Three tools with complete schemas and descriptions, but significant quality gaps. Tool names lack action verbs (calculate, advanced_math, expression_eval are weak). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameters have types and descriptions, but lack constraints and validation guidance. No output schema documented. Error handling is minimal, errors return plain text without recovery guidance. expression_eval has a critical security flaw (unsafe eval). No tool annotations (readOnlyHint, destructiveHint). Composition is reasonable but naming could be clearer.

• Rename tools with action verbs: 'calculate' → 'compute_arithmetic', 'advanced_math' → 'compute_trigonometric', 'expression_eval' → 'evaluate_expression'. Follow verb_noun pattern to improve LLM intent inference.
• Replace unsafe Function() eval in expression_eval with a safe math expression parser (e.g., math.js or expr-eval library). Validate input against a whitelist of allowed functions and operators.
• Document output schemas explicitly. E.g., calculate returns {type: 'object', properties: {result: {type: 'number'}, expression: {type: 'string'}}}. Include in tool registration.
• Enhance error messages with recovery guidance. E.g., 'Division by zero error: divisor (b) must be non-zero. Retry with a different value.' Include the invalid value and constraint violated.
• Add tool annotations to all three tools: set readOnlyHint: true since all are read-only and safe to retry. This signals to agents that these calls are idempotent.