Server provides 2 tools with reasonable parameter schemas and descriptions, but has significant gaps in clarity, error handling guidance, and output documentation. Tool names follow verb_noun conventions (generate_image, edit_image) which is good. However, descriptions lack LLM-optimization: they are moderately detailed but do not clearly distinguish WHEN to use each tool vs alternatives, or what the response structure looks like. Parameters are well-typed with enums for size/quality, but descriptions are generic ('The text prompt...') without guidance on prompt engineering or length limits. Output schemas are not documented in the provided code, no indication of what fields are returned or how pagination/multiple images are handled. Error handling mentions logging but provides no recovery guidance (e.g., 'if API rate-limited, retry after X seconds'). The code shows helper functions for image validation and file saving, but these safeguards are not reflected in tool descriptions, so LLMs won't know input constraints exist.

• Add explicit output schema documentation in tool descriptions. State: 'Returns a JSON object with fields: images (array of objects with url, local_path, size_bytes), request_id, created_at. If n > 1, the images array contains multiple entries.' Include example response structure.
• Expand tool descriptions to 100 - 150 characters and add WHEN-to-use guidance. Example: 'generate_image: Create new images from text prompts using DALL-E 3. Use this when the user wants original images. If editing existing images, use edit_image instead.'
• Document 'prompt' parameter constraints in descriptions. Add: 'Prompt should be 1 - 4000 characters, descriptive (e.g., "golden retriever sitting in a sunny garden, oil painting style"). Vague prompts (e.g., "dog") produce lower-quality results.'
• Add dependency hints to parameter descriptions. For edit_image, change 'image_path' description to: 'Absolute file path to the PNG image to edit (e.g., /home/user/photo.png). If you only have a filename, call a file listing tool first to locate the full path. Maximum 25MB.'
• Implement error response templates that guide recovery. Catch validation errors early and return: '{"error": "Image validation failed: PNG required. Input was JPEG. Convert to PNG using PIL (Image.open(...).save(..., format="PNG")) and retry."}'
• Document pagination/batching behavior. If n=5 generates 5 images in one response, clarify in tool description and return schema whether they are in a flat array or nested. Add: 'Returns up to n images. No pagination needed; all requested images return in a single response.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry