Mnemo provides 6 tools with visible schemas and descriptions, but quality is inconsistent. All tools have basic schemas with required fields and enums where applicable (e.g., memory_type in 'remember'). Descriptions are present but generic, most are 20-60 characters, well below the 50-200 char baseline for LLM-optimized descriptions. Parameter descriptions are minimal or missing entirely. Error handling is not visible in the tool definitions. The 'forget' tool is destructive but has no confirmation or dry-run pattern documented. Tool composition is sound (each does one thing), but descriptions lack the 'when to use' and 'what it returns' context needed for LLM selection.

• Expand all tool descriptions to 50-200 characters. Include: WHAT the tool does, WHEN to call it (in contrast to similar tools like recall vs. search), and WHAT it returns. Example: 'remember: Stores a fact, skill, code pattern, or preference with a key and optional tags. Use this to save information for later recall. Returns the memory ID and confirmation.' Current: 'Store a memory'.
• Add full parameter descriptions for all inputs. For 'tags' in remember: 'Comma-separated or array of labels to categorize the memory. Example tags: python, debugging, frontend. Optional.' For 'memory_types' in search: 'Filter results by memory type. Valid values: fact, skill, preference, code_pattern. If omitted, searches all types.'
• Document output schemas. After each tool definition, add a 'Returns' section. Example for 'remember': Returns {memory_id: string, key: string, created_at: ISO8601, memory_type: string}. This enables downstream tool chaining.
• Implement confirmation or dry-run for 'forget'. Offer a confirm_deletion pattern: when forget is called, return {status: 'pending_confirmation', memory_id: X, preview: {...}}. Require a follow-up call to actually delete. Alternatively, add a dry_run parameter (default false).
• Clarify the recall vs. search distinction. Rename 'recall' to 'recall_best_match' and update description: 'Returns the single most relevant memory matching the query. Use when you need one focused result. For multiple results, use search.' Rename 'search' description: 'Returns up to N memories matching the query, ranked by relevance. Use for discovery or when multiple results are relevant.'

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained