Four tools with basic schemas and minimal descriptions. All tools have input schemas with typed parameters, but descriptions are generic (10-50 chars) and lack context for LLM selection. No output schemas documented. Parameters lack constraints (enums, ranges, formats). No error handling guidance. Tool names follow verb_noun pattern but descriptions do not explain WHEN to use each tool or what distinguishes search_nodes from search_memory_facts. Missing idempotency hints for write operation (add_memory). No pagination guidance despite max_* parameters suggesting result limiting.

• Expand tool descriptions to 100-200 chars. For add_memory: 'Add a memory episode to the graph. Use this to record new facts, events, or context about an agent or user. Returns episode_id for reference in future queries.' For search_nodes vs search_memory_facts, clarify: 'search_nodes finds graph vertices (entities); search_memory_facts finds semantic relationships and facts.'
• Document output schemas for all tools. Example for get_episodes: {episodes: [{id: string, name: string, body: string, created_at: ISO8601}], total: integer, next_cursor?: string}. This enables agents to chain calls and extract IDs.
• Add parameter constraints. max_episodes: {type: integer, minimum: 1, maximum: 100, description: 'Limit results to 1-100 episodes'}. Add enum for group_ids if only certain values are valid.
• Add error handling guidance. 'If group_id not found, returns {error: 'group_not_found', available_groups: [...]}' This lets agents self-correct without extra discovery calls.
• For add_memory, add idempotency guidance: 'Idempotent: calling twice with same name/group_id/body returns same episode_id. Safe to retry on network failure.' Consider returning {episode_id, created: boolean} to signal if this was a new insert.

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained