QuantGPT
co.quantgpt由 quantgpt.co 发布官方发布者v1.0.1更新于 Oct 24, 2026
The data and trading platform for AI: decades of US market data, backtests in seconds.
概览
让 AI 助手接入 QuantGPT 的美股数据与研究平台,用于回测、股票评级、因子与宏观查询以及组合研究。
- 可以做什么
- 通过托管端点提供 76 个工具(68 个已启用):run_backtest 与 run_sweep 在数十年美股数据上做前向回测,grade_stock 与 grade_table 给出 A+ 到 F 评级,query_warehouse 对研究仓库执行只读 DuckDB SQL。还覆盖组合构建、宏观序列、加密资产、预测市场与事件,并提供 search 与 fetch 以引用策略库和论坛记录。发现层(available_categories、activate_category、call_tool)可展开更多工具类别。
- 适用场景
- 当助手需要用可复现的证据回答量化市场问题时适用:检验策略想法、筛选或评级美股、查询因子、基本面或宏观数据,或审视投资组合。它面向研究与教育,而非实盘下单,因为没有任何工具会向券商下单或改单。免费方案整个账户仅 5 次 AI 调用。
- 运行要求
- 远程 streamable HTTP 端点 OAuth 2.1 登录(Claude、ChatGPT、Gemini 应用),或以 Authorization: Bearer 或 X-API-Key 头传入 QuantGPT 密钥(qg_sk_...)。免费方案无需信用卡;付费方案每月 49 或 149 美元。需要能访问该端点的网络。
安装
在 SourceWeft 中
- 打开 控制台中的 QuantGPT,将其添加到工作区。
- 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
- 为需要使用其工具的对话启用该服务。
Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。
交给智能体安装
把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。
阅读 https://sourceweft.com/mcp/co.quantgpt%2Fmcp/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。其他 MCP 客户端
把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。
{
"mcpServers": {
"quantgpt": {
"type": "http",
"url": "https://terminal.quantgpt.co/mcp"
}
}
}README
QuantGPT (quantgpt.co)
QuantGPT (quantgpt.co) is the data and trading platform for AI. This is the official repository of QuantGPT LLC.
Not affiliated with other GitHub projects named QuantGPT. The product, its hosted MCP server and this repository are QuantGPT by QuantGPT LLC, at https://quantgpt.co.
The data and trading platform for AI: decades of US market data, backtests in seconds.
Claude and ChatGPT connect over MCP to decades of US market data, and backtests run in seconds on QuantGPT Cloud. On QuantGPT Pro, your own strategy runs there too and trades in your own brokerage account. Start with a 3-day free trial of QuantGPT or QuantGPT Pro, cancel any time.
Updated October 2026. This repository holds no server code. It is the public listing for the hosted server: how to connect each client, what every tool does, example prompts, plans and limits.
What your AI can do with it
- Backtest any idea in seconds. Describe a strategy in plain words; your AI turns it into criteria (up to 8 filters, a rank, a holding count) and runs it walk-forward against a benchmark on decades of US market data, with a cost model and a decade-by-decade table. Up to 100 variants run in one call.
- Grade any US stock. A+ to F on six dimensions against the whole market, point in time, with the drivers, the peers and a chart pack. Up to 25 tickers in one call, or the leaderboard for a sector.
- Query the data itself. Prices, total returns, fundamentals joined by filing date, 74 backtest factors, FRED macro, insider filings and short interest, through ready-made calls or read-only SQL over the whole warehouse.
- Start from the literature. 175 documented strategies from papers, books and practitioners, 53 of them runnable in one call.
- Work on a portfolio. A frontier over your tickers, a review of your holdings, the trade list from your criteria to the book they imply today, paper books marked nightly, and alerts when a name enters or leaves a screen.
- Go past US stocks. Macro regimes, commodity spot series, weather studies, crypto with funding rates, memecoins and the 24/7 stock tape. Prediction markets are coming soon.
- Chart and cite. Branded charts with the numbers that drew them, and
searchandfetchso ChatGPT deep research can cite the library, the studies and the forum. - Run your own strategy on QuantGPT Pro. QuantGPT Cloud hosts your strategy as your own code, on your schedule, and it trades in your own brokerage account under your keys. You set it up in the terminal; no MCP tool places an order.
Contents
- Connect your AI
- What your AI can call
- Strategies you can ask about
- Plans and limits
- What the data is, and is not
- Disclaimer
Connect your AI
Claude, ChatGPT and the Gemini app sign in to QuantGPT through OAuth: no key to paste. The connection lands on your account page as a key labelled with the client's name, and you can revoke it there. Claude Code, Gemini CLI, Cursor and any other client below use a key.
Get a key
- Sign in at https://terminal.quantgpt.co (new here? start a 3-day free trial).
- Open your account page. Under Keys for your AI, type a label and press Issue key.
- Copy the
qg_sk_...value now. Only its hash is stored, so it is shown once.
Claude (claude.ai and Claude Desktop)
- Customize > Connectors, then + > Add custom connector (on Team or Enterprise an Owner may need to add it first).
- Name it QuantGPT, paste
https://terminal.quantgpt.co/mcp, leave the OAuth client fields empty, and click Add. - Press Connect. Claude opens QuantGPT's sign-in page: sign in and approve. The connection shows on your account page as
connector: Claude.
Guide on quantgpt.co: https://quantgpt.co/learn/connect-claude.html
Claude Code
Add --scope user to have it in every project. Then ask Claude Code to call describe_capabilities.
ChatGPT
- In ChatGPT, Settings > Security and login > turn on Developer mode. OpenAI offers it on Plus, Pro, Business, Enterprise and Education accounts, on the web.
- Go to chatgpt.com/plugins and select the plus button. Name the app QuantGPT and, under Connection, enter
https://terminal.quantgpt.co/mcp. - ChatGPT registers itself and sends you to QuantGPT's sign-in, now or on its first call. Sign in and press CONNECT; there is no key to paste. The connection shows on your account page as
connector:plus the name ChatGPT registered. In a chat, choose Developer mode from the + menu and select QuantGPT.
On a Business, Enterprise or Education workspace, developer mode can depend on the workspace's policy, so an admin may have to allow it first. ChatGPT deep research works through a connector's search and fetch tools; QuantGPT ships both, so deep research can search the library, the studies and the forum and cite what it used.
Guide on quantgpt.co: https://quantgpt.co/learn/connect-chatgpt
Gemini
Gemini CLI, with a key:
-s user saves it in ~/.gemini/settings.json in your home folder. Without it, Gemini CLI saves to the current project's .gemini/settings.json, where a key can end up in a git repository.
Or add it to ~/.gemini/settings.json (every project) or .gemini/settings.json (one project):
Run /mcp inside Gemini CLI to see the server and its tools.
The Gemini app: at gemini.google.com open Settings > Connected Apps, and under Custom apps choose Add a custom app. Paste https://terminal.quantgpt.co/mcp, then sign in to QuantGPT. Google decides who can add custom apps (in September 2026: 18 or over, in the US, signed in with a personal Google Account, Keep Activity on, set up in the web app): https://support.google.com/gemini/answer/17209137
Guide on quantgpt.co: https://quantgpt.co/learn/connect-gemini
Cursor
In .cursor/mcp.json (one project) or ~/.cursor/mcp.json (every project):
Set QUANTGPT_KEY to your qg_sk_... key in the environment Cursor starts from, so the key never sits in the file.
Guide on quantgpt.co: https://quantgpt.co/learn/connect-cursor
Any other MCP client
A request without a valid key gets 401 and a WWW-Authenticate header that points OAuth clients to https://terminal.quantgpt.co/.well-known/oauth-protected-resource.
No connector option on your plan? The setup prompt at https://terminal.quantgpt.co/pastepack.txt gives any chat AI the same context in one block you paste, without MCP. Signed-in members find the Claude and ChatGPT steps, this address and the prompt library at https://terminal.quantgpt.co/connect.
Your first call
Ask your AI to call describe_capabilities: it returns the whole map (every factor, operator, benchmark, library category and methodology rule). On connect the server lists 14 core tools (15 once pre_trade_brief opens) and 5 discovery tools; the other 57 sit in seven categories. If your AI asks where the market stands, market_snapshot answers in one call. Your AI calls available_categories, then activate_category, to list a category for your key, or runs any tool by name through call_tool. With a key-based client you can add ?tools=all to the URL to list every tool at once.
What your AI can call
77 tools, 69 of them switched on today; the 8 marked (coming soon) answer with a coming-soon note until they open. None of them places, modifies or cancels an order at any broker. A few write to your own QuantGPT account: saved studies, paper books, sealed books, alerts, event rules and the panels on your terminal.
Listed on connect (15)
Discovery (5)
Equities: category equities (21)
Portfolio: category portfolio (8)
Macro: category macro (4)
Alternative markets: category alt_markets (13)
Events: category events (3)
Charts: category charts (3)
Your terminal: category desk (5)
Strategies you can ask about
A sample from the library of 175 documented strategies (53 of them runnable), drawn from papers, books and practitioners. They are listed so you can test them, not as promises. A library run is the closest the warehouse's factor vocabulary gets to the published rule, and every run says what it approximates and how it differs from the source. A name that starts "Inspired by" marks a run that only loosely follows its source. Run one and read its losing years before you trust it.
Ask: "Run the library strategy piotroski-2000, show me its losing years and its worst drawdown, and tell me what the run approximates from the paper." More prompts: examples/prompts.md, examples/first-session.md and the Strategy Library walk-throughs in examples/library/. Browse the whole library at https://quantgpt.co/library/.
Plans and limits
- QuantGPT and QuantGPT Pro start with a 3-day free trial: card up front, cancel any time. The trial has the plan's full daily limits.
- An AI call is one tool call. Connecting, listing tools and pings do not count. The count is per account, across all of its keys.
- The daily count resets at 00:00 UTC.
- A per-minute limit also applies: 120 requests a minute per account. Past it the server answers
429; try again a minute later. - Some tools depend on the plan:
portfolio_reviewopens on QuantGPT Pro,count_betson QuantGPT and QuantGPT Pro,strategy_alertsallows 3 alerts on QuantGPT and unlimited alerts on QuantGPT Pro, andpaper_booksallows 1 book on QuantGPT and 5 on QuantGPT Pro. - Teams: Business is custom (seats, limits built for your volume) and starts with a conversation at https://quantgpt.co/contact.
Plans: https://quantgpt.co/#plans. Start: https://terminal.quantgpt.co/join.
What the data is, and is not
- Companies that later delisted are only partly in the market-wide backtest universe from 1998: some have no prices or no factor data, and each market-wide result states the share it measured. Before 1998 the universe holds survivors only, so a market-wide result leans toward survivors. Criteria backtests start in 1998 unless you choose an earlier year. A backtest on a list of tickers you choose, and a rule on a single index, are not affected. The server tells your AI to say so whenever it reports a market-wide result.
- Fundamentals are joined by filing date. About 9% of rows still carry the quarter-end as the filing date, so those quarters are seen early; a fix is in progress.
- Factor data begins in 1990, the earliest start year a criteria backtest accepts. Macro series are the latest vintage; revisions are not modelled.
- Results are gross by default. Every
run_backtestresult also carries turnover and net figures at 5, 10, 25 and 50 bps per side. - Every
run_backtestresult returns its worst drawdown, a decade-by-decade table and the full return series, so the losing years are in the result; its chart pack draws them year by year. - The same criteria return the same numbers, to the last digit.
- Futuretests replay the record of past episodes. They do not forecast.
- Public data checks, no account needed: https://terminal.quantgpt.co/verify
Disclaimer
QuantGPT is research and education software. Nothing it returns is investment advice or a recommendation to buy, sell or hold anything, and nothing is tailored to you. Every backtest, study, screen, futuretest and library entry is hypothetical and describes the past; past performance, real or simulated, does not indicate or guarantee future results.
No tool on this server places, modifies or cancels an order, and none connects to a broker. Where a tool produces a trade list or a signal, you or your own software act on it, at your own broker, under your own keys. On QuantGPT Pro, QuantGPT Cloud hosts your own strategy as your own code and sends the orders your code makes to your own brokerage account, under your keys; QuantGPT never decides, sizes or times a trade. That runs in the terminal, not through this server.
Full terms: https://quantgpt.co/terms. Risk and methodology disclosures: https://quantgpt.co/disclosures. Privacy: https://quantgpt.co/privacy. Security: https://quantgpt.co/security.
QuantGPT LLC. Questions: https://quantgpt.co/contact.
来源:README.md,提交 a3e0c58
工具
0版本历史
1- v1.0.1最新Oct 13, 2026
评分
已满足适用检查项的 85%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通必需25 分
- 已声明认证方式必需10 分
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分