Server has 12 tools with schemas and descriptions, but quality is inconsistent. Descriptions are present but often generic or overly technical (Indonesian language, domain jargon). Parameter descriptions lack actionable constraints (ranges, formats, enums). No error handling guidance visible. Output schemas are not documented. Tools combine multiple concerns (e.g., curate_youtube_clips does analysis + segmentation + layout recommendation). No tool annotations (readOnlyHint/destructiveHint) despite clear read/write distinctions. Naming is generally acceptable but some tools are compound (e.g., 'fetch_and_trim_clip' combines download + extraction).

• Split 'fetch_and_trim_clip' into 'download_youtube_video' + 'trim_video_segment'. Split 'analyze_cover_spatial_and_hooks' into 'analyze_cover_spatial_composition' + 'analyze_cover_visual_hooks'. Each tool should have one clear responsibility.
• Add numeric constraints to all numeric parameters. E.g., 'scene_threshold (0.0 - 100.0, default 27): ContentDetector sensitivity. Lower = more scenes detected.' Document units (seconds, pixels, percentage) explicitly.
• Document output schemas for all tools. E.g., curate_youtube_clips returns {candidates: [{start_sec, end_sec, layout_9_16, sub_scenes}], ...}. Include field types and descriptions so LLMs know what to extract.
• Add error handling guidance to tools calling external services. E.g., 'If Gemini API fails: check api_key, retry with exponential backoff, or fall back to heuristic scene detection.' Categorize errors as retryable vs user-fixable.
• Add tool annotations to all tool definitions. Mark WRITE tools with destructiveHint=true (render_adaptive_video, burn_captions_to_video, etc.). Mark READ_ONLY tools with readOnlyHint=true (curate_youtube_clips, analyze_scene_composition, etc.). This enables agents to plan safely.