The calculator_server.py example demonstrates solid foundational tool definition quality. All 6 tools have explicit names, descriptions, and properly typed input schemas. Naming follows verb_noun convention (add, subtract, multiply, divide, power, square_root). Descriptions are concise and action-oriented (10-50 chars), meeting the baseline for clarity. All parameters have type annotations (float) and descriptions. However, output schemas are not explicitly documented, the code implies numeric returns but lacks formal schema documentation in the tool definitions. Error handling is present in divide() and square_root() (ValueError for division by zero and negative square root), but error messages lack LLM-actionable recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are declared despite all tools being READ_ONLY and idempotent. The server is an evaluation/testing framework itself (pymcpevals), not a production tool server, which explains the minimal example scope.

• Add explicit output schema documentation to each tool. E.g., for add: 'Output: {type: number, description: "Sum of a and b (float)"}'. This enables downstream tool planning.
• Declare tool annotations via fastmcp's tool decorator or metadata. E.g., @server.tool(hints={'readOnlyHint': True, 'idempotentHint': True}). All 6 tools should declare idempotentHint=True.
• Enhance error messages with actionable recovery steps. Replace 'Cannot divide by zero' with 'Cannot divide by zero. Ensure the divisor (b parameter) is not zero. Try another denominator value.'
• E.g., 'Exponent (float): Power to raise the base to. Use positive for multiplication, negative for division, fractional for roots.'
• Consider adding parameter constraints via JSON Schema. E.g., for square_root's 'n' parameter: {minimum: 0, description: 'Number to calculate square root of (must be >= 0)'}.