The Kubrick MCP server provides 4 tools with basic schemas and descriptions. However, definitions have significant quality gaps: descriptions are generic and lack LLM-optimized guidance; parameter descriptions are minimal; output schemas are not documented; error handling guidance is absent; and tool naming could be clearer. The server is functional but falls short of production-grade quality. Average per-tool score: 52.

• Document output schemas for all tools. For process_video: return {job_id: string, status: 'processing'|'ready'|'error', estimated_duration_ms?: number}. For get_video_clip_from_*: return {clip_path: string, start_frame: number, end_frame: number, duration_seconds: number, metadata: object}. For ask_question_about_video: return {answer: string, confidence: 0-1, relevant_timestamps: [{start_sec: number, end_sec: number, caption: string}], sources: string[]}.
• Expand all descriptions to 150-300 chars using the pattern: '<action>. Use this when <condition>. Returns <output_type>. Note: <prerequisite or limitation>.' Example for get_video_clip_from_user_query: 'Search a video for scenes matching a text query using semantic video understanding. Use this to find moments relevant to a user question without knowing timestamps. Returns clip metadata with frame range and duration. Note: requires process_video() first.'
• Add format constraints to all path parameters. Example: 'video_path (string): Absolute or relative path to video file. Supported: .mp4, .mov, .avi, .mkv. Max size 2GB. Example: /data/video.mp4 or ./videos/sample.mov'
• Document parameter relationships in descriptions. Clarify: Does process_video() need to run before the search tools? Can search tools be called in any order? If process_video has a job_id output, can search tools accept job_id as an alternative to video_path?
• Add error guidance to all descriptions. Example: 'If the video file is not found, the tool returns {error: "FILE_NOT_FOUND", hint: "Check the path and ensure the file exists"}. If a query returns no results, try a broader or more specific query, or verify the video contains relevant content.'