Server has 5 tools with complete input schemas and descriptions. Naming follows verb_noun convention (generate_image, edit_image, get_image_task, generate_video, get_video_task). Descriptions are present and moderately detailed (100-200 chars), explaining what each tool does and when to call it. However, output schemas are NOT documented, LLMs cannot see what fields to expect from responses. Parameter descriptions are good but lack explicit constraints (e.g., no min/max for polling intervals, no guidance on retry behavior). Error handling is minimal, no recovery guidance or error classification. Security: API key is properly injected via environment variable, not exposed as a parameter.

• Document output schemas for all tools. For get_image_task, specify: {state: 'succeeded'|'error'|'pending', images?: [{url: string, width: number, height: number}], error_message?: string}. For get_video_task: {state, progress: number, video_url?: string, error_message?: string}.
• Add explicit error handling and recovery guidance. E.g., 'If task_id is invalid, returns error_code=404. If polling times out after 10 retries, returns error_code=timeout, retry with exponential backoff or check task status manually via the API.'
• Enforce array length constraints in schema. For edit_image.image_urls, add maxItems: 10 to the schema definition. For generate_video.image_urls in REFERENCE mode, add maxItems: 3.
• Clarify auto_download timeout behavior. Add to description: 'If auto_download=true and polling exceeds 5 minutes, returns partial result with task_id for manual polling. Does not hang indefinitely.'
• Add per-tool permission/scope declarations. E.g., 'Requires DUOMI_API_KEY with image-generation scope.' This enables least-privilege agent configuration and audit trails.