Two tools with complete schemas and descriptions, but significant gaps in parameter clarity and output documentation. find_image has 7 parameters with descriptions, but several are vague (e.g., 'query' lacks format guidance; 'target' uses cryptic notation like 'aN, vN, iN, pN'). tag_image has good action enum but target/tags parameters accept anyOf string|array without clear guidance on when to use each. Neither tool documents output schema structure. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint). Descriptions are adequate (150-200 chars) but could be more LLM-optimized.

• Document target parameter format explicitly: 'One of: image ID (a1, v2, i3, p4), filename in scratchpad folder, or absolute path. Use image IDs from prior find_image results.'
• Add output schema documentation to find_image: specify fields in structuredContent.ranks array (e.g., rank, image_id, similarity_score, metadata).
• Clarify query syntax with examples: 'Search instruction. Examples: "red car", "Model: stable-diffusion, Tag: landscape". Hard filters (Model:, LoRAs:, Tag:, Size:, Source:, Origin:, Timestamp:) use AND logic.'
• Split tag_image target/tags into separate required/optional variants or add enum for action to clarify which params apply (e.g., 'list_tags requires scratchpadFolder only; add_tag requires target and tags').
• Add tool annotations via server.registerTool: find_image with readOnlyHint=true; tag_image with destructiveHint=true for remove_tag/remove_all_tags actions.

Inferred effective spec: 2025-06-18+.

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