Graphiti Local provides 6 read-only tools with complete JSON schemas and reasonable descriptions. All tools have input schemas with typed parameters and descriptions. However, descriptions are generic and lack LLM-optimized guidance on when/why to use each tool. Parameter descriptions are minimal (e.g., 'Search query string' without format hints). No output schemas documented. Error handling and recovery guidance absent. Tool naming is clear (verb_noun pattern) but composition could be tighter, search_nodes and search_memory_facts overlap conceptually.

• Expand tool descriptions to 100 - 200 characters. Example: 'Search graph entities by name or property. Use this to find nodes (people, companies, concepts). Returns node UUIDs, names, and types. If you need relationships/facts, use search_memory_facts instead.'
• Document output schemas for all tools. Specify fields returned (e.g., search_nodes returns [{uuid, name, entity_type, properties}]). This enables LLMs to chain tools and extract required IDs.
• Add parameter constraints and examples. E.g., 'max_nodes: integer, 1 - 100 (default 10). Limit results to prevent context overflow.' For group_ids: 'Array of group identifiers. Obtain via get_status or configuration.'
• Add error handling guidance. E.g., 'If no results found, try broadening the query or removing entity_type filters. If UUID is invalid, use search_nodes to find the correct UUID.'
• Clarify tool selection. Add to search_nodes: 'Use for entity discovery. For relationship facts, use search_memory_facts.' Add to search_memory_facts: 'Use for facts/relationships. For entity properties, use search_nodes.'