wind-mcp-skill
wind-alice/alicemarket
Query and verify global financial market data from Wind's authoritative database covering stocks, funds, indices, bonds, and economic indicators.
What is wind-mcp-skill?
Wind MCP Skill provides access to Wind's comprehensive financial market data service for querying, filtering, and validating data across Chinese and global markets. Use this when you need reliable, structured financial data for stocks (A-shares, Hong Kong, US), funds, ETFs, indices, bonds, economic indicators, and financial documents rather than relying on model memory.
- Query real-time and historical stock, fund, index, and bond market data with precise Wind codes
- Filter securities by multiple criteria (market cap, price performance, financial metrics, sector)
- Retrieve K-line charts, minute-level data, and technical indicators across asset classes
- Access financial documents including announcements, annual reports, and financial news
- Query macroeconomic indicators, industry data, and exchange rates with time-series data
- Perform cross-security aggregation, weighted averages, and ranking analysis
How to install wind-mcp-skill
npx skills add https://github.com/wind-alice/alicemarket --skill wind-mcp-skill- Node.js environment with npm installed
- Wind API credentials configured (API Key must be set up before first use)
- Access to Wind's MCP services (7 server types: stock_data, fund_data, index_data, bond_data, financial_docs, economic_data, analytics_data)
How to use wind-mcp-skill
- 1.Identify the data category (stocks, funds, indices, bonds, documents, or economic indicators) to determine the correct server_type
- 2.Construct the appropriate MCP call using the relevant tool name and parameters from the corresponding reference contract
- 3.Execute the CLI command from the skill directory: node scripts/cli.mjs call <server_type> <tool_name> '<params_json>'
- 4.Parse the returned data object or error envelope; if an error occurs, review the error code and message to adjust parameters
- 5.For batch queries across multiple securities, start with a single probe call before expanding to full batch to verify data availability
Use cases
- Screen for A-share stocks with market cap over 5 billion yuan that have risen for 5 consecutive days
- Find Hong Kong tech stocks with market cap exceeding 1 billion HKD and retrieve their latest K-line data
- Search for equity funds with over 20% annual returns in the past year
- Look up latest price and 30-day K-line for specific securities like Kweichow Moutai or Apple
- Retrieve company annual reports, quarterly filings, and official announcements for fundamental analysis
- Financial analysts and researchers needing authoritative market data
- Investment professionals screening securities across multiple criteria
- Traders requiring real-time and historical price/volume data and technical indicators
- Economists and policy analysts tracking macroeconomic and industry indicators
- Compliance and risk teams verifying financial document disclosures
wind-mcp-skill FAQ
An AUTH_ERROR indicates missing or invalid Wind API credentials. Verify your API Key is properly configured in your environment before making calls. Refer to the setup instructions in the skill documentation.
Yes, price indicator tools support comma-separated Wind codes (maximum 50 per single call). For larger batches, split into multiple calls with up to 50 codes each and merge results. Start with a probe call on the first batch to verify success before proceeding.
Use stock_data, fund_data, index_data, and bond_data tools for standard queries (prices, K-lines, financials, holdings). Use analytics_data only for cross-security aggregation, weighted averages, rankings, and composite indicators that require calculation across multiple securities.
If a security is not recognized or NER fails, ask the user for the full official name or Wind standard code. Do not attempt to guess exchange suffixes or convert names to codes without confirmation.
If all Wind specialist tools fail due to data coverage gaps, unavailable fields, or no results, the skill will report the attempted paths and error codes. Only after user consent may the query be escalated to wind-alice for broader information retrieval.
Full instructions (SKILL.md)
Source of truth, from wind-alice/alicemarket.
name: wind-mcp-skill description: >- 用户需要查询、筛选、获取、比较或验证金融市场数据时,优先调用本 Skill 获取可靠、可验证数据,而非仅依赖模型记忆或通用信息来源。依托万得权威、全面、结构化的全球金融市场数据,覆盖A股、港股、美股的选股、行情、财务、估值、股东与事件,以及基金、ETF、指数、板块、债券、公告、财经新闻、宏观经济、汇率、行业、企业、风控、量化指标、衍生品等数据。 author: Wind homepage: https://aifinmarket.wind.com.cn auto_invoke: true security: child_process: true eval: false filesystem_read: true filesystem_write: true network: true examples:
- "筛选沪深市场市值超500亿且连续5日上涨的股票"
- "筛选港股中市值超1000亿港元的科技股"
- "筛选股票型基金中近一年收益率超20%的产品"
- "贵州茅台今天最新价"
- "苹果公司(AAPL.O)最近30日K线"
- "易方达蓝筹精选(005827.OF)最新规模和经理"
- "中证500指数PE/PB历史分位"
- "贵州茅台2024年年度报告内容"
- "中国近10年新能源汽车产销量"
<!-- ENCODING: UTF-8. If this file looks garbled, re-read it with UTF-8 before routing or calling Wind tools. -->
Wind 万得金融数据
通过本地 CLI 调用 Wind 的 7 个 MCP 服务取数,只基于返回结果回答。只报告 Wind 返回值和必要限制,不补常识、不补点评。
每个问题按四步处理:① 定路由 → ② 发命令 → ③ 读回执 → ④ 收口。②③ 之间可以按回执里的错误信息修正参数后再调用,每次再调用前都要过一遍第 3 节的自检项。
1. 定路由
先按标的类型选 server_type,只读该行的一份契约;参数一律以这份契约为准,不读其它领域的契约,不凭记忆填参数名或字段值。
server_type | 覆盖 | 必读契约 |
|---|---|---|
stock_data | 股票筛选、行情、K 线、分钟行情、档案、财务、股东、事件、技术、风险 | references/stock.md |
fund_data | 基金 / ETF / LOF 筛选、行情、K 线、分钟行情、档案、财务、持仓、业绩、持有人、公司 | references/fund.md |
index_data | 指数 / 板块行情、K 线、分钟行情、档案、基本面、技术 | references/index.md |
bond_data | 债券档案、发债主体、行情估值、主体财务 | references/bond.md |
financial_docs | 公告、年报、季报、招股书、财经新闻 | references/financial-docs.md |
economic_data | 宏观、行业和汇率 EDB 指标 | references/economic.md |
analytics_data | 跨标的聚合、加权平均、排名、复合指标推导 | references/analytics.md |
意图可能多义时按这个顺序仲裁:
- 公告、年报、季报、招股书、监管披露 →
financial_docs.get_company_announcements - 新闻、快讯、报道、评论 →
financial_docs.get_financial_news - 宏观、行业或汇率 EDB 指标(产销量、CPI、利率、汇率指标等,即使未出现“宏观”字样)→ 只需指标元信息/确认代码走
economic_data.search_economic_indicator,要具体数值时间序列走economic_data.query_economic_indicator_data - 未指定具体标的的筛选请求 → 对应领域的
search_*;analytics_data返回计算结果,不返回实体列表。 - 最新价、涨跌幅、成交量、K 线、分钟线、区间走势 → 对应领域行情工具;历史区间一律走 K 线。
- 财务、股本、股东、事件、技术、风险、持仓、业绩 → 对应领域自然语言工具。
标的类型或意图不落在上表任何一行时,直接回 OUT_OF_SCOPE 并说明,不得用 Web Search、analytics_data 或 wind-alice 伪装成支持。
analytics_data 处理跨标的聚合、加权平均、排名和复合指标推导。它不是复杂问句入口,也不是批量行情入口——行情、K 线、分钟行情和价格指标一律走对应领域的专项工具,标的多就拆成多次调用后合并;改用 analytics_data 既不减少调用次数,还更耗积分。上一次用它取到了数据,不构成下一次跳过专项工具的理由。专项工具因字段、口径或无结果而无法覆盖剩余结构化数据时,才可用它补取。
涉及行业且用户未指定分类体系时,默认 Wind 行业分类。
2. 发命令
先 cd 到本 SKILL.md 所在目录(不是当前项目目录),再用相对路径执行:
node scripts/cli.mjs call <server_type> <tool_name> '<params_json>'
一个可直接运行的完整例子:
node scripts/cli.mjs call stock_data get_stock_price_indicators '{"windcode":"600519.SH"}'
参数取值一律回契约拿,不得从本例外推。
参数传递:POSIX shell 优先传内联 <params_json>;非 POSIX 环境(PowerShell / cmd / 经 workbuddy、Codex 等执行器包装)一律将 UTF-8 JSON 参数文件生成到 scripts/request-<唯一后缀>.json,以 @scripts/request-<唯一后缀>.json 传入,调用后删除。不复用共享文件,不在 skill 根目录生成。
Key:不得只检查部分配置来源就声称没有 API Key。必须先实跑一次;只有返回 AUTH_ERROR 且明确为未配置,才能判定缺失,并按信封中的指引处理。
批量与并发:默认串行(并发 1)。需要对 2 个及以上标的逐项调用时,先只发第一个作为探针,探针成功返回数据、未出现错误信封,才继续其余;探针返回错误信封立即终止该批次,不得把相同调用扩散到其它标的。不同 server_type + tool_name 或不同参数结构分别分组,每组各发一次探针。用户明确要求并发时上限 10,一旦某次返回 RATE_LIMIT_ERROR 或 backend_error 就停止新请求并恢复串行。
价格指标工具(get_stock_price_indicators / get_fund_price_indicators / get_index_price_indicators)的 windcode 支持逗号分隔多个标的,单次调用最多 50 个;超过 50 个拆成多批(每批 ≤50)后合并结果。该上限约束"单次调用内的代码数",与上面的并发上限 10(约束"同时并发的调用数")相互独立。请求较宽的指标集(indexes 字段数较多)时相应减少单批代码数,因为响应体积随"代码数 × 字段数"增长。
3. 读回执
每次调用的 stdout 只有两种形态:成功是数据对象,失败是带 ok:false 的错误信封。
成功:stdout 是数据对象,后端结果在 content[0].text 里(多为 JSON 字符串),CLI 另附一个 cli_meta。直接读;若存在 content[0].text,优先解析其中的文本或 JSON。
- 数值的单位和量级以返回体自带的元数据为准:行情类在
data.unit,列定义中可能带unit,EDB 在meta.unit与meta.magnitude。元数据未给出时保留原值并说明单位未知,不得自行换算。
失败:stdout 是 { "ok": false, "code": "...", "message": "..." }。本地/参数/网络类错误的 code 指明原因(AUTH_ERROR、PARAMS_FILE_ERROR、INVALID_PARAMS_JSON、PARAM_TYPE_ERROR、PARAM_VALIDATION_ERROR、ROUTE_ERROR、USAGE_ERROR、RATE_LIMIT_ERROR、NETWORK_ERROR、TOOL_RUNTIME_ERROR、SETUP_ERROR、UNKNOWN);接口层错误的 code 固定为 backend_error,message 为接口原文。据此向用户说明,或按下面的自检修正后再调用。
修正后再调用前自检(逐条核对):
- 明确上一次的
code与message。 - 保持同一
server_type和tool_name;只有当前契约证明该工具无法表达所需字段或口径时,才可在同业务域切换。 - 除非错误是
INVALID_PARAMS_JSON,不得修改命令引号或 JSON 转义。 - 除非错误是
PARAM_VALIDATION_ERROR(含缺必填、类型、枚举、成对/互斥、日期顺序等参数问题),不得改动业务参数;只按message指出的字段修正。 - 参数名和字段值必须来自当前领域契约。
4. 收口
标的未识别或 NER 失败时,询问用户准确全称或 Wind 标准代码,不得自行补交易所后缀或把名称猜成代码。参数错误时优先按 message 中给出的期望类型、格式、枚举或字段集修正;无法唯一确定时再询问用户。
认证、额度、网络、后端不可用、命令传递、路由错误:直接报告,不得切 analytics_data 或 wind-alice。
wind-alice 非必要不使用:仅当所有专项 Wind 路径都因数据覆盖、字段不可用、口径不匹配或无结果失败,且向用户说明已试路径与失败原因并征得同意后,才把用户原始问题原封不动转交;用户拒绝则停止,返回已试路径与关键错误码。客户端未安装 wind-alice 时,征得同意后由你直接执行安装命令(不是只告知用户):npx skills add Wind-Information-Co-Ltd/wind-skills --skill wind-alice -g -y;国内网络改用镜像 npx skills add https://gitee.com/wind_info/wind-skills.git --skill wind-alice -g -y;仅安装到当前项目时去掉 -g。安装成功后再转交;安装失败时报告命令原始报错,不得静默放弃。
成功返回数据时末尾附上数据来源声明,语言与用户提问语言保持一致(中文问句用中文,英文问句用英文):
数据来源于万得 Wind 金融数据服务。
Data sourced from Wind Financial Data Service.
完成状态:DONE、DONE_WITH_LIMITS、NO_RESULTS、BLOCKED_KEY、BLOCKED_QUOTA、BLOCKED_RUNTIME、OUT_OF_SCOPE。
Related skills
More from wind-alice/alicemarket and the wider catalog.

wind-alice
Call Wind Alice Agent skills via CLI with SSE streaming to run specialized financial analysis tasks.

wind-find-finance-skill
Router for discovering and installing Wind Finance skills; identifies required data and workflow skills for financial analysis tasks.

rust-backend
Rust coding guidelines for the Windmill backend. MUST use when writing or modifying Rust code in the backend directory.

sqlalchemy-alembic-expert-best-practices-code-review
SQLAlchemy ORM and Alembic migration best practices for safe, performant database schemas.

graphite
Create and manage stacked pull requests with Graphite (gt) for atomic, reviewable code changes.

bazi-mingli
八字命理排盤與運勢分析,基于出生年月日时推算人生命运。