This MCP server has inconsistent tool definition quality with several critical gaps. While some tools (analyze_video_from_url, transcribe_video) have well-structured schemas and reasonable descriptions, others lack sufficient parameter documentation or unclear naming conventions. The server defines 11 tools across two files (Video-Understander/src/mcp_server.py and mcp-server/mcp_sse_server.py), but code inspection reveals incomplete schema enforcement and missing error handling patterns. Tool descriptions range from adequate (194 char baseline met by most) to minimal. Critical issues: (1) Parameters like 'analysis_type' and 'summary_type' lack sufficient context about when to use each enum value; (2) Several tools (get_job_status, get_system_stats) have minimal descriptions that don't explain WHEN an LLM should call them; (3) No visible output schema documentation for most tools; (4) Error handling is absent from visible definitions.

• Expand all tool descriptions to 50-200 characters, following pattern:tool-description. Example: 'cancel_job' should be 'Cancel a running video analysis job. Immediately stops processing and releases resources. Partial results are discarded. Cannot be undone.'
• Document output schemas for all tools using JSON Schema. Include examples: 'Returns {transcription: string, segments: [{text, start_ms, end_ms, confidence}], language: string, duration_ms: integer}'.
• Add enum value descriptions in parameter docs. Replace 'Type of summary to generate' with 'Type of summary: brief (1-2 sentences), detailed (paragraph), bullet_points (list format), chapters (timestamped sections)'.
• Constrain free-form string parameters. For scene_criteria, either define an enum (e.g., scene_type in [action, dialogue, transition, closeup, wide_shot]) or add format guidance: 'Comma-separated scene descriptors or predefined types: action, dialogue, objects, people, text'.
• Add error handling with recovery guidance. Return structured errors: '{error: string, recovery: string, retryable: boolean}'. Example: 'Invalid URL. Use a direct video link or URL from YouTube, Instagram, or TikTok. Retry with a valid URL or call search_videos() first.'
• Implement confirmation flow for destructive operations (cancel_job). Return intermediate result 'job_status requires confirmation' with action: user must explicitly call cancel_job_confirm(job_id) to complete.
• Align enum definitions across similar tools. Choose one canonical set of analysis_types and use it consistently. Document each value's purpose and when to use it.

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

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