VidCap provides 8 YouTube-focused tools with consistent naming (verb_noun pattern: youtube_getInfo, youtube_getCaption, etc.) and basic descriptions. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; no recovery guidance or actionable error messages. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Parameter descriptions lack format constraints, ranges, or examples of valid values. The server lacks pagination guidance for list-returning tools (youtube_getComments, youtube_search). Overall: functional but below production baseline.

• Document output schemas for all 8 tools. Include field names, types, and sample structures. Example: youtube_getInfo returns {videoId, title, duration, uploadDate, channelId, viewCount, likeCount, description}.
• Enhance descriptions with LLM-optimized guidance: 'Use youtube_getInfo for metadata (title, duration, channel). Use youtube_getCaption for transcripts. Use youtube_getSummary for AI-generated summaries. Use youtube_getScreenshot for visual frames.' This disambiguates tool selection.
• Add format constraints to parameters. Example: 'timestamp: integer seconds (0-86400) or YouTube format (HH:MM:SS)'. 'model: one of [gpt-4, gpt-3.5-turbo, claude-3]'. 'ext: one of [json3, srv1, srv2, srv3, ttml, vtt]'.
• Add readOnlyHint annotation to all tools via MCP SDK. This signals to agents that all operations are safe to retry without side effects.
• Document pagination behavior for youtube_getComments and youtube_search: 'Returns up to maxResults items (default 20, max 100). Use pageToken to fetch next page. Returns total count if available.'

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

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