Server has 7 tools with complete schemas and descriptions. Naming follows verb_noun convention (get_, list_, run_). Descriptions are present and substantive (avg ~150 chars), exceeding the 10-char minimum. However, parameter descriptions lack depth, many are generic ('A list of...') without format constraints, enums, or validation guidance. Output schemas are not documented. Error handling is absent from visible code. No tool annotations (readOnlyHint, destructiveHint). The run_report and run_realtime_report tools accept complex nested objects (FilterExpression, OrderBy) with no schema detail or examples, forcing LLMs to guess structure. Missing: per-parameter type hints in descriptions, output schema documentation, error recovery guidance.

• Document output schemas for all tools. Include field names, types, and whether fields are always present or conditional. Example: 'Returns {accounts: [{id: string, name: string, properties: [{id: string, displayName: string}]}]}'.
• Add enum constraints to categorical parameters. E.g., for date_ranges, specify format: 'ISO 8601 date strings (YYYY-MM-DD) or relative ranges (last_7_days, last_30_days, last_90_days)'.
• Expand parameter descriptions with format and validation rules. E.g., 'limit: integer, 1 - 250,000 (default 10,000). Larger limits may timeout.' instead of 'The maximum number of rows to return'.
• Implement structured error responses with classification and recovery hints. E.g., '{error: 'invalid_property_id', message: 'Property 123 not found', retryable: false, suggestion: 'Call get_account_summaries() to list valid properties'}'.
• Add readOnlyHint: true to all tool definitions in coordinator.py via MCP tool annotations to signal safety to LLMs and prevent unnecessary confirmation prompts.