Single tool with basic schema and description. Tool name follows verb_noun pattern (get_user_courses). Description is present (73 chars) but lacks context on when to use it vs alternatives, prerequisites, or what the return structure contains. Input schema uses Zod with type definitions and descriptions for all three parameters, which is good. However, output schema is not documented, LLMs cannot predict the response structure. Error handling returns plain text errors without recovery guidance. No tool annotations (readOnlyHint, etc.). Overall: functional but below production baseline (194 char avg description, 100% A+ tools document output).

• Document the complete output schema in the tool description or as a structured response example. Include field types and meanings: 'Returns array of {id: number, name: string, code: string, state: string, needsGradingCount?: number, enrollmentState?: string}'.
• Expand tool description to 150-250 chars. Add: prerequisites (environment variables), use case (discovering user's courses before fetching assignments), and what data is returned.
• Convert enrollment_state parameter to an enum with valid Canvas values: ['active', 'invited', 'completed', 'rejected']. Remove the free-form string.
• Enhance error messages with recovery guidance. Catch specific HTTP status codes (401, 403, 404, 500) and return actionable messages: 'User not found. Verify user_id is correct. Call search_users() if you only have a name.'
• Add readOnlyHint annotation to the tool definition to signal to agents that the call is safe to retry without side effects.