MCP server for image and video generation
Server has 6 tools with explicit schema registration in src/index.ts. All tools have input schemas and descriptions, but there are significant quality issues: (1) Tool naming is verb-based but two tools contain 'and' patterns (batch_generate_images combines multiple concerns), (2) Descriptions are minimal or in mixed languages (batch_generate_images is entirely in Chinese), (3) Parameter descriptions exist but lack guidance on constraints, formats, and use cases, (4) No output schema documentation for any tool, (5) No error handling guidance in descriptions, (6) Missing actionable context for LLM decision-making. Per-tool analysis: generate_image (naming: 75, description: 60, schema: 65, overall: 67); batch_generate_images (naming: 50, description: 20, schema: 70, overall: 47, Chinese description fails English-speaking LLM optimization); generate_video (naming: 75, description: 65, schema: 70, overall: 70); query_video_result (naming: 80, description: 55, schema: 60, overall: 65); wait_for_video (naming: 75, description: 60, schema: 65, overall: 67); configure_models (naming: 75, description: 35, schema: 60, overall: 57, lacks explanation of why/when to use, side effects). Average: 52.
批量生成多张图像,支持并行处理和批次管理
Configure default models and settings
Generate images using BigModel CogView models
Generate videos using BigModel CogVideoX models
Query the result of an asynchronous video generation task
Wait for video generation to complete and return the result
batch_generate_images description is entirely in Chinese ('批量生成多张图像,支持并行处理和批次管理'). English-speaking LLMs cannot infer intent or use the tool effectively. Tool naming alone is insufficient for proper tool selection.
No output schema documented for any of the 6 tools. LLMs cannot predict the structure of responses, making downstream chaining, field extraction, and error handling unreliable. Users of these tools cannot know what fields to expect.
configure_models lacks explanation of when/why to call it or what side effects it has. Description is only 35 chars ('Configure default models and settings'). Does it persist across sessions? Does it affect only current execution? Is it required before other tools? No guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Error handling guidance missing from all descriptions. No indication of retryability, expected failure modes, or recovery steps. E.g., generate_video and query_video_result interact (async task pattern), but descriptions do not explain this relationship or what to do if query_video_result returns 'task not found'.
Parameter descriptions lack actionable constraint guidance. E.g., 'model' enum is listed but description for batch_generate_images says '使用的模型' (in Chinese). 'size' parameter descriptions say 'e.g.' instead of formal regex/format constraints. 'delay_between_batches' has no explanation of why/when to adjust it.
batch_generate_images combines multiple concerns: batch management, parallelization, and delay handling. This violates single-responsibility. Should split into separate tools or at minimum clarify when each parameter (parallel, max_concurrent, delay_between_batches) applies.
No dry-run, confirmation, or safety mechanism for destructive/expensive operations. generate_image and generate_video call external APIs that may incur costs. No way to preview cost or confirm before execution. Agents cannot safely explore options.
query_video_result and wait_for_video both accept task_id and query async task state, but descriptions do not clarify when to use which, or explain polling semantics, timeout behavior, or what 'task not found' means.