dj-mcp demonstrates solid fundamentals: all 4 tools have clear verb-noun naming (lookup_, get_, list_, add_), comprehensive descriptions (100-200 chars, well above 20-char minimum), and complete input schemas with type definitions and constraints. Pydantic validation is present. However, output schemas are not formally documented in the code, only inferred from ToolResult.json() calls. Error handling is basic (returns ToolResult.error() with messages) but lacks recovery guidance or error classification. No tool annotations (readOnlyHint/destructiveHint) despite clear read/write semantics. Tenant isolation is correctly enforced server-side, preventing agent bypass. Parameter descriptions are solid (e.g., 'exact' vs 'partial' matching, health_score_below threshold), but some lack format hints (e.g., date ranges, ISO 8601 expectations). Overall: good naming and descriptions, complete schemas, but missing output documentation and error recovery patterns.

• Document output schemas explicitly. For each tool, add a 'Returns' section in the docstring or a separate schema definition. Example: lookup_student returns {students: [{id, name, email, plan, mrr, health_score, status, enrollment_date}], row_count}.
• Add tool annotations to the schema. Mark add_student_note with destructiveHint=true and lookup_student/get_student_usage/list_at_risk_students with readOnlyHint=true. This signals to agents which tools are safe to retry.
• Enhance error messages with recovery guidance. Instead of 'Student not found in your account', return 'Student not found. Try lookup_student(query=partial_name) to search by name, or verify the student_id is correct.'
• Clarify parameter formats in descriptions. For 'days', add '(integer, 1-365)'. For 'note', specify 'Plain text, max 2000 chars; markdown not supported.' For dates, use ISO 8601 format in examples.
• Consider adding a batch_add_student_notes(student_ids: list[int], note: str, note_type: str) tool to support bulk operations without sequential calls.