Seedance MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| SEEDANCE_API_KEY | Yes | Your provider's dk_ API key, required for authenticated operations. Can also be set in the credential file specified by SEEDANCE_CONFIG_FILE. | |
| SEEDANCE_BASE_URL | No | The API origin. HTTPS is required outside loopback. | https://sd2.bantakal.com |
| SEEDANCE_CONFIG_FILE | No | Override the credential file location. Environment variables take precedence over the file. | ~/.config/seedance-mcp/.env |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| seedance_healthA | Read-only connectivity check; returns API health, whether a dk_ API key is configured, and the local config file path. Does not authenticate or consume credits. A configured key is not proof it is valid; use seedance_list_models to verify authentication. |
| seedance_list_modelsA | List currently enabled video models, names and credit prices. Authenticated read-only request. Call before generation; disabled models return 404. Returns the provider data array. |
| seedance_upload_imageA | Upload a local PNG/JPEG/WEBP image (maximum 5 MB) as a reference asset. Sends the image to the configured Seedance API and returns its asset id, mime and byte count. Use IDs in reference_asset_ids in the same order as @Image1, @Image2 in your prompt. |
| seedance_create_videoA | Create ONE asynchronous video task, consuming account credits (documented Seedance 2.0 price: 20 credits; confirm current price via list_models). Returns provider task id, statusTask, cost and other metadata immediately. Explicitly defaults to seedance-2.0 / 10s / 9:16 / 720p. Never automatically retry submission on network failure: it may have succeeded. Poll with get_video every 10–20 seconds; failed/interrupted tasks are refunded by the provider. |
| seedance_get_videoA | Query task metadata and statusTask: queued/submitted/running/succeeded/failed/interrupted. Poll no faster than once every 10–20 seconds. Terminal states are succeeded, failed and interrupted. A signed result URL may expire; query again to refresh it. Download completed video within 24 hours for retention. |
| seedance_download_videoA | Download a succeeded task as an MP4 to an absolute local output path. Returns path, bytes and mime. Creates parent directories; refuses to overwrite an existing file. Streams at most 512 MB with a 90-second network deadline. Partial files are removed on failure. Uses the authenticated file endpoint, with no key forwarding across redirect origins. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 6 tools
Each tool targets a distinct stage or resource: health/connectivity, model catalog, image upload, task creation, task polling, and file download. The only potentially overlapping pair (health vs. list_models) is explicitly differentiated in the descriptions, with list_models named as the way to verify authentication.
All names use a consistent seedance_ prefix with verb_noun patterns (list_models, upload_image, get_video, create_video, download_video). The lone deviation is seedance_health, which has no verb and slightly breaks the pattern, though it remains readable and unambiguous.
Six tools is well-scoped for an asynchronous video generation service: one per lifecycle stage with no redundant or filler tools. Nothing appears missing that would require a separate tool beyond optional extras like cancellation.
The full workflow is covered: connectivity check, model/price discovery, reference asset upload, task creation, status polling, and MP4 download with clear terminal-state handling. Minor gaps exist — no cancel task, task listing, or asset management — but agents can work around these using create/get/download.