A-Share Full-Stack Data Toolkit
10-layer architecture, 47 data endpoints, 15 data sources, zero-auth — one skill file integrates A-share market, research, capital flow and more.
- Source repo
- simonlin1212/a-stock-data
- Stars
- ★ 11k
- Last updated
- 7d ago
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeCodex (Partial support)OpenAI API (Needs adaptation)
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Individual investors using Claude Code to quickly query stock valuations (e.g., 'Estimate 688017, give PE/PEG/absorb time').
- Not a fit if
- Teams needing fully offline or air-gapped operation; all data comes from external public APIs
- Users wanting a backtesting engine or strategy framework; this only fetches data
- Overseas deployments without a China IP; mootdx times out
- Source review
- 52/100 · Major gaps 3 safety controls not found
What does this agent do, and when should you use it?
a-stock-data is a self-contained Skill file designed for AI coding assistants (like Claude Code, Codex) that aggregates A-share (Chinese stock market) data from 15 public data sources into a directly usable toolkit. It features a 10-layer architecture covering market quotes, research reports, signals, capital flow, news, fundamentals, announcements, limit-up tracking, ETF options, and investor interaction, totaling 47 endpoints. The repository distributes a single SKILL.md file; users copy it to the skill directory and it works. Embedded Python code executes directly, compatible with Claude Code, Codex, OpenClaw, and other context-injection AI tools. It emphasizes zero-auth (except iwencai) and mitigates IP bans through data source prioritization and fallback strategies. Licensed under Apache-2.0, it suits developers needing rapid A-share data for investment research.
The agent reads SKILL.md and exposes data-fetching functions to the AI assistant. Users make natural language requests like 'check 688017 valuation', and the agent parses and calls appropriate functions. Operations include: calling mootdx for K-lines, five-level order books, and tick-by-tick data; Tencent Finance for PE/PB/market cap; Eastmoney endpoints for research reports, dragon-tiger lists, capital flows, margin trading, etc., with all Eastmoney calls rate-limited via em_get() to prevent IP bans; asking iwencai for natural-language research search (requires API key); fetching stock news via Eastmoney and Cailianshe telegraph; retrieving announcements from cninfo; quoting ETF options with Greeks, etc. It includes four built-in research workflows (single-stock valuation, batch comparison, themed research, new ticker research) and provides fallback sources (e.g., official exchange APIs) for degradation.
- Individual investors using Claude Code to quickly query stock valuations (e.g., 'Estimate 688017, give PE/PEG/absorb time').
- Quantitative researchers fetching dragon-tiger list data via natural language to analyze institutional moves (e.g., 'Did 002475 appear on the dragon-tiger list recently? Which brokers are buying?').
- Financial news editors using Cailianshe telegraph for real-time market alerts and monitoring major news.
- Investment analysts using iwencai for themed research (e.g., 'Recent reports on humanoid robotics supply chain, especially screws and reducers').
- Traders monitoring limit-up ladders and themes for short-term decisions (e.g., 'How many limit-ups today? What's the highest streak? What's the break rate?').
- Options traders querying IV and Delta of ETF options (e.g., 'What are the ATM option IV and Delta for 50ETF?').
How do you install or deploy this agent?
Installation: 1) Create skill directory mkdir -p ~/.claude/skills/a-stock-data; 2) Download SKILL.md: curl -o ~/.claude/skills/a-stock-data/SKILL.md https://raw.githubusercontent.com/simonlin1212/a-stock-data/main/SKILL.md; 3) Install Python dependencies: pip install mootdx requests pandas stockstats. For Codex/OpenClaw users, paste the SKILL.md content into your system prompt or project context file. iwencai requires an API key (https://www.iwencai.com/skillhub).
How do you use this agent?
After installation, in Claude Code simply say 'Help me check 688017's valuation' and it activates automatically. Other examples: say 'Which stocks are strong today and what themes?' for hot stocks; 'Is 000858's main capital flowing in or out today?' for capital flow; 'Pull up recent news and announcements for 300476'. All calls are natural language, no command memorization. For Codex/OpenClaw, include the SKILL.md content in context.
What are this agent's strengths and limitations?
- Zero-auth: All data sources except iwencai are free and keyless, out-of-box usable.
- Clear architecture: 10 layers covering everything from quotes to investor interaction; 47 endpoints.
- Built-in fallbacks: If primary sources are blocked, official exchange endpoints and others provide degradation.
- Single self-contained file: Easy to distribute and portable, works offline.
- Relies on multiple third-party data sources; interfaces may change or close (e.g., Eastmoney has rate limits).
- Some sources like iwencai require API key, not completely free.
- Beijing Stock Exchange old code segments are deprecated; new 920 codes needed, otherwise stale quotes.
- Requires domestic IP for mootdx; overseas deployment may timeout.
How does this agent compare with similar options?
In the A-share data space, common alternatives include akshare (Python wrapper) and tushare (requires token). This project removed akshare dependency in V3.0 due to intermediate failure points; tushare needs credits. Compared to commercial terminals like Wind or Choice, this is free but may lack full breadth.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| A-Share Full-Stack Data Toolkit This agent | 52 · Major gaps | Agent plugin / skillFree | ★ 11k | 7d ago | Python | Claude Code |
| OpenBB Open Data Platform | 37 · Major gaps | Library / SDKFreemium | ★ 74k | today | Python | — |
| FinRobot | 51 · Major gaps | CLIFree + model costs | ★ 8.2k | 20d ago | Python | OpenAI API |
| ValueCell Financial Agent Hub | 48 · Major gaps | Desktop appFree + model costs | ★ 11k | 8mo ago | Python | OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: README states zero-auth, no API keys (except iwencai), data sources are public HTTP APIs, no excessive permissions or sensitive data collection; dependencies are minimal (mootdx, requests, pandas, stockstats), no known vulnerabilities or malicious behavior; external effects limited to network requests with fallback strategies; source attribution clear (author, email, X). Deductions: no user confirmation mechanism, data flow transparency partial (how data is used by agent not explained), sensitive data handling not mentioned, rollback mechanism not mentioned.
Evidence: README internally consistent, endpoint count explained; dependency availability discussed (mootdx unmaintained but protocol works, httpx conflict has workaround); failure messages specific (403, HTTP 000, stale quotes). Deductions: no runtime verification, but static review does not require it.
Evidence: clear target users (AI coding assistant users), rich scenarios (valuation, research, fund flows); capability boundaries clear (endpoint list, source priority); trigger descriptions specific (example phrases); environment fit discussed (overseas needs proxy, Python version). Deductions: none.
Evidence: clear information architecture (ten-layer, endpoint list); detailed install notes (3 steps); naming stable (version numbers, endpoint names); examples and FAQ extensive; known limitations documented (BSE old codes, Eastmoney IP bans); License Apache-2.0; CHANGELOG present; maintenance responsibility clear (author, Issue channel). Deductions: none.
Evidence: high output usability (structured data, built-in workflows); high marginal value (integrates multiple sources, fallbacks); reasonable cost-benefit (free, zero-auth). Deductions: none.
Evidence: some claims have specific data (e.g., 2026-07-31 test), but most claims lack independent verification; cross-source corroboration limited (fallbacks mentioned but not verified); fact vs inference separation unclear (e.g., 'no IP ban' is inference). Deductions: lack of traceable test evidence.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Static review cannot verify actual runtime behavior; all claims based on README and code structure.
- Dependency mootdx is unmaintained, potential compatibility issues; monitor alternatives.
- Eastmoney APIs risk IP bans; use rate limiting and fallbacks.
- BSE old codes may cause data errors; use new codes.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v3.10.2 v3.10.2 — PE 消化年数修正（#58）
Release notes excerpt
十五层、87 个入口、34 个来源都不变。只修 full_valuation() 的 PE 消化年数。 修复 full_valuation() 在盈利不增长时把 PE 消化年数报成 0 年（#58） 前向 PE 高于 30x、下一年一致预期盈利持平或下降时，digest_years 原来返回 0，读起来像「已经消化到 30x」， 实际是永远消化不到。同一份文档里的 pe_digestion() 对这种情况本来就返回无穷大，full_valuation() 没有复用它。 现在改为调用 pe_digestion()，得不出有限年数时返回 None（JSON 里是 null）： - 当年 EPS 为正、前向 PE 不高于 30x：0，不变。 - 当年 EPS 为正、前向 PE 高于 30x、预期增长为正：返回消化年数，不变。 - 前向 PE 高于 30x，而下一年预测缺失或盈利不增长：None（原来是 0）。 - 当年 EPS 缺失或不为正：None（原来是 0）。 「完整调研流程」里「流程 B：批量估值对比」的打印对 None 显示「无法估算」，不再打出「消化=None年」。 感谢 @KennyMcSimpson 发现问题并提交修复。 测试 …View on GitHub ↗ -
v3.10.1 v3.10.1 — 成交量单位更正（#57）/ 逐笔提速与盘后核对
Release notes excerpt
十五层、87 个入口、34 个来源都不变，返回的数据也不变。§1.2 只改文档；§1.4 逐笔和 §13.7 期货日 K 修的是 v3.10.0 发版后补做代码复审查出的问题，§1.4 多了一个 frame.attrs["complete"]。 修复 §1.2 tencent_kline() 的成交量单位写错了：科创板是股，不是手（#57） §1.2 的返回列、函数说明，以及「备用源速查」里的腾讯分钟 K 线字段说明，都把成交量写成「手」。 2026-10-05 对照 §1.3 通达信官网盘后包（单位：股）实测：腾讯 K 线对科创板（688 / 689）返回股，其余返回手—— 沪深主板、创业板、B 股、ETF（含科创板 ETF 588）、LOF、REITs 1 手 = 100 股 / 份，可转债 1 手 = 10 张，指数也是手。 日周月线、分钟线、复权与否都是同一规则。 照旧文档换算，科创板的成交量会大 100 倍；备用源速查里「成交额 = 量(手) × 100 × 均价」对科创板、可转债（1 手 = 10 张）也是错的，现按品种写全，指数不适用。 返回值本身不改，免得老用户拿到的数在不知情的情况下变了；只把文档改成实测结果。 §1.4 逐笔的量科创板也是手，那一节原本就写对了，补了一句「和 §1.2 不同」。 …View on GitHub ↗ -
v3.10.0 v3.10.0 — 当日逐笔 / 期货日 K（含大商所）/ 行情层重排
Release notes excerpt
新增 2 个能力入口，来源数不变：十五层、87 个入口（82 主入口 + 5 备胎）、34 个来源。 通达信公开服务器的 K 线 / 盘口 / 逐笔命令 2026-09 起返回空（#52），本版补上逐笔的替代来源，并把行情层的可用来源排到前面。 Breaking Changes - Layer 1 重新编号，可用来源在前、mootdx 放最后留档。旧 → 新： 1.2 腾讯财经 → 1.1、1.5 腾讯 K 线 → 1.2、1.6 通达信盘后包 → 1.3、（新增）腾讯逐笔 → 1.4、 1.3 百度 → 1.5、1.4 新浪复权因子 → 1.6、1.1 mootdx → 1.7。函数名与签名都没变，只影响按 § 编号查找。 本文件更早的条目和 SKILL.md 开头的历史版本说明仍用旧编号。 新增 - §1.4 tencent_ticks(code)：腾讯当日逐笔成交（分笔，约 3 秒一笔），替代失效的 mootdx transaction()。 覆盖沪深个股（含创业板、科创板）与 ETF，只有最近一个交易日，北交所与指数直接报错。 …View on GitHub ↗
FAQ
pip install mootdx causes httpx version conflict; what to do?
pip install --no-deps "httpx>=0.27.1" to upgrade httpx; mootdx still works as it uses TCP protocol, not httpx. Or use a separate venv.