QuantGPT is a hosted market data and backtesting service that you reach through the AI client you already pay for. Connect Claude, ChatGPT or any other MCP client, ask for a backtest in plain English, and the server runs it against its own stored US market history and sends the results back into the chat. The pitch is aimed at traders who want to test ideas without building a research stack. This review covers what the product does, what it costs, where the limits sit and how much of it has been checked by anyone other than the vendor. scored.tools has not run it; everything below comes from QuantGPT's own pages and our scoring data.
What QuantGPT is
QuantGPT LLC sells three things in one subscription. There is a data warehouse of US stock prices and fundamentals, plus a compute layer that runs backtests on QuantGPT's servers. On the Pro plan you also get a hosted runtime for your own trading code. The AI does the asking. Per the homepage, the model sends the rules, the server computes them on stored history, and the numbers come back from the server rather than from the model's memory. That design matters, because a chat model left alone will happily invent a price series.
QuantGPT also says it never decides, sizes or manages a trade. Orders, where there are any, go from your own code to your own brokerage account.
Key Features
MCP connection for Claude, ChatGPT and other clients
The MCP server lives at terminal.quantgpt.co/mcp. Claude and ChatGPT connect with OAuth, and other MCP clients connect with a key. The two named tools are query_warehouse and run_backtest. On the ChatGPT side, the homepage notes that custom MCP apps need ChatGPT's developer mode, which OpenAI offers on Plus, Pro, Business, Enterprise and Education accounts on the web. Free ChatGPT users are out.
The data
The pricing page describes decades of US stock history, with fundamentals joined by filing date, which is the right way to avoid look-ahead bias. Delisted companies are included from 1998; before 1998 the data holds survivors only. The homepage example runs every trading day from 1999 to 2026, and the warehouse is refreshed nightly, per the vendor. The homepage also lists crypto, commodities and FRED macro series, but the equity data is US only. If you trade European or Asian stocks, this is the wrong tool.
Backtests and the strategy library
Both paid plans include 200 backtests a day. The library holds 175 documented strategies drawn from research papers and investor letters, and the vendor says 55 of them run in one click. Examples include the Piotroski F-Score, Greenblatt's Magic Formula and post-earnings drift. You can also build criteria in the web terminal without code.
The homepage demo backtests a 200-day trend rule on the S&P 500 since 1999. QuantGPT reports a worst fall of -24.4% for the rule against -56.8% for buy and hold, with similar annual growth. The footnote says the figures exclude dividends and are gross of trading costs. Treat that as a sales example. A trend filter on one index over one period proves nothing about your strategy.
Journal, forum and extras
Every plan gets a trade journal, shareable results, a forum and a leaderboard. The Monitor section covers end-of-day prices, a call audit, insider filings, a halal screen and valuation. Pro adds an Excel valuation model and private forum rooms.
QuantGPT Cloud (Pro only)
Pro lets you host your own strategy code on QuantGPT's servers, on your schedule, sending orders to your own broker under your own keys. The vendor says broker keys are sealed in the browser and only opened on the server that runs your bot, and that a 30-day paper window is available. We found no public documentation on which brokers are supported or how the runtime handles failures, so ask before you wire real money to it.
Pricing Breakdown
| Plan | Price | MCP limits | Backtests | Main additions |
|---|---|---|---|---|
| QuantGPT | $49 a month ($490 a year) | 500 tool calls a day, 3 keys | 200 a day | Ask AI, library, journal, forum, Monitor, insider filings, halal screen |
| QuantGPT Pro | $149 a month ($1,490 a year) | 1,500 tool calls a day, 5 keys | 200 a day | Pro forum rooms, Excel valuation model, QuantGPT Cloud |
| Business | Custom, from $1,499 a month | Set per customer | Set per customer | Team seats, data licence for your product, direct support |
Both self-serve plans start with a 3-day trial at full daily limits, with a card taken up front. Annual billing saves about 17% (two months free), per the pricing page. Business starts with a sales call.
The real cost is higher than the sticker. You also need a Claude or ChatGPT plan that supports MCP connectors, so budget for that on top. The other catch is that Pro does not raise the backtest cap. You pay three times as much for three times the tool calls plus the hosted runtime, and still get 200 backtests a day. If you only want more backtests, Pro does not give you any.
QuantGPT's own pricing page puts competitor prices side by side. It says QuantConnect lists a $0 tier with unlimited backtesting on one backtest node, with algorithms written in Python or C#. If you can code, that free tier is a serious alternative to a $49 monthly bill.
Pros & Cons
Pros
- Works through MCP, so you keep the AI client you already use.
- Numbers come from the server's computation instead of the model's memory, per the vendor.
- Daily limits for backtests and tool calls are published, so you know where the ceiling is.
- Delisted stocks from 1998 and filing-date fundamentals address two common backtest biases.
- Annual billing cuts the price by about 17%.
- Extras such as insider filings, a halal screen and an Excel valuation model come included.
Cons
- No free tier, and the trial needs a card.
- Equity coverage is US only, and survivor-only before 1998.
- 500 tool calls a day on the base plan can run out fast, since one chat question may trigger several calls.
- Pro keeps the 200-backtest daily cap.
- No public docs page, REST API or SDK that we could find.
- Results shown are gross of trading costs unless stated.
- Almost every claim comes from QuantGPT's own site. Glama's MCP directory lists the connector as untested.
Who Is It For
The best fit is a trader who already pays for Claude or ChatGPT, trades US stocks, and wants to test ideas by asking questions instead of writing Python. Factor and value tinkerers will get the most from the library and the filing-date fundamentals. Pro makes sense if you want a hosted place to run your own bot against your own broker and do not want to manage a server.
It is a poor fit for anyone who trades outside US equities or who needs intraday data (the vendor describes daily prices). Developers comfortable in Python have cheaper routes, starting with QuantConnect's free tier. If your workflow is chart-led, TradingView and TrendSpider are the usual comparisons, though we have not scored them here.
Verdict
QuantGPT scores 5/10 on our rubric. Ease of use gets 6, helped by a simple sign-up and no code required. Pricing value gets 5, held back by the card-up-front trial and the extra AI subscription you need. API integration also gets 5: MCP works with any client, but there are no public docs and no REST API or SDK. Problem fit gets 5 and output quality gets 4, because the only evidence of quality is the vendor's own sample charts.
The design is sound. Pushing the maths to a server and keeping the model on question duty is the right way to use an LLM for backtesting, and the survivorship and filing-date handling show the builders know the usual traps. The pricing is plain. What is missing is outside proof. Our recommendation: if you trade US stocks, already pay for Claude or ChatGPT and do not want to code, take the 3-day trial on the $49 plan and use it to rebuild a strategy you already know well, then compare the output against your own figures. Skip Pro until the base plan's call cap actually blocks you. If you code, start with QuantConnect's free tier first.