Mnemo has 21 tools with consistent naming (verb_noun pattern: mem_*, skill_*, workflow_*, persona_*) and complete input schemas visible in the TypeScript plugin file. However, descriptions are uniformly terse (10-50 chars), falling well below the 50-200 char baseline for LLM-optimized tools. Parameter descriptions exist but are minimal ('Memory title', 'Search query'). No output schemas are documented, critical for tools like mem_search and mem_context that return structured data. Error handling and recovery guidance are absent. The server is HTTP-based (good for protocol readiness) but tool definitions appear inferred from a TypeScript plugin rather than directly registered in Go MCP code, capping individual tool scores at 50 per the rubric.

• Expand all tool descriptions to 80-150 characters. Include WHEN to use (e.g., 'Use mem_save to record a decision or bug fix for future reference. Use mem_capture_passive for automatic context from the session.'), prerequisites, and what the tool returns.
• Document output schemas for all tools. For mem_search, specify: returns array of { id: string, title: string, content: string, type: string, score: number, created_at: ISO8601 }, total: number. For mem_context, specify the structure of memories, skills, and workflows sections.
• Add parameter constraints to descriptions. E.g., 'topic_key: 2-32 lowercase alphanumeric + underscore, used to group related memories' and 'scope: one of project|team|personal, determines visibility and retention'.
• Implement confirmation for mem_delete. Return { confirmed: false, message: 'Delete memory "<title>"? Call mem_delete with confirmed=true to proceed.' } on first call, then execute on second call with confirmed=true.
• Add error recovery guidance. E.g., mem_search returning no results should suggest: 'No memories found. Try mem_timeline to see recent entries, or mem_stats to check project memory count.'