Discogs MCP has well-structured tool definitions with complete schemas and reasonable descriptions. All 5 tools follow verb_noun naming (search, get_*). Schemas are properly typed with required fields declared. However, descriptions are generic and lack LLM-optimization guidance (e.g., no 'when to use' context, no dependency hints). Parameter descriptions are minimal (1-5 words). Output schemas are not documented, callers cannot predict response structure. Error handling is basic (no recovery guidance). The server lacks pagination guidance for search results despite returning potentially large datasets.

• Document output schemas for all tools. E.g., 'Returns: {id: number, title: string, artists: [{name: string, id: number}], formats: string[], year: number, ...}'. Include field types and nesting structure.
• Expand tool descriptions to 50-200 characters with LLM-optimization guidance. E.g., 'search: Find releases, masters, artists, or labels by title, artist, format, year, or genre. Use this first if you only have a name. Returns paginated results (default 25 per page, max 100).'
• Enhance parameter descriptions with format/constraint hints. E.g., 'per_page: Results per page (1-100, default 25). Larger values increase response size; use 25-50 for typical queries.'
• Add recovery guidance to error messages. E.g., 'Discogs: not found (HTTP 404). Try search() with a partial title or artist name to find the correct ID.'
• Migrate API key from _apiKey parameter to server-side environment variable (DISCOGS_TOKEN). Update callTool() to read from process.env instead of args.