Server has 8 tools with complete JSON schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (save_, retrieve_, update_, delete_, list_, search_, get_, clear_). Descriptions are present (avg ~100 chars) but lack actionable context for LLM selection. Parameters have types and descriptions, but many lack constraints (enums, ranges, patterns). No output schemas documented. Error handling is minimal, no recovery guidance or categorization. Destructive operations (delete_memory, clear_memories) lack confirmation patterns. Vector embedding logic is present but undocumented in tool descriptions.

• Add output schemas to all tool descriptions. Example: 'Returns: {id: number, content: string, category: string, tags: string[], created_at: ISO8601, updated_at: ISO8601}'
• Add min/max constraints to numeric parameters: limit (1-100, default 10), offset (0+, default 0), threshold (0.0-1.0, default 0.5)
• Implement confirmation pattern for destructive tools: require explicit confirm=true flag and return a confirmation_id before executing delete_memory or clear_memories
• Add error handling with recovery guidance: 'Memory not found (ID: 123). Try list_memories() to see available IDs.' Include error categories (not_found, invalid_input, server_error)
• Document embedding behavior in tool descriptions: 'When embedding=true, saves a vector representation for semantic_search. When false, only keyword search works. Default: true.'