Bilibili MCP server has well-structured tool definitions with complete JSON schemas and consistent naming patterns. All 10 tools follow verb_noun convention (bilibili_*). Descriptions are present and adequate (avg ~80 chars), though some lack actionable context for LLM selection. Parameter schemas are properly typed with enums, ranges, and constraints. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is absent from visible code. No tool annotations (readOnlyHint, destructiveHint) despite all tools being read-only. Missing descriptions for some enum values (e.g., duration filter meanings are terse).

• Add output schema documentation for all tools. Example: 'Returns: {videos: [{bvid, title, author, views, duration, pubdate}], total: int, page: int}'. This enables LLMs to chain tools and extract needed fields.
• Add readOnlyHint: true to all tool definitions in the Tool object. This signals to agents that these tools are safe to call without confirmation.
• Expand tool descriptions to 100-150 chars with actionable context. E.g., 'Get trending videos on Bilibili. Use this to discover popular content; use bilibili_search for keyword-based lookup. Returns paginated results (default 20 per page).'
• Implement error handling that returns structured errors with recovery guidance. E.g., 'User UID 999999999 not found. Try bilibili_search with a username to find the correct UID.' instead of raw HTTP errors.
• Document enum values in parameter descriptions. E.g., for duration: 'Duration filter (0=all, 1=<10min, 2=10-30min, 3=30-60min, 4=>60min)' so LLMs understand all options without reading schema.

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

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