This server exposes 5 financial data tools from the AKShare library. All tools are READ_ONLY wrappers around external library functions that return pandas DataFrames formatted as markdown/csv/json. Critical deficiencies: (1) NO input schemas are visible in the source code, tools are registered via @mcp.tool() decorator on wrapped functions with no explicit parameter validation or schema definition; (2) Descriptions are extremely generic (all follow the pattern 'From akshare.X - returns DataFrame formatted as markdown, csv, or json'), under 100 characters, providing no actionable guidance on when/why to use each tool or what data they return; (3) Parameter documentation is completely absent, the underlying akshare functions likely have parameters (filters, date ranges, etc.) but none are described in the tool registration; (4) Output schemas are undocumented, LLMs cannot predict what fields a stock index query returns; (5) No error handling, validation, or recovery guidance; (6) Tool names do not follow verb_noun convention (should be get_stock_zh_index_spot, list_global_indices, etc.). The server is essentially a thin passthrough with minimal MCP-level metadata.

• Rename all tools to follow verb_noun convention: 'get_stock_zh_index_spot', 'list_global_indices', 'get_stock_zh_a_spot', 'get_stock_earnings_report', 'get_stock_zt_pool'. This enables LLMs to infer intent from the name.
• Write substantive descriptions (100-250 chars each) explaining WHAT data each tool returns, WHEN to use it, and any key prerequisites. Example: 'get_stock_zh_index_spot: Fetch real-time spot price and volume data for Chinese stock indices (SSE, SZSE, CSI). Returns symbol, price, change_percent, volume. Use to monitor index trends or verify market open status.'
• Inspect akshare function signatures (via inspect.signature) and expose their parameters as tool input parameters with proper types, enums, and descriptions. E.g. if stock_zh_index_spot_em() accepts a 'date' parameter, add it to the tool schema as input_date: string (ISO 8601).
• Document output schemas for each tool. Run sample akshare calls, inspect the returned DataFrame columns, and define the output structure in the tool response documentation. Example: 'Returns JSON array of objects: [{symbol: str, price: float, change_percent: float, volume: int, timestamp: ISO8601}]'.
• Add error handling in output_format() wrapper: catch exceptions (network timeouts, data parsing errors) and return actionable error messages. Example: 'Failed to fetch stock data: timeout after 30s. Retry in 30s or try a different time window.' This enables LLM to self-correct.
• Add optional output_format parameter to tool inputs (default to server global, but allow per-call override). This gives LLMs flexibility to request markdown for readability or JSON for structured parsing.