Server has 8 tools with basic schemas and descriptions, but significant gaps in parameter documentation, error handling, and output schema clarity. Cognee tools (1-4) have reasonable descriptions (80-150 chars) and input schemas with types. Mem0 tools (5-8) have descriptions in Chinese with minimal English context, making LLM integration unclear. Most tools lack output schema documentation. Parameter descriptions are sparse or missing entirely (e.g., 'metadata' in add_memory has no type hint). No error handling guidance, no recovery paths, no permission gates. Tool naming is verb-first (good), but composition issues exist: cognee_add + cognee_cognify should be one atomic operation or clearly sequenced. Schemas present but incomplete, many parameters lack descriptions or type constraints.

• Translate all Mem0 tool descriptions to English (50-150 chars each). Example: 'add_memory' → 'Store a fact or decision in persistent memory. Automatically deduplicates and resolves conflicts. Use for user corrections, discoveries, or session summaries.'
• Document output schemas for all tools. E.g., cognee_search should return {results: [{entity: string, type: string, relationships: string[]}], total: number}. Include in tool description or as a separate schema field.
• Add type and bounds to all parameters: metadata → {type: 'object', properties: {type: {type: 'string'}, project: {type: 'string'}}}; top_k → {type: 'integer', minimum: 1, maximum: 100, default: 5}.
• Add error handling guidance to delete_memory: 'If memory_id not found, returns error with available memory IDs. If deletion fails, returns retryable error with guidance.'
• Merge cognee_add + cognee_cognify into a single 'cognee_add_and_process' tool, or add explicit dependency hint: 'After calling cognee_add, always call cognee_cognify with the same dataset_name to extract entities and relationships.'