Server has 9 tools with complete schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (query-entity, get-entity-metadata, create-entity-record, etc.), which is good. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Parameter descriptions exist but are minimal (10-30 chars). No output schemas documented. Error handling is absent, no recovery guidance, no categorization of retryable vs. fatal errors. Destructive tools (delete-entity-record) lack confirmation patterns. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Prompts feature is implemented but not integrated with tool discovery.

• Document output schemas for all tools. Example: create-entity-record should return {recordId: string, createdAt: ISO8601, entityName: string}. This enables LLMs to chain tools (e.g., create then update).
• Expand tool descriptions to 150-250 chars with WHEN guidance. Example: 'Query entities using OData filters. Use this for filtered searches; use list-entities to discover all entities, get-entity-metadata for schema info.'
• Add tool annotations to schema: readOnlyHint=true for get_* and query_*, destructiveHint=true for delete_*, idempotentHint=true for get_* and list_*. Enables LLM to reason about safety.
• Implement confirmation pattern for delete-entity-record: require a confirm_delete=true parameter or return a confirmation request (MRTR pattern) before executing deletion.
• Add error handling with recovery hints. Example: 'Entity not found. Call list-entities to see available entities, or check the entityName spelling.' Categorize errors as retryable (network timeout) vs. user-fixable (invalid entity name) vs. fatal (permission denied).