This server has significant definition quality gaps. All 5 tools have basic descriptions (10-100 chars, mostly under 50), but lack depth on WHEN to use them and prerequisites. Input schemas are present but minimal: most parameters lack type constraints (enums, ranges). No output schemas documented anywhere. Parameters like 'hook_type' have enums (good), but 'keyword', 'hashtag', 'topic', 'niche' are bare strings with no format guidance. Error handling is generic ('Error: {message}') with no recovery guidance. Tool implementations return free-text instructions rather than structured data, forcing LLMs to parse unstructured responses. No tool distinguishes between free and premium tiers in a machine-readable way.

• Add output schemas to all tools. Define structured JSON responses with typed fields (e.g., {videos: [{title, views, likes, url}], total_count, next_cursor}). Return data, not instructions.
• Expand parameter descriptions to 50-150 chars each. Include format guidance: 'keyword (string, 1-100 chars, e.g. "#AI" or "marketing tips")'. Add constraints like min/max length or regex patterns.
• Implement proper secret injection for DEESEEK_API_KEY. Check for the key at startup, validate it, and return a clear error message if missing: 'DEESEEK_API_KEY not configured. Set it in your environment and restart the server.'
• Replace generic error handling with actionable messages. E.g., 'Hashtag not found. Try a different hashtag or use search_trending_videos to discover popular tags.' Categorize errors as retryable or user-fixable.
• Add tool selection guidance in descriptions. E.g., search_trending_videos: 'Use this to discover trending content by keyword. Returns video URLs and engagement metrics. Call analyze_hashtag next to deep-dive into a specific hashtag.'

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

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