Server has 19 tools with complete input schemas and descriptions in Korean. Naming follows verb_noun pattern (memory_save, memory_search, etc.). Descriptions are present but brief (avg ~80 chars, below the 194-char baseline). Schemas are well-structured with enums and type constraints. However, output schemas are not documented in the source, parameter descriptions lack depth (e.g., 'TTL 원본 문자열' without format guidance), and error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Security concern: memory_import accepts raw JSON data as string parameter without validation hints.

• Document output schemas for all tools. For memory_search, specify: {results: [{id, title, content, score, store, tags}], total_count}. For memory_list, specify pagination fields (offset, limit, total).
• Add tool annotations to the tool registration. Mark memory_delete, memory_bulk_delete, memory_reindex, memory_import, memory_restore with destructiveHint=true. Mark memory_search, memory_list, memory_get, memory_stats with readOnlyHint=true.
• Translate all descriptions and parameter docs to English. Use concise, LLM-optimized descriptions (50 - 200 chars). E.g., 'Save a memory to one of four stores (general, images, files, secrets). Supports TTL, tags, priority, and relationships.' instead of the current Korean text.
• Add format constraints to parameters. For ttl, specify: 'ISO 8601 duration format (e.g., 3d, 12h, 1w). Max 365 days.' For scoreThreshold, specify: 'Float 0.0 - 1.0, default 0.5.'
• Add error recovery guidance to tool descriptions. E.g., memory_import: 'If JSON is invalid, returns error with line number. Ensure all memory objects have required fields: content, store.' memory_search: 'If no results found, try broader query or lower scoreThreshold.'

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

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry