video-context-mcp has 11 tools with clear verb-noun naming (ingest_video, get_transcript, search_videos, etc.). All tools have descriptions (avg ~80 chars, within baseline 34-392). Input schemas are present and typed with Zod validation. However, parameter descriptions are sparse or missing in several tools (e.g., crop_region lacks description for x/y/width/height semantics), and output schemas are not documented in the source. Error handling is basic, no recovery guidance or actionable error messages visible. Tool composition is sound (single responsibility), but descriptions could be more LLM-optimized per pattern:tool-description (10-1024 chars, currently many are ~50-80 chars, which is on the short side for complex tools like ingest_video).

• Document output schemas for all 11 tools. Include field names, types, and descriptions so LLMs know what to expect. Example: ingest_video should document {video_id, status, transcript_segments, frame_count, ocr_blocks, embedding_count}.
• Expand tool descriptions to 100-200 chars per pattern:tool-description baseline. Add context: 'Use peek_frame to view a single keyframe at a timestamp; use crop_region to extract a specific region with OCR. Use search_videos for semantic search across all videos.'
• Add parameter descriptions to crop_region: 'x, y, width, height are normalized coordinates (0.0-1.0) where (0,0) is top-left and (1,1) is bottom-right. E.g., x=0.1, y=0.2, width=0.3, height=0.4 extracts a 30%×40% region starting 10% from left, 20% from top.'
• Implement error recovery guidance in tool handlers. Return structured errors like {error: 'video_not_found', message: 'Video ABC123 not found.', suggestion: 'Call list_videos() to see available videos or ingest_video() to add a new one.'}
• Add dry-run support to delete_video and prune_cache. E.g., delete_video should accept {video_id, dry_run: true} to preview what would be deleted before confirming. Prevents accidental data loss.