Server provides 8 tools with complete JSON schemas and basic descriptions. However, descriptions are generic and lack actionable context for LLM selection. Parameter descriptions are minimal (1-5 words). No error handling guidance, no output schema documentation, and no tool annotations (destructiveHint, readOnlyHint). Tool names follow verb_noun convention but descriptions do not explain WHEN to use each tool or what happens on failure. Schemas are present but descriptions fall short of the 50-200 char LLM-optimized baseline (avg ~60 chars, mostly under 20 chars for params).

• Expand tool descriptions to 50-200 chars. Example: 'create_entities: Add new entities to the knowledge graph. Each entity has a name, type, and list of observations. Filters out duplicates by name. Use this before creating relations between entities.' This explains WHAT, WHEN, and WHAT HAPPENS.
• Add output schema documentation for read_graph and search_nodes. Specify returned fields (entity names, types, observations, relation types) so LLMs know what data is available for downstream calls.
• Add destructiveHint annotations to delete_* tools and readOnlyHint to read_graph and search_nodes. This enables clients to warn users or require confirmation before executing destructive operations.
• Expand parameter descriptions with format and constraint details. E.g., 'entities: Array of entity objects. Each must have: name (string, 1-255 chars), entityType (string, e.g. person, place, concept), observations (array of strings, each 1-1000 chars). Duplicate names are filtered.'
• Add error handling guidance in tool descriptions. E.g., 'delete_entities: Deletes entities and all relations referencing them. If an entity does not exist, it is silently skipped. Returns success even if no entities were deleted.' This prevents LLM confusion on partial failures.