Server has 6 tools with basic schemas and descriptions, but significant gaps prevent higher scores. Tool names follow verb_noun convention (query, get_record, list_records, create_record, update_record, delete_record), which is good. However, descriptions are minimal (10-30 chars), parameter descriptions are sparse or missing, and output schemas are not documented. The 'body' parameter in create_record and update_record is dangerously vague (type: object, description: 'Record data'). No enum constraints on recordType. Error handling and recovery guidance absent. Schemas present but incomplete, missing type constraints, ranges, and field documentation. This is typical of early-stage MCP servers.

• Document output schemas for all tools. Example: query returns {rows: [{field: value, ...}], total_count: number}. get_record returns {id, type, fields: {}}. This enables LLM planning and field extraction.
• Replace free-form 'body' parameter with explicit field schemas. Use a discriminated union or nested object: {fields: {customer_name: string, email: string, ...}} with required/optional markers. Or provide a discovery tool (describe_record_type) that returns valid fields.
• Add enum constraint to recordType parameter. List valid NetSuite record types (customer, item, vendor, employee, etc.) so LLMs pick from known options instead of guessing.
• Expand tool descriptions to 80 - 150 characters. Include WHEN to use (e.g., 'Use to fetch a single record by ID. For multiple records, use list_records.'). State what fields are returned and any prerequisites (e.g., 'Requires valid NetSuite account ID').
• Add error handling with recovery hints. Example: 'Record not found. Verify the record type and ID are correct. Use list_records to search by name.' Categorize errors as retryable (network timeout) vs user-fixable (invalid ID) vs fatal (permission denied).