The server declares 2 tools in the spec (get_video_details, get_playlist_details) but the source code shows evidence of many more functions in the docstring ('14 functions including transcript extraction'). Only 2 tools are visible in the provided source excerpt. Both visible tools have minimal descriptions (under 100 chars) and very sparse parameter documentation. Input schemas are present but lack richness, parameters have type and description, but no enums, constraints, or validation guidance. No output schemas are documented. The code loads secrets from a credentials file rather than environment variables, violating secure credential handling. Error handling is minimal, no recovery guidance, no categorized errors, no user-fixable vs retryable classification. The tool names are reasonable action verbs ('get_*') but descriptions lack LLM-optimization details like when to use each tool and what structure they return.

• Expand tool descriptions to 100 - 200 characters. For get_video_details: 'Fetch detailed metadata for a YouTube video including title, description, duration, view count, upload date, channel, and like count. Use this to gather context about a video before extracting its transcript or analyzing its metadata.'
• Document output schemas. For get_video_details, explicitly list returned fields: {video_id (string), title (string), description (string), duration_seconds (integer), view_count (integer), upload_date (ISO 8601 string), channel_id (string), channel_name (string), like_count (integer)}.
• Add parameter format constraints. For video_input, document: 'Accepts YouTube URLs (youtube.com/watch?v=*, youtu.be/*) or 11-character video IDs matching [a-zA-Z0-9_-]{11}. LLMs should validate format before passing.'
• Move API_KEY loading to environment variables. Replace load_api_key() with: API_KEY = os.getenv('YOUTUBE_API_KEY') and raise error if not set. Remove credentials.yml file dependency.
• Add error handling with recovery hints. Wrap API calls in try/except blocks and return structured error responses: '{"error": "Video not found", "hint": "Check that the video ID is valid and the video is not private or deleted."}' instead of raising exceptions.
• Add tool annotations to declare read-only behavior. Use FastMCP's annotation system (e.g., @tool(...) with readOnlyHint=true) to signal that these tools are safe to retry and have no side effects.
• Verify and register all '14 functions' mentioned in the docstring. If they exist in the code, add them to the tool registration list. If not, remove the claim from the project description.