qlib-sync

本地 A 股数据，
始终最新，从不损坏。 Your Qlib A‑share data, current and never half‑written.

qlib-sync 把每天的社区数据包装进 Qlib 默认读取的 ~/.qlib/qlib_data/cn_data，先校验、后替换，任何一步出错都保留原有数据。你的 AI 助手也能直接读取它。 qlib-sync installs the daily community release into ~/.qlib/qlib_data/cn_data, where Qlib already looks. It verifies first and replaces second, and keeps your existing data if anything goes wrong. Your AI assistant can read it too.

uvx --python 3.12 qlib-sync cn update

或者 pip install qlib-sync。已是最新时约 1 秒结束。 Or pip install qlib-sync. When nothing is new, it finishes in about a second.

查看源码View the source 完整说明 Read the docs

更新是什么样子What an update looks like

先看状态，再更新：带进度下载、核对校验值、汇报新增了哪些交易日。再运行一次，因为没有新数据，一秒结束。 Check the status, then update: the download shows progress, the checksum is confirmed, and new trading days are listed. Run it again and it finishes in a second, because nothing is new.

真实录制，2 倍速播放。 Recorded for real, played at 2× speed.

为什么不直接 wget 加 tarWhy not just wget and tar

Qlib 的说明推荐用一行 wget 下载社区数据包。平时没问题，直到遇到这些情况： Qlib's README suggests a one-line wget of the community release. That works until one of these happens:

四格漫画 / Four-panel comic: a download dies at 99% at night; the next day 550 MB arrive and nothing changed; the qlib-sync robot verifies the data, stamps it, and keeps the old copy safe; the researcher relaxes and asks an AI for Moutai's recent closes.
  • 下载到一半断了The download breaks halfway 解压出半截数据，或者旧数据已被覆盖才发现失败。 You extract half a dataset, or find out after the old one is already overwritten.
  • 上游修正后重新发布The release gets corrected 上游发现数据错误后，用同一个日期重新发布（#163）。只看日期永远拿不到修复。 Upstream republishes a date after fixing data (#163). Comparing dates alone never picks up the fix.
  • 每次都下载 550 MBEvery check costs 550 MB 即使什么都没变。qlib-sync 先读一个 500 字节的清单。 Even when nothing changed. qlib-sync reads a 500-byte manifest first.
Update flow: check the manifest, download, verify size and sha256, stage beside the live data, swap and read with Qlib. Any failure puts the previous dataset back.

investment_data 自己的说明也给出了手动校验流程（manifest 加 validate_archive.py）。qlib-sync 每次自动执行同样的检查，并补上自动跳过、整体切换和失败回滚。 The investment_data README describes a manual check with the manifest and validate_archive.py. qlib-sync runs the same checks on every update and adds the skip, the atomic swap, and the rollback.

这些校验能挡住损坏、不完整或被篡改的数据包，挡不住上游正常发布的数据里本身的数值错误。需要时可以用 --keep-backup 保留一份旧数据。 These checks catch broken, truncated, or tampered packages. They cannot catch wrong values inside a correctly published release; keep a rollback copy with --keep-backup if you need one.

让 AI 助手读你本地的数据Let your AI assistant read your data

qlib-sync 自带一个本地 MCP server。MCP 是开放标准，Claude、Codex、Cursor、CodeBuddy、Trae、通义灵码、Cherry Studio 等都支持，背后用 DeepSeek、通义千问、Kimi、GLM 还是 Claude 都可以。数据始终留在你的电脑上。 qlib-sync includes a local MCP server. MCP is an open standard supported by Claude, Codex, Cursor, CodeBuddy, Trae, Tongyi Lingma, Cherry Studio and more, whichever model they run: DeepSeek, Qwen, Kimi, GLM, or Claude. Your data stays on your computer.

  • Claude Desktop

    从最新发布下载 qlib-sync.mcpb，双击安装。 Download qlib-sync.mcpb from the latest release and open it.

  • 命令行工具Command-line agents

    claude mcp add qlib-sync -- uvx --python 3.12 qlib-sync mcp
codex mcp add qlib-sync -- uvx --python 3.12 qlib-sync mcp

  • CodeBuddy、Trae、通义灵码、Cursor 等Cursor, CodeBuddy, Trae, Tongyi Lingma, and others

    在 MCP 设置里手动添加，粘贴： Add a server in the MCP settings and paste: 

    { "mcpServers": { "qlib-sync":
  { "command": "uvx", "args": ["--python", "3.12", "qlib-sync", "mcp"] } } }

    Cherry Studio、VS Code、Gemini CLI、Qwen Code 的写法见完整说明。 Cherry Studio, VS Code, Gemini CLI, and Qwen Code are covered in the README.

Claude Code 中的真实对话：茅台最近 5 个交易日收盘价，从 1452.00 到 1402.30。在 Codex 中问同样的问题，得到的是同一组数字。 A real session in Claude Code: Moutai's last five closes, from 1,452.00 to 1,402.30. Codex returns the same figures for the same question.

提供的工具：cn_status 查数据状态，cn_update 更新，cn_health_check 检查，cn_get_prices 取日线（默认真实成交价），cn_list_pool 查沪深 300 等成分股。 Tools: cn_status, cn_update, cn_health_check, cn_get_prices (real traded prices by default), and cn_list_pool for CSI 300 and other pools.

在 notebook 里用From a notebook

回测前确保数据是最新的，原来的 Qlib 代码不用改。 Make sure the data is current before a backtest. Your Qlib code stays the same. 

from qlib_sync import update_cn

result = update_cn()   # raises UpdateError on failure; existing data unchanged
print(result.latest_date, result.new_trading_days)

import qlib
qlib.init(provider_uri="~/.qlib/qlib_data/cn_data")