Server has 8 well-defined read-only tools with reasonable naming conventions (all start with 'get_' or 'search_'), and most have descriptions between 50-150 characters. Input schemas are present with type definitions and descriptions for parameters. However, there are significant gaps: output schemas are completely undocumented (no visibility into what these tools return), parameter descriptions lack detail about expected formats and constraints, and error handling guidance is absent. The tools are query-focused with low risk, but lack the comprehensive documentation expected for production agent integration. No tool annotations (readOnlyHint) are present despite all tools being read-only operations.

• Add complete output schema documentation for all 8 tools. Specify field names, types, and descriptions. Example: get_mood should document it returns {zones: [{name: string, comfort: number 0 - 100, joy: number 0 - 100, frugality: number 0 - 100}], timestamp: ISO8601}
• For get_mood, add enum constraint to 'zone' parameter or document valid zone names in the description (e.g., 'Optional zone name. Valid zones: living_room, bedroom, kitchen')
• For get_brain_graph and get_habitus_patterns, specify numeric bounds on 'limit' parameters (e.g., 'limit: 1 - 500, default 20') and document what 'neighbors' or 'patterns' actually contain
• For get_preferences, document 'min_confidence' with clearer guidance: 'Minimum confidence score (0.0 - 1.0, default 0.3). Higher values return only high-confidence patterns; lower values include exploratory/uncertain preferences.'
• For search_memory, document expected format of 'query' parameter and what 'limit' applies to (e.g., max topics, max messages per topic)
• Add error handling section to each tool description. Example: 'Returns empty list if no matching patterns found. Returns error if min_confidence is outside 0 - 1 range.'
• Add readOnlyHint=true annotation to all 8 tools to signal they are safe idempotent queries
• Document pagination/limits: if any tool can return large result sets, add explicit note: 'Results capped at [limit] items to avoid context window exhaustion. Use limit parameter to request fewer results.'
• Add 'When to use' context to tool descriptions. Example for get_brain_graph: 'Use this to explore relationships between entities in learned patterns. Complements get_habitus_patterns (which shows rules) and search_memory (which shows past conversations).'