Sibyl Memory MCP has well-structured tool definitions with consistent naming (verb_noun pattern: memory_remember, memory_recall, memory_search, etc.) and complete input schemas. All 8 tools have descriptions and typed parameters with descriptions. However, descriptions are terse (10-50 chars, well below the 50-200 char LLM-optimized baseline), lack actionable context about when to use each tool, and omit error recovery guidance. Output schemas are not documented. Parameter descriptions are minimal (e.g., 'entity category' for category param). No tool annotations (readOnlyHint, destructiveHint) despite clear risk classifications. Error handling is present but not visible in schema definitions.

• Expand tool descriptions to 50-200 chars with WHEN/WHY context. Example: 'memory_remember: Store a long-lived entity (user profile, project context) in a tiered storage system. Use for facts that persist across conversations. Specify tier (HOT=immediate access, COLD=archive) based on access frequency.'
• Document output schemas for all tools. Example: 'memory_search returns {results: [{category, name, body, tier, created_at}], total: int, next_cursor?: string}'
• Add tool annotations via MCP schema extensions. Mark memory_recall, memory_search, memory_list, memory_get_state with readOnlyHint=true; mark memory_remember, memory_set_state, memory_record_event with destructiveHint=true; mark memory_forget with idempotentHint=true.
• Enhance parameter descriptions with format/constraint guidance. Example: 'category: entity category (alphanumeric, 1-50 chars, e.g. "user_profile")' and 'tier: storage tier, HOT for frequent access, COLD for archival'.
• Document pagination for memory_list and memory_search. Specify: 'Returns up to {limit} results. If total > limit, include next_cursor for pagination. Default limit=50 to prevent context window exhaustion.'