ImgLume MCP demonstrates solid tool design with clear naming, comprehensive parameter schemas, and good descriptions. All 6 tools follow verb_noun conventions (search_gallery, get_inspiration, enhance_prompt, list_models, generate_image, generate_video). Schemas are well-structured using Zod with proper type constraints, enums, and ranges. Descriptions are substantive (100-200 chars typical) and explain WHEN to use each tool. Error handling is present with custom ImgLumeApiError class and actionable messages. Main gaps: output schemas are not formally documented in tool definitions (only visible in implementation), some parameter descriptions could be more prescriptive about constraints, and no explicit idempotency or retry guidance for generation tools.

• Add formal output schema documentation to each tool definition. Document fields, types, and pagination structure so LLMs can plan downstream calls. Example: 'Returns {items: [{id, previewUrl, promptPreview, ...}], page, totalPages, totalItems}'.
• Add explicit idempotency statement to generate_image and generate_video: 'Calling with identical prompt and model within 24 hours may return cached result; no duplicate charge.' This guides agent retry logic.
• Enhance referenceImages parameter description with format constraints: 'Local file paths (JPEG, PNG, WebP, max 10MB) or HTTPS URLs. Relative paths resolved from current working directory.'
• Document search_gallery result limits and pagination strategy in description: 'Returns up to 20 results per page. Use page parameter to iterate; limit parameter controls per-page count (1-20, default 5).'
• Improve error messages for upload/generation failures with actionable recovery hints. Example: 'Reference image upload failed: unsupported format. Try JPEG or PNG.' instead of generic error.