Server has 17 tools with basic schemas and descriptions, but significant quality gaps. Most tool descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameter descriptions exist but lack constraints, ranges, and format guidance. No output schemas documented. Error handling returns status/error fields but lacks recovery guidance. Tool naming is mostly verb-noun compliant, but several tools lack specificity (e.g., 'remember'/'recall' are vague; 'forecast' vs 'weather_info' overlap). No tool annotations (readOnlyHint/destructiveHint). Security concerns: calculate_expression uses eval() with restricted globals but still risky; no secret injection pattern visible. Composition issues: memory tools (remember/recall/list_memory_keys/clear_memory) are overly granular; clear_memory is destructive but lacks confirmation pattern.

• Add output schema documentation to all tools. Specify return type, field names, and types. E.g., create_customer returns {id: string, name: string, email: string, phone: string, status: string}.
• Replace eval() in calculate_expression with ast.literal_eval or a dedicated math parser (sympy). Document allowed functions and operators in the description.
• Consolidate memory tools: merge remember/recall into store_memory(key, value) and retrieve_memory(key); merge list_memory_keys/clear_memory into manage_memory(action: 'list'|'clear', key?: string). Reduces cognitive load and tool count.
• Add pagination to list_customers: accept offset/limit, return {customers: [...], total: int, offset: int, limit: int}.
• Add confirmation pattern to clear_memory: require a dry_run parameter or a separate confirm_clear_memory(key) tool. Prevent accidental data loss.

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

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