memory-mesh has 5 tools with basic descriptions and visible schemas via FastMCP decorators. However, descriptions are minimal (10-60 chars, well below the 194-char baseline), parameter descriptions are sparse or missing, and output schemas are not formally documented. No error handling guidance, no enum constraints on status parameters, and no security considerations for destructive operations. Tool naming is reasonable (verb_noun pattern), but parameter documentation is insufficient for LLM reasoning.

• Expand tool descriptions to 100-200 chars. Example: 'save_context' → 'Save a memory entry with optional tags and session context. Detects semantic conflicts with existing memories and returns both the saved entry and any conflicts for resolution. Use this to persist facts, decisions, or context that should be available across agent sessions.'
• Add enum constraints to status parameters: list_conflicts(status: Literal['pending', 'resolved'] = 'pending'). Document in description: 'Filter by status: pending (unresolved conflicts) or resolved (already decided).'
• Document output schemas in docstrings. Example: 'Returns: {"memory": {"id": str, "content": str, "agent_id": str, ...}, "conflicts": [{"id": str, "entries": [...]}]}'
• Add error handling guidance. Example for delete_context: 'Raises ValueError if memory_id not found. Returns {"success": true} on success, {"success": false, "error": "Memory not found"} on failure. If deletion fails, verify the ID with search_context first.'
• Implement confirmation pattern for destructive ops. Add optional 'confirm=false' param to delete_context; when false, return {"preview": {...}, "confirm_required": true}. Require explicit confirm=true to execute deletion.