mcp-starter is a teaching example with basic tool definitions. All three tools have explicit schemas and descriptions, but descriptions are minimal (10-30 chars), parameter descriptions lack detail, and output schemas are not documented. No error handling guidance, no parameter constraints (enums), and no indication of which tools modify state. Naming is clear (verb_noun pattern), but the server lacks production-grade polish. Suitable for learning; not for production use.

• Expand tool descriptions to 50 - 150 characters. Example: 'add_note: Create and persist a new note with a title and body. Returns the note ID and confirmation. Use this to capture user ideas or tasks.' This tells LLMs what the tool does, when to use it, and what to expect.
• Add parameter descriptions with constraints. Example: title: 'Short title (1 - 100 chars, required)' and body: 'Note content (1 - 5000 chars, required)'. Include format hints if applicable.
• Document output schemas explicitly. Example: add_note returns {id: integer, title: string, body: string}. list_notes returns [{id, title}]. delete_note returns {success: boolean, message: string}.
• Add destructiveHint annotation to delete_note. Use types.Tool(..., destructiveHint=True) to signal to LLMs that this operation is irreversible and should be confirmed.
• Implement error classification. Return structured errors: {error: 'not_found', message: 'No note with id #X', recovery: 'Call list_notes() to see available IDs'}. This guides LLM recovery logic.