akshare-jc-mcp

Unified MCP server for Chinese stock (A-share) market data.

Why akshare-jc-mcp?

Install

pip install git+https://github.com/bzetu/akshare-jc-mcp

Or for development:

git clone https://github.com/bzetu/akshare-jc-mcp
cd akshare-jc-mcp
pip install -e .

MCP Configuration

Add to your MCP client config (Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "akshare-jc": {
      "command": "python3.13",
      "args": ["-m", "akshare_jc_mcp"]
    }
  }
}

Requires Python 3.11+ and a working akshare installation (auto-installed as dependency).

Tool: get_data

Parameters

Features

Typical Usage

消息面+基本面查询（一次性获取全部）：

    get_data(symbol="600733", features=["news", "inner_trade", "financial", "fund_flow", "concept", "hsgt_summary"], news_recent_n=10, recent_n=3)

日K技术分析（默认120条，约6个月）：

    get_data(symbol="600733", features=["hist_data", "realtime"], hist_interval="day", hist_day_n=120, hist_indicators=["KDJ","MACD","RSI","BOLL","SMA"])

月K长线趋势（默认36个月=3年）：

    get_data(symbol="600733", features=["hist_data", "realtime"], hist_interval="month", hist_month_n=36, hist_indicators=["KDJ","MACD","RSI","BOLL","SMA"])

预计算技术分析（保持默认返回兼容，显式开启）：

    get_data(symbol="600733", features=["hist_data"], hist_analysis=true)

With hist_analysis=true, hist_data.data becomes {"raw": [...], "analysis": {...}}. The analysis is compact: MA/MACD/RSI/KDJ/BOLL/volume state, regression trend, MACD divergence, support/resistance, single/multi-bar patterns, double-top/double-bottom candidates, latest cross events, and an evidence-backed mechanical score. bar_period_end is the period label for weekly/monthly bars; source_last_trading_day is the actual latest underlying daily session. An unfinished period is marked intraday_partial. Major patterns use explicit candidate, confirmed, and invalidated statuses, and their key points identify whether they came from high or low. Minute K-lines skip divergence, support/resistance, and pattern detection. The default response remains the original raw array.

For compact multi-timeframe reports, set hist_raw_n=0 for analysis-only month/week requests and hist_raw_n=10 for a daily request that needs recent candles. hist_raw_n only changes the returned raw array; the complete requested history is still used to calculate analysis.

年K超长线（默认10年）：

    get_data(symbol="600733", features=["hist_data", "realtime"], hist_interval="year", hist_year_n=10, hist_indicators=["KDJ","MACD","RSI","BOLL","SMA"])

1分钟K盘中超短线（仅交易时段有数据，全天最多480条）：

    get_data(symbol="600733", features=["hist_data", "realtime"], hist_interval="1min", hist_indicators=["KDJ","MACD","RSI","BOLL","SMA"])

限售解禁+增发数据（计算解禁成本）：

    get_data(symbol="600733", features=["restricted_release", "additional_issuance"])

Response Format

JSON array, one entry per feature:

[
  {"feature": "news", "data": [...], "error": false, "error_reason": null},
  {"feature": "fund_flow", "data": [...], "error": false, "error_reason": null},
  ...
]

error: true means the feature failed — LLM should fall back to web search or inform the user.

Design & Reliability

Problems Solved

Reliability

All HTTP calls share a single requests.Session to minimize connection overhead. Features that depend on eastmoney APIs fall back to alternative endpoints (push2delay when push2/push2his are unavailable). Total 6-feature fetch (news+inner_trade+financial+fund_flow+concept+hsgt_summary) completes in 2-3s (~15KB).

Development

pip install -e .
python -c "from akshare_jc_mcp.server import mcp; print(mcp.name)"

License

MIT