Four tools with complete schemas and descriptions, but quality is uneven. format_greeting and compute_stats have adequate descriptions (60-80 chars) and clear schemas. ShoppingCart tools lack depth: descriptions are terse (20-30 chars), and ShoppingCart_total_items has zero input parameters with no explanation of what it returns. Naming is inconsistent: ShoppingCart_add_item and ShoppingCart_total_items use PascalCase with underscores (non-standard verb_noun pattern). No output schemas documented. Error handling and recovery guidance absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in metadata.

• Rename ShoppingCart_add_item → add_item and ShoppingCart_total_items → get_cart_total to follow verb_noun convention. Prefix with action verb so LLMs parse intent from the name alone.
• Expand ShoppingCart_total_items description to 50+ chars: 'Return the total quantity of all items in the shopping cart. Returns an object with total_items (integer) and item_count (integer).' Document the output schema explicitly.
• Add tool annotations: set destructiveHint=true on ShoppingCart_add_item (modifies state), readOnlyHint=true on format_greeting and ShoppingCart_total_items. This guides agent planning and prevents accidental misuse.
• Document output schemas for all tools. E.g. format_greeting returns {greeting: string}; compute_stats returns {mean: number, min: number, max: number, count: integer}.
• Add error handling: compute_stats should validate input is non-empty and all elements are numeric. Return actionable error: 'numbers array is empty or contains non-numeric values. Provide a list of valid numbers.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry