FinDataMCP provides 5 financial data tools with consistent naming (all get_* verbs) and reasonable descriptions (100-250 chars). However, input schemas are minimal, only ticker and optional limit/days_back parameters with basic type info. No output schemas are documented, forcing LLMs to infer response structure. Error handling returns success/error flags but lacks actionable recovery guidance. Parameter descriptions are present but sparse (e.g., 'Stock ticker symbol' is generic; no format hints like 'e.g., AAPL'). No enum constraints, no pagination guidance despite tools returning lists. Descriptions mention returned fields but don't formally document the output schema structure.

• Document output schemas as JSON Schema objects. For get_financial_snapshot, define: {type: 'object', properties: {valuation_metrics: {...}, profitability_metrics: {...}, ...}, required: [...]}. This enables LLMs to extract fields reliably.
• Enhance error messages with recovery guidance. Instead of {'success': False, 'error': 'API key not set'}, return {'success': False, 'error': 'API key not set', 'recovery': 'Set FINANCIAL_API_KEY environment variable and restart the server'}.
• Add format constraints to parameter descriptions. Change 'Stock ticker symbol' to 'Stock ticker symbol (uppercase, 1-5 chars, e.g., AAPL, MSFT)'. Add min/max to limit params: 'Number of trades to return (1-1000, default: 100)'.
• Document pagination and result limits explicitly. Add to get_insider_trades and get_institutional_ownership: 'Returns up to [limit] records. No cursor-based pagination; use limit to control result size.'
• Add tool annotations (readOnlyHint) to all tools. All 5 are read-only; annotate them with readOnlyHint=true so agents know they are safe to retry without side effects.