Seedance MCP Server: Make Seedance Videos from Claude, ChatGPT and Cursor

What the server can do

The server exposes 9 tools: 6 only read or quote and are free, and 3 start paid work. Each tool is a thin wrapper around one Open API endpoint, so limits, idempotency, billing and refunds behave exactly as they do over HTTP.

Connect your assistant

Every client uses the same URL: https://aiunivid.com/api/mcp. How the sign-in starts depends on the client, as shown for each one below; it opens an AiuniVid page where you sign in and click Allow. Each authorization becomes a connection you can revoke on its own later.

Claude (claude.ai and the desktop app)

• In Claude (claude.ai or the desktop app), open Settings › Connectors and choose Add custom connector.
• Name it AiuniVid and paste this URL:
• Add it, click Connect, sign in to AiuniVid and click Allow. Turn the connector on in a chat to use it.

ChatGPT

• In ChatGPT, open Settings › Apps & Connectors › Advanced and turn on Developer mode.
• Create a connector named AiuniVid with this MCP server URL and OAuth authentication:
• Save, click Connect, sign in to AiuniVid and click Allow. Pick the connector from the + menu in a chat.

Claude Code

claude mcp add --transport http --scope user aiunivid https://aiunivid.com/api/mcp

Then: In Claude Code, type /mcp, pick aiunivid and choose Authenticate.

Cursor

{
  "mcpServers": {
    "aiunivid": { "url": "https://aiunivid.com/api/mcp" }
  }
}

Then: In Cursor Settings › MCP, click Needs login next to aiunivid.

VS Code with GitHub Copilot

{
  "servers": {
    "aiunivid": { "type": "http", "url": "https://aiunivid.com/api/mcp" }
  }
}

Then: Start aiunivid from the MCP servers list.

Codex, Gemini CLI, OpenCode and other agents are covered, with a copy-and-paste setup prompt for each, in the MCP section of the API docs.

What a request looks like

You do not call tools by name; you describe the clip. The assistant turns that into a quote, asks you, and only then creates the job. A typical exchange:

You: Make a 5-second vertical 9:16 clip with Seedance 2.0 Fast at 720p:
     a barista slides a latte across the counter, slow push-in, warm morning light.

Assistant (quote_video): That will cost 79 credits (about $0.79). Go ahead?

You: Yes.

Assistant (create_video, then get_video every 20 seconds or so):
     Done. Here is the download link: https://…

Generation usually takes one to five minutes. For paid calls the assistant can pass its own idempotency key; if the connection then drops, retrying with that same key does not charge you twice.

What it costs

MCP uses your AiuniVid credits at the same rates as the website and the API. One credit is $0.01, credits come in packs with no subscription, and quoting is free. The cheapest 5-second clip is 29 credits on Seedance 2.0 Mini at 480p. A 5-second 1080p clip on Seedance 2.5 is 455 credits. The table below is for clips without input media; a reference video is billed for its own seconds as well. Full rates are on the pricing page.

For 1080p on Seedance 2.5, ask the assistant for a draft first: it renders a 480p preview, and render_final turns the take you keep into the 1080p final. See how Seedance 2.5 draft mode works.

Hosted server or your own ModelArk key?

Search for a Seedance MCP server and you will also find open-source servers that you install and run with your own BytePlus ModelArk API key. Both approaches work; they suit different people.

If you already have a ModelArk account and run large volumes, your own key is cheaper per second. If you want to try Seedance from a chat today, or share access without handing out a cloud key, the hosted server is simpler.

Limits and safety

• Real faces are refused. Uploaded media with a recognizable real person is rejected by the upstream model. Ask the assistant to create the character with generate_image first, or use a verified portrait from list_characters; keeping one character across a series is covered in consistent characters in Seedance.
• External media must be public HTTPS URLs. The assistant cannot upload a file from your disk through MCP; host it first, or start from an image generated with generate_image. Saved characters and verified portraits from list_characters are passed by their asset:// reference instead.
• Rate limits match the Open API: 10 create requests and 60 read requests per minute.
• No browser? On a server or in CI, send an API key instead of using OAuth:

claude mcp add --transport http aiunivid https://aiunivid.com/api/mcp \
  --header "Authorization: Bearer $AIUNIVID_API_KEY"

Frequently asked questions