MCP server for Atlas Cloud AI image and video generation using Seedance models. Supports text-to-image, text-to-video, image-to-video, and reference-to-video generation with storyboard grid splitting and audio integration.
7 tools with schemas present but significant quality gaps. Tool names are action-oriented (generateImage, generateVideo, pollPrediction, downloadMedia, uploadMedia, splitGridImage, verifyModels) but descriptions vary widely in completeness. Most parameter descriptions are present but lack constraint details (ranges, enums, formats). Output schemas are not documented. Error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint). Risk classifications exist but not integrated into descriptions. Average tool score: 42/100.
Download generated media (image or video) from a URL to a local file path
Generate an image using the configured image model with a text prompt and optional size parameter
Generate a video using the configured video model with text prompt, optional input image, duration, resolution, aspect ratio, and audio references
Poll the status of an image or video generation prediction until completion or timeout
Split an N x M storyboard grid image into individual numbered frames in reading order (left-to-right, top-to-bottom)
Upload local media files (images or audio) to Atlas Cloud for use as input references
Verify that configured model IDs are available in Atlas Cloud's live catalog
No output schemas documented. LLMs cannot predict what fields generateImage, generateVideo, pollPrediction, downloadMedia, uploadMedia return. Breaks tool chaining and forces agents to guess field names.
Missing constraint documentation. Parameters like 'size' (generateImage), 'duration' (generateVideo), 'rows'/'cols' (splitGridImage) lack min/max bounds, format specs, or enum values. LLMs will pass invalid values.
No error recovery guidance. Tools like pollPrediction (timeout), downloadMedia (network failure), uploadMedia (file not found) lack actionable error messages. LLMs cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
No tool annotations. Risk classifications (WRITE, READ_ONLY) exist in metadata but are not exposed as tool annotations (destructiveHint, readOnlyHint, idempotentHint). MCP clients cannot infer safety properties.
Descriptions lack WHEN-to-use guidance. generateImage vs generateVideo distinction is clear, but pollPrediction, downloadMedia, uploadMedia lack context on when agents should call them vs alternatives. Incomplete docstrings per pattern:tool-description.