mcp-eastmoney
README.md
# mcp-eastmoney
> 🇨🇳 A 股投研数据底座 + 策略筛选引擎 — 让 Claude / Cursor / Codex 等 MCP 客户端拥有完整的"数据 → 计算 → 推荐 → 后验"闭环
[](https://www.python.org/downloads/)
[](https://opensource.org/licenses/MIT)
[](https://modelcontextprotocol.io)
[]()
[English](#english) | [中文](#中文)
---
## 中文
**mcp-eastmoney** 不只是一个行情查询工具——它是一个**分层完整的个人量化研究基础设施**:SQLite 时序+截面双库、缠论/筹码/因子计算引擎、YAML 驱动的策略筛选漏斗、以及"真实可成交入场价"的推荐绩效闭环。全部数据源免费免 Key。
### ✨ 核心特性
- 🆓 **完全免费** — 腾讯/Baostock/AKShare/同花顺/新浪等公开源,零 API Key
- 🐉 **龙头战法** — 五维"识别真龙"评分(带动/领涨/抗跌/流动/承接,门槛一票否决)
- 🧠 **缠论引擎** — 分型/笔/线段/中枢/背驰/三类买卖点(含置信度与止损位)
- 📊 **16 因子库** — 动量/波动/流动性/技术/估值/质量 六类因子横截面排名
- 🎯 **策略漏斗** — YAML 策略筛选 → 趋势终审 → 排雷 → ATR 定仓,一键输出候选包
- 📈 **板块轮动** — 90 个同花顺行业指数 N 日排行 vs 大盘基准
- 🔄 **推荐闭环** — 落库推荐 → 真实入场价后验(涨停顺延开盘、连续一字作废)→ 分策略胜率归因
- 📦 **20 张表 / 33 个 Tool** — eod 定时 14 步全自动,幂等可重跑
### 🏗️ 架构
```
server.py — 33 个 MCP Tool
├── data/ 数据层(8 客户端 + 20 张表唯一写入口)
│ tencent / baostock / akshare / 新浪财务 / 三源新闻 / 东财备源
├── compute/ 计算层(纯函数零网络:缠论/指标/因子/筹码/异动/趋势/仓位/龙头)
├── recommend/ 推荐层(策略YAML筛选 + pipeline漏斗编排 + 龙头扫描)
├── strategies/ 策略定义(放量突破/双低价值/缩量回踩,可自定义覆盖)
└── refresh.py 编排层(eod 每日14步 / weekly / 12个CLI任务)
```
### 🛠️ Tools 一览(33 个)
| 分组 | Tools |
|------|-------|
| 行情 | get_stock_quote / search_stock / main_fund_rank / sector_fund_flow / get_kline |
| 研究计算 | get_indicators / get_chan_analysis / get_chip_distribution / trend_score / position_plan / factor_rank / factor_catalog |
| 推荐漏斗 | list_strategies / screen / **run_pipeline** / scan_dragons / screen_runs |
| 情绪事件 | get_zt_pool / get_lhb / get_earnings_forecast / sector_rotation |
| 基础数据 | get_news / get_stock_concepts / get_concept_members / get_stock_industry / get_financial_indicator |
| 闭环管理 | manage_watchlist / check_alerts / refresh_daily / save_recommendation / get_recommendation_stats / save_report / list_reports |
### 🚀 快速开始
```bash
# 安装运行(uv 推荐)
uvx mcp-eastmoney
# 或
pip install mcp-eastmoney && mcp-eastmoney
# 初始化数据(可选,eod 也可每日定时自动做)
uv run python -m mcp_eastmoney.refresh basic
uv run python -m mcp_eastmoney.refresh index
uv run python -m mcp_eastmoney.refresh eod # 每日收盘后全套
# 接入 Claude Desktop
```
```json
{
"mcpServers": {
"eastmoney": { "command": "uvx", "args": ["mcp-eastmoney"] }
}
}
```
### 🐉 三十秒看懂推荐漏斗
```
market_snapshot(全市场每日截面 5,222 只)
→ run_pipeline("shrink_pullback") # 策略筛选+趋势终审+排雷+定仓
→ AI 终审候选包(缠论/筹码/龙虎榜/业绩催化)
→ save_recommendation(skill_version=策略名)
→ eod 每日真实入场价后验 → 分策略胜率归因
```
打板风格另有 `scan_dragons`(龙头识别)入口。
### ⚠️ 数据说明
- 实时行情约延迟 15 分钟(腾讯免费源),仅供研究参考
- 筹码分布为换手衰减三角分布近似模型(东财同思路本地推演,非官方精确值)
- 龙头评分为日线级近似(原版 dragon-quant 的分钟级子因子已降维)
- **不构成任何投资建议**
### 🤝 贡献
欢迎 PR!开发约定:数据层不计算、计算层零网络零落表、写库只走 db.py、全部幂等 upsert。
```bash
git clone https://github.com/<your-username>/mcp-eastmoney.git
cd mcp-eastmoney && uv sync
uv run pytest tests/ -q --ignore=tests/test_live.py # 53 个离线测试
```
### 📜 License
MIT
---
## English
**mcp-eastmoney** is a full-stack personal equity research infrastructure as an MCP server — not just quotes. It ships a layered architecture (data / compute / recommend / orchestration), a complete Chan-theory engine, a 16-factor cross-sectional ranking library, YAML-driven screening pipelines, a dragon-leader (龙头战法) five-dimension scorer, sector rotation analytics over 90 THS industry indices, and a recommendation feedback loop that evaluates picks at **realistically fillable entry prices** (limit-up deferral to next open, consecutive one-limit boards voided). All data sources are free and keyless.
Key tools: `run_pipeline` (one-shot screening funnel), `scan_dragons` (limit-up leader identification), `get_chan_analysis` (Chan structure), `sector_rotation` (industry momentum vs benchmark), `save_recommendation` → `get_recommendation_stats` (per-strategy win-rate attribution).
Quick start: `uvx mcp-eastmoney`, then add to any MCP client. Daily data via `python -m mcp_eastmoney.refresh eod`. Data delayed ~15 min; not financial advice. MIT License.
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessNo issues