This server has 17 tools with significant definition gaps. Most tools have basic descriptions (10-100 chars, well below the 194-char baseline), but critically, input schemas are inferred from code rather than explicitly visible in schema definitions. Parameters lack type constraints and enums. No output schemas are documented. Error handling is absent, tools return raw results without recovery guidance. The server mixes read-only and write operations without clear separation or confirmation patterns. Tool names are reasonable (verb_noun style) but descriptions are too terse to guide LLM selection.

• Add explicit JSON Schema definitions for all tool inputs with type, enum, minLength, maxLength, and pattern constraints. E.g., 'action' in record_preference_feedback should be enum: ['confirm', 'reject', 'correct'].
• Expand descriptions to 100-200 chars. Include WHEN to call (e.g., 'Call after capturing preferences to resolve conflicts between overlapping rules') and WHAT it returns (e.g., 'Returns resolved preference with merged text and conflict ID').
• Document output schemas for all tools. Specify returned fields, types, and which IDs/references are included for chaining (e.g., preference_id, agent, session_id).
• Add dry-run or confirmation parameters to destructive tools (apply_fitting_plan, reject_fitting_plan). Return a preview of changes before committing.
• Rename hook_* tools to clarify intent. If they are internal callbacks, move them to a separate 'hooks' capability. If they are agent-callable, rename to inject_session_preferences, capture_user_message, etc.