Financial data aggregation MCP server with support for multiple data providers (Tushare, Wind). Provides tools for accessing stock fundamentals, market data, financial statements, and macroeconomic indicators.
This MCP server exposes 15 financial data tools via FastMCP with basic input schemas and Chinese-language descriptions. Critical issues: (1) Tool names lack action verbs, stock_basic, bak_basic, cn_gdp, cn_cpi, cn_ppi, cn_m, sf_month, cn_pmi are nouns, not imperative verbs. LLMs struggle to infer intent from 'stock_basic' vs 'get_stock_basic'. (2) Descriptions are minimal (10-50 chars) and many are in Chinese, blocking English-speaking LLMs. (3) No output schemas are documented anywhere in the source. LLMs cannot predict what fields to expect or plan downstream calls. (4) Parameters lack depth, most have only a type and brief description with no constraints, enums, or format guidance. (5) Zero error handling visible; no guidance on what to do if a lookup fails or the API returns no results. (6) Tool registration uses dynamic decorator pattern (decorate_async_functions) rather than explicit schema registration, making schema verification difficult. (7) The sayHiWind tool is placeholder-level and adds no value.
获取某支股票在指定时间范围内的基本面数据,数据从2016年开始。
获取上市公司财务三大报表之资产负债表数据。
获取上市公司财务三大报表之现金流量表数据。
获取CPI居民消费价格数据,包括全国、城市和农村的数据。
获取国民经济之GDP数据。
获取货币供应量之月度数据。
获取采购经理人指数(PMI)数据。
Tool names are all nouns (stock_basic, cn_gdp, bak_basic, sf_month) with no action verb. LLMs infer intent from names and will struggle to distinguish 'stock_basic' from other stock tools or understand it retrieves data.
Descriptions are minimal (10-50 chars) and many are in Chinese ('获取股票的名称、代码等基础信息'). English-speaking LLMs cannot understand these descriptions, and they lack context on WHEN to use each tool or what it returns.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 29 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 18 | - | v1 |
获取PPI工业生产者出厂价格指数数据。
获取股票未复权的日线行情数据。
获取上市公司财务三大报表之利润表数据。
礼貌的回应用户的打招呼
获取月度社会融资数据。
获取LPR贷款基础利率。贷款基础利率(Loan Prime Rate,简称LPR),是基于报价行自主报出的最优贷款利率计算并发布的贷款市场参考利率。
获取股票的名称、代码等基础信息。
获取上市公司基础信息。
No output schemas documented in source. LLMs cannot predict what fields each tool returns, preventing downstream tool composition and forcing wasteful discovery calls. All 15 tools affected.
Parameters lack constraints, enums, or format validation. 'market' accepts free-form strings when it should be an enum (主板, 创业板, 科创板, etc.). 'ts_code' has no length constraints or pattern. Date parameters (start_date, end_date) lack format validation (YYYYMMDD vs YYYY-MM-DD ambiguity).
Zero error handling or recovery guidance in visible code. If a stock code does not exist, the API response is not documented. If date ranges are invalid, no actionable error is returned. LLMs cannot self-correct.
Parameter descriptions are too brief. 'fields' is described as '从Fields中选取需要查询的字段' (Chinese) with no indication of valid field names, whether it accepts a string or array, or what happens if omitted. All tools with 'fields' param affected (14 tools).
Tool registration uses dynamic decorator pattern (decorate_async_functions in server.py) rather than explicit schema definition. Schema is inferred from function signatures; no explicit JSON Schema registration visible. This makes validation and schema inspection difficult.
sayHiWind tool is placeholder-level (empty input, greeting response) and adds no domain value. It clutters the tool list and wastes the LLM's reasoning budget.
Mutually exclusive parameters not documented. E.g. 'daily' tool has trade_date vs (start_date + end_date) as alternatives, and 'income' has ann_date vs f_ann_date vs (start_date + end_date) vs period. LLMs will pass multiple conflicting params unless explicitly told not to.
No pagination parameters visible for tools that could return many results (daily stock data, financial statements). Without limit/offset/cursor support, LLM context can be blown if a query returns hundreds of rows.