Memory MCP has 19 tools with consistent naming (memory_* prefix) and descriptions present for all. However, parameter descriptions are minimal (often single-word or 5-10 chars), and output schemas are not documented in the source. Input schemas are visible and properly typed. Error handling is basic, no recovery guidance or actionable error messages evident. Tool composition is reasonable (each does one thing), but descriptions lack the 50-200 char LLM-optimized range (most are 40-80 chars). No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications in the spec.

• Add output schema documentation to every tool. Document the response structure (fields, types, example values) so LLMs know what to expect and can chain tools correctly.
• Implement tool annotations: add readOnlyHint=true to memory_session_start, memory_search, memory_get_rules, memory_asoode_status; add destructiveHint=true to memory_delete_rule; add idempotentHint=true to memory_task_start, memory_task_done.
• Expand parameter descriptions to 30-80 chars with format/constraint hints. E.g., 'Project slug (2-50 lowercase alphanumeric chars; if omitted, uses active project from memory_use)' instead of 'Project slug'.
• Add error recovery guidance to tool descriptions. E.g., memory_search: 'Returns empty if project not found. Call memory_session_start(project) first to initialize.' memory_task_claim_next: 'Returns error if queue is empty; call memory_task_add to create tasks.'
• Add a memory_delete_rule_confirm tool or support a 'confirm' parameter on memory_delete_rule to prevent accidental rule deletion. Similarly, document memory_session_end's side effects (releases task claims) prominently.