Server has 7 tools with basic Zod schemas and descriptions, but significant quality gaps. All tools have descriptions (10-60 chars, mostly short), and input parameters are typed via Zod. However, descriptions are generic and lack actionable context for LLM selection. No output schemas documented. Parameter descriptions are minimal. Error handling is present but not recovery-focused. No tool annotations (readOnlyHint/destructiveHint). Tool names follow verb_noun pattern but lack clarity on side effects. Average per-tool score: 42/100.

• Expand tool descriptions to 100-200 chars. Example: 'Swap faces between two images using AI. Requires clear face in both images. Returns processed image URL and task ID. Use for creative effects, not identity fraud.'
• Document output schema explicitly. Define return type as {taskId: string, usage: number, imageUrls: string[]} so LLMs know what fields to extract.
• Add tool annotations: mark all WRITE tools with destructiveHint=true, mark show_image with readOnlyHint=true. This signals retry safety and side effects.
• Enhance parameter descriptions with constraints. Example: 'URL of image to swap (JPEG/PNG, max 10MB, min 256x256px, clear face required)'.
• Improve error messages with recovery guidance. Instead of 'Image URL is required', return 'Image URL is required. Provide a direct HTTPS URL to a JPEG or PNG image (max 10MB).'.

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

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