadId. Spends plan credits by the length of the audio.
    • audio
      • target_language
      • source_language
      • voice
  • save_character

    Saves an image from your workspace as a reusable @character in your KreatorFlow library.
    • name
      • upload_id
      • prompt
  • list_characters

    Lists the reusable characters saved in your KreatorFlow library.
    • limit
      • type
  • list_history

    Lists your recent images, videos and audio, each with the uploadId the edit, Kitsune and face swap tools take.
    • type
      • limit
      • surface
  • list_voices

    Lists the voices generate_speech can use: KreatorFlow's built-in voices (name, language, description) and your cloned voices. Pass an entry's voice value as generate_speech's voice. No inputs.
  • list_models

    Lists the KreatorFlow image models you can pick from (ids, names and tags). No inputs.
  • list_templates

    Lists KreatorFlow templates: official and community ones for each tool (image, video, Stage, Kitsune, Kartoonz, Lip Sync...) and cross-tool styles, with what each asks you to fill in and its settings. Never a locked recipe.
    • tool
      • compatible_with
      • query
      • source
      • sort
      • limit
      • cursor
  • get_template

    One template in full: what you fill in (slots), its settings and which are locked, the tools a style works in, and its example output. The recipe text is shown only when the author opened it or it is yours.
    • id
  • use_template

    Uses a template. A tool template runs in its own tool; a style runs in any tool it works in. Starts the job like the matching generate tool (same credits or Unlimited pass), or with run false returns the ready params, missing slots and credit estimate without spending.
    • id
      • tool
      • slots
      • prompt
      • base_template
      • overrides
      • run
      • use_pass
  • save_template

    Saves a finished result of yours as your own template (private by default). Pick the words people replace; reference pictures always become slots. As a style it saves just the look so other tools can use it.
    • job_id
      • kind
      • title
      • description
      • slots
      • style_line
      • works_in
      • visibility
  • publish_template

    Changes who can see one of your templates: private, unlisted (anyone with the link) or public (the marketplace). Public needs a verified email, an account a day old and a finished result of yours as the cover.
    • id
      • visibility
      • cover_job_id
  • upload_media

    Copies a public image, video or audio URL into your workspace and returns an uploadId for the edit, Kitsune, face swap, voice and character tools.
    • url
      • kind
  • create_upload

    Uploads a file from your own computer with no public link: returns a private upload URL for your account (15 minutes, one file), the exact curl command to run and an uploadId. AI tools that can run commands (Claude Code, Cursor, Codex) run the command; in a web chat you open the returned upload page and drop the file there. Same size and type limits as uploads on the site.
    • filename
      • kind
      • size_bytes
      • content_type
  • confirm_upload

    Checks that the file for a create_upload uploadId has arrived (size and real type) and makes the uploadId usable. The other tools also do this the first time they get the uploadId.
    • upload_id
  • get_job

    Checks a generation job and returns its status (QUEUED, PROCESSING, COMPLETED or FAILED) and, once it is done, the result URL. Works for image, video, edit, Kitsune and Lip Sync jobs.
    • job_id
  • ask_kiryu

    Talk to Kiryu, KreatorFlow's AI director, and let it drive the studio: it answers and makes the cards for what you ask (video, image, voiceover, sound, edits, captions, lip sync), and with max_credits above 0 it runs them for you, never past that cap. Same plan, consent and free-plan rules as the website; the cards it runs cost credits at the connector's price (the studio price plus 10%). Keep the conversationId to carry on ("is it done? now add captions").
    • message
      • conversationId
      • attachments
      • max_credits
      • mode
      • run_pending
  • get_kiryu_chat

    Reads a Kiryu chat (recent messages, each card's status and job, the files made in it), or lists your recent Kiryu chats, including ones started on kreatorflow.ai. Free.
    • conversationId
      • limit
  • stage_create_scene

    Plans a video scene in Stage, the free 3D shot planner: Kiryu turns your idea into shots with simple 3D people, sets, camera moves and lighting, or writes a What If (hook, escalation, payoff, spoken lines). Returns Kiryu's reply, the shot list and a link that opens the scene in Stage for your account. Saved @characters and reference pictures can play the people. Free: Stage makes a plan, not a video; open the link on a computer and press Make it real for the video (credits shown first), or use generate_video for a video right now.
    • idea
      • what_if
      • aspect
      • length_seconds
      • characters
      • references
  • stage_edit_scene

    Changes a Stage scene in plain words ("make it night", "closer on her face", "add a shot where he slips") and returns Kiryu's reply, what changed and the updated link. Works on a scene from stage_create_scene or a shared Stage link (edited as your own copy). Free: no credits.
    • scene
      • instruction
      • shot
      • references
  • stage_list_examples

    Lists Stage's example scenes (What Ifs, genre scenes, starters and the showcase scenes that were made real) with a link that opens each one in Stage, a preview video and a prompt to try with stage_create_scene. Free.
    • kind
  • stage_make_it_real

    Explains how to turn a Stage scene into a real video: Make it real renders in the browser, so this returns the scene's link and what to press in Stage (open it on a computer, press Make it real). Stage shows the credits first; this call spends nothing. For a video right now, use generate_video.
    • scene

Set it up in your app

Every app connects to the same server URL, https://kreatorflow.ai/api/mcp, and signs in with your KreatorFlow account.

Claude (web and Claude Desktop)

  1. 1In Claude, open Customize, then Connectors, click + Add and choose Add custom connector (on Team and Enterprise plans an Owner adds it first under Organization settings, then you click Connect).
  2. 2Name it KreatorFlow, paste the server URL and click Continue.
  3. 3Sign in with your KreatorFlow account, press Allow, then click Add.
  4. 4In a chat, click + at the lower left, open Connectors, turn KreatorFlow on and ask for an image or a video.

Server URL

https://kreatorflow.ai/api/mcp

Claude Code

  1. 1Run the command below in your terminal.
  2. 2Start Claude Code, type /mcp, pick kreatorflow and sign in when the browser opens.
  3. 3Check the connection with claude mcp list.

Command

claude mcp add --transport http --scope user kreatorflow https://kreatorflow.ai/api/mcp

ChatGPT

  1. 1In ChatGPT, open the Plugins page, click + and choose Add custom MCP server (your account or workspace settings decide whether custom servers are allowed).
  2. 2Name it KreatorFlow, paste the server URL under Connection and choose OAuth as the authentication.
  3. 3Confirm the warning, sign in with your KreatorFlow account and press Allow.

Server URL

https://kreatorflow.ai/api/mcp

Cursor

  1. 1Click Add to Cursor on the KreatorFlow MCP page, or add the config below to ~/.cursor/mcp.json under mcpServers.
  2. 2Open Cursor Settings, then MCP, and click Connect next to kreatorflow.
  3. 3Sign in with your KreatorFlow account in the browser and press Allow.

~/.cursor/mcp.json

{
  "mcpServers": {
    "kreatorflow": {
      "url": "https://kreatorflow.ai/api/mcp"
    }
  }
}

Add to Cursor

VS Code (GitHub Copilot)

  1. 1Click Add to VS Code on the KreatorFlow MCP page, or add the config below to.vscode/mcp.json (or run MCP: Add Server from the Command Palette).
  2. 2Start the server from the MCP view and sign in with your KreatorFlow account when VS Code asks.
  3. 3Use it from Copilot Chat in Agent mode.

.vscode/mcp.json

{
  "servers": {
    "kreatorflow": {
      "type": "http",
      "url": "https://kreatorflow.ai/api/mcp"
    }
  }
}

Add to VS Code

Windsurf

  1. 1Open Windsurf Settings, then Cascade, then MCP servers, and choose to edit the raw config.
  2. 2Add the config below to mcp_config.json and refresh the server list.
  3. 3Sign in with your KreatorFlow account when Windsurf opens the browser.

mcp_config.json

{
  "mcpServers": {
    "kreatorflow": {
      "serverUrl": "https://kreatorflow.ai/api/mcp"
    }
  }
}

Other MCP clients

  1. 1Any client that supports remote MCP servers over Streamable HTTP with OAuth can connect to the server URL directly.
  2. 2Clients that only run local servers can use the mcp-remote bridge with the command below.
  3. 3Sign in with your KreatorFlow account when the browser opens.

Command (stdio bridge)

npx -y mcp-remote https://kreatorflow.ai/api/mcp

How billing works

Plan credits, studio price + 10%

The MCP server spends the credits on your plan, at 10% more than the studio price (the extra covers running the connector): a Seedream 5.5 image at 2K is 2 credits and a 5-second 720p Seedance 2.5 clip is 29 credits. Unlimited pass clips use your pass as usual.

Audio prices

Text to speech (generate_speech) is 2 credits per 1,000 characters, charged in whole 1,000-character blocks (a clip can be up to 5,000 characters). Music and sound (generate_sound) is 1 credit per 10 seconds of audio, at least 2 credits. A cloned voice is 60 credits, transcription is 1 credit per 5 minutes of audio and speech translation is 3 credits per minute of audio. These are Audio studio prices; through MCP they cost 10% more.

Paid plans only

Connecting needs a paid plan (Creator, Pro, Max or Business). Plan rules apply as usual, so 4K video needs Pro or above.

Stage planning is free

The Stage tools spend no credits. Kiryu's builds share one daily allowance with Stage on the site, and Make it real in Stage shows the credits before anything is made.

Failed jobs are refunded

If a generation fails or is blocked, its credits go back to your balance automatically, just like in the studio.

Check your balance

Ask the assistant to call get_account to see your plan and remaining credits before a big batch.

Developer server: tokens

The developer MCP server bills prepaid Developer API tokens instead of plan credits. One token is $0.05, and a 5-second 720p Seedance 2.0 clip is 11 tokens.

No double charging

Both servers run your request through the same billing as the studio or the API, so a generation is charged once, in one place.

For developers: the API key MCP server

Building an agent or a backend? https://kreatorflow.ai/v1/mcp takes a kf_ API key instead of a sign-in and bills Developer API tokens.

Send your key as Authorization: Bearer kf_... (or an x-api-key header) to https://kreatorflow.ai/v1/mcp. Every tool calls the same /v1 endpoints as the Seedance API, with the same scopes, prices and job records. Files on your computer go up with create_upload (it returns a ready curl command) and confirm_upload, then the upload_id works as a reference image, frame, clip or lip sync photo. The full tool schemas are in the developer docs.

Inputs marked * are required.

  • list_models

    Lists the image and video model ids with their limits and token costs. No inputs.
  • create_upload

    Uploads a file from your own computer with no public link, like POST /v1/uploads: returns an upload_id, a private upload URL for your account (10 minutes, one file) and the exact curl command to run. Same size and type limits as the REST upload. Free.
    • file_name
      • mime_type
      • file_size
      • kind
  • confirm_upload

    Checks that the file for a create_upload upload_id has arrived (size and real type), like POST /v1/uploads/complete, and makes the upload_id usable as reference_image_upload_ids, first_frame_upload_id, reference_video_upload_ids or image_upload_id.
    • upload_id
  • generate_image

    Creates a Seedream or Wan image job through POST /v1/images and returns a job to poll with get_job.
    • prompt
      • model
      • aspect_ratio
      • reference_image_urls
      • reference_image_upload_ids
      • character_id
      • seed
      • watermark
      • idempotency_key
  • generate_video

    Creates a Seedance or Wan video job through POST /v1/videos and returns a job to poll with get_job.
    • prompt
      • model
      • duration
      • aspect_ratio
      • resolution
      • first_frame_url
      • last_frame_url
      • reference_image_urls
      • first_frame_upload_id
      • last_frame_upload_id
      • reference_image_upload_ids
      • reference_video_upload_ids
      • reference_audio_upload_ids
      • generate_audio
      • idempotency_key
      • video_role
      • keep_clip_sound
      • shots
      • character_ids
      • voice_ids
      • character_voices
  • edit_video

    Transforms, restyles or extends an existing clip with Seedance 2 through POST /v1/video-edits.
    • edit_mode
      • source_video_url
      • source_video_upload_id
      • prompt
      • model
      • idempotency_key
  • lip_sync

    Makes the person in a photo speak a voice track through POST /v1/lip-sync and returns a queued job to poll with get_job. Only use photos of people who agreed (Acceptable Use Policy).
    • image_upload_id
      • audio_upload_id
      • audio_job_id
      • prompt
      • resolution
      • idempotency_key
  • list_voices

    Lists the voice ids you can pass to generate_speech (needs the models:read scope). No inputs.
  • generate_speech

    Turns text into speech (same as POST /v1/audio). Finishes in one call: the returned job is already completed.
    • text
      • voice
      • model
      • speed
      • volume
      • pitch
      • emotion
      • emotion_scale
      • idempotency_key
  • generate_sound

    Generates music or a sound effect from a description. Finishes in one call: the returned job is already completed.
    • prompt
      • model
      • speed
      • volume
      • pitch
      • idempotency_key
  • get_job

    Polls a job by id (needs the jobs:read scope). When it is completed, providerOutputUrl holds the result.
    • job_id
  • get_balance

    Returns your remaining Developer API token balance (needs the models:read scope). No inputs.

MCP server questions

What is the KreatorFlow MCP server?

KreatorFlow is a remote MCP server (https://kreatorflow.ai/api/mcp, OAuth) that lets Claude, ChatGPT, Cursor and VS Code make Seedance 2.5/2.0 video, images, Lip Sync and audio on your plan credits; developers use /v1/mcp with an API key. You ask for an image, a video, a voiceover or an edit in the chat, the assistant calls KreatorFlow, and the result comes back as a link, made with Seedance and Seedream on your own account.

Which apps can I connect?

Claude (web and Claude Desktop), Claude Code, ChatGPT, Cursor, VS Code with GitHub Copilot, Windsurf and any other client that supports remote MCP servers over Streamable HTTP with OAuth. Clients that only run local servers can connect through the mcp-remote bridge. The server URL is https://kreatorflow.ai/api/mcp for all of them.

How much does a generation cost through MCP?

10% more credits than the same thing in the studio, taken from your plan credits; the extra covers running the connector. For example, a Seedream 5.5 image at 2K is 2 credits and a 5-second 720p Seedance 2.5 clip is 29 credits. Text to speech is 2 credits per 1,000 characters in the studio, plus 10% through MCP. Unlimited pass clips made through MCP use your pass as usual. Ask the assistant to call get_account to see your balance, and failed jobs are refunded automatically.

Do I need an API key?

No. The connector signs you in with your KreatorFlow account through OAuth, so there is nothing to copy. Developers who prefer keys can use the developer MCP server at https://kreatorflow.ai/v1/mcp with a kf_ API key, billed in Developer API tokens.

Do I need a subscription to use the KreatorFlow MCP server?

Not for the developer MCP server. https://kreatorflow.ai/v1/mcp takes a KreatorFlow API key and spends prepaid Developer API tokens (top-ups from $20, one token is $0.05), with no subscription. It works in Claude Code, Cursor, n8n's MCP Client Tool and other agents that can send the key as a header. The one-URL connector at https://kreatorflow.ai/api/mcp, which signs in with OAuth for Claude, ChatGPT, Cursor and VS Code, spends plan credits instead and needs a paid plan (Creator, Pro, Max or Business) or the $3 48-hour trial on Creator, Pro or Max (it rolls into the plan you picked unless you cancel within the 48 hours: Creator continues at $12 for the first month and $25/month, Pro at $49/month, Max at $99/month). Free accounts see "A paid plan is required" on the connector sign-in screen, and what you can generate follows your plan, for example 4K video needs Pro or above.

Which models does it use?

Images use Seedream 5.5 unless the assistant passes another Seedream model. Video uses Seedance 2.5 by default, at 5 seconds and 720p unless you ask for something else. Powered by ByteDance's Seedance 2.0 and Seedream models via BytePlus, a ByteDance company.

Can my assistant make voiceovers and other audio?

Yes. generate_speech reads your text aloud in a built-in voice or one of your cloned voices, and generate_sound makes music, sound effects or dialogue from a prompt. clone_voice, transcribe and translate_speech cover voice cloning, speech to text and speech translation. Speech, sound, transcripts and translations come back in the same reply, with no job to check. Text to speech (generate_speech) is 2 credits per 1,000 characters, charged in whole 1,000-character blocks (a clip can be up to 5,000 characters). Music and sound (generate_sound) is 1 credit per 10 seconds of audio, at least 2 credits. A cloned voice is 60 credits, transcription is 1 credit per 5 minutes of audio and speech translation is 3 credits per minute of audio.

The connection stopped working. What do I do?

Remove the KreatorFlow connector in your app and add it again, then sign in. Sign-ins renew on their own while you use them, and a connection left unused for 30 days asks you to sign in again. If it says your account no longer has access, check that you are still a member of the workspace and that its plan is active.

Generate video from your AI assistant

Pick a paid plan, add the KreatorFlow connector and ask for your first Seedance clip.

Keep exploring