Ask Claude about your portfolio.
Connect QuantStark to Claude, Cursor or ChatGPT and ask in plain words: how is it doing, where is the risk, which bets did you not mean to take. The answers come from the same factor model and diagnostics that run in the dashboard, computed on your own holdings, not guessed.
https://mcp.quantstark.com/mcpRead-only. Included with every account. Works with Claude, Claude Desktop, Claude Code, Cursor and ChatGPT.
Up 59.4% against 69.9% for the S&P 500 since January 2022, Sharpe 0.73, worst drawdown 22.7%. It trailed, but not by taking more risk: beta is 0.83 and predicted volatility 12.0%.
Most of that risk is factor-driven (10.6% of the 12.0%), and Technology is the largest factor after the market. The book is 49% tech against 37% for the index, with VTI and XLK looked through.
Two things worth a look: 11% is sitting in cash, and NKE and DIS fell 76% and 28% over the period.
Figures from the connector for a 16-holding sample portfolio, as of 2026-09-14. Answer shortened. Every answer states its as-of date.
01 Setup
Connect in three steps.
No code, nothing to install. Sign in once, approve once, then ask.
Have a QuantStark account
Sign in and load at least one portfolio, by file upload or a linked brokerage. The connector reads the same books you see in the dashboard.
Add the connector in your assistant
Claude and Claude Desktop: Settings, then Connectors, then Add custom connector. ChatGPT: Settings, then Connectors, then Create, with developer mode on. Cursor: add an MCP server with a url field. Paste the address in each case. No API key, nothing to install.
https://mcp.quantstark.com/mcpApprove, then ask
You are sent to QuantStark to sign in and see exactly what the app will be able to read. Approve, and you are back in your assistant, ready to ask.
In Claude Code, add it from the terminal instead, then run /mcp to sign in.
claude mcp add --transport http quantstark https://mcp.quantstark.com/mcp02 For investors
Every chapter of the dashboard, one question away.
Ask in whatever words you use for it. The connector routes the question to the right tool and answers with your own numbers.
“How is my Fidelity portfolio doing against the S&P 500 this year?”
Your portfolio
“What are my biggest style tilts, and which ones are bets I did not mean to take?”
Your portfolio
“Which three positions drive most of my predicted risk?”
Your portfolio
“Explain the concentration finding on my main account in plain words.”
Your portfolio
“What is my max drawdown over the last year?”
Your portfolio
“Compare the factor exposures of my two accounts.”
Your portfolio
“What kind of company is ASML by the numbers: size, value, momentum?”
Market, no portfolio needed
“Find large-cap healthcare names with low volatility and a value tilt.”
Market, no portfolio needed
“Compare VOO and QQQ on look-through factor exposures.”
Market, no portfolio needed
Factor directions are stated in words, so a high Size exposure reads as “larger companies”, never as a number you have to interpret. Screener results are candidates for research, not recommendations.
03 Playbooks
Two flows to start with.
Each one is a prompt you can paste as-is. The answers below are what the connector returned for a 16-holding sample portfolio, shortened.
Before adding a name, find out whether it adds to a bet you already have.
I am thinking of adding 5% NVDA to my portfolio. Does that add to a bet I already have?Yes. NVDA is already 3.0% of the book, next to AAPL 13.5%, XLK 12.7%, MSFT 4.9% and CRM 1.8%. Technology is 49% of the portfolio against 37% for the S&P 500 with the ETFs looked through, and it is the top concentration finding. Another 5% takes it to about 54%.
It also moves the style tilts toward NVDA rather than away. NVDA reads Growth +0.67, Value -0.40, Leverage -0.67. The book today sits slightly on the other side of growth (-0.07) and value (-0.04), so the add pulls both toward NVDA's profile.
If the goal is diversification, this is not it. If you want more of the tech bet knowingly, you now know its size.
A read of your exposures, not a recommendation. Nothing here places or suggests a trade. Sample portfolio, figures as of 2026-09-14.
Screen the market for what your portfolio is short of, not for what is trending.
Find large, steady, reasonably valued names outside Technology that would balance my portfolio.The book already leans large and steady (Size +0.11, Volatility -0.12) with modest tilts elsewhere, so the imbalance is sector, not style: Technology at 49% against 37%. The screen kept Healthcare, Financial Services, Industrials and Consumer Defensive, and asked for Value above +0.3, Volatility below -0.2 and Size above +0.3.
Eight names came back: BRK.B, UNH, BAC, C, WFC, RY, HSBC and JPM.
Two things to see before doing anything. JPM is already 4.8% of the book. And six of the eight are banks, so the screen found a sector, not eight independent ideas; UNH and BRK.B are the two that are not.
Candidates for research, not recommendations. The screener ranks by the criteria you set; it does not know your goals or taxes. Sample portfolio, figures as of 2026-09-14.
04 Under the hood
Ten tools, all read-only.
The assistant picks the tool; you never name one. Seven read a portfolio you have approved access to, three read public securities data and need no portfolio at all.
- List portfolios
list_portfoliosYour saved portfolios and their ids. The assistant calls this first when you name a book.
- Analyze portfolio
analyze_portfolioThe overview: return against benchmark, risk band, value and P&L, factor coverage, and the top diagnostic findings.
- Get holdings
get_holdingsWhat a portfolio holds, heaviest first, with weights.
- Get performance
get_performanceRealized returns and risk-adjusted measures: total and money-weighted return, volatility, max drawdown, Sharpe, tracking error.
- Get predicted risk
get_riskForward-looking volatility, tracking error, beta, and which factors contribute most. Predicted, not realized.
- Get factor exposures
get_factor_exposuresHow a portfolio tilts on Value, Size, Momentum, Volatility, Leverage and Growth, or on industries, against its benchmark.
- Explain a finding
explain_findingOne diagnostic finding from the overview, explained with the portfolio's own numbers.
- Get security profile
get_security_profileWhat a single stock or ETF is: name, sector, market cap, and style factor exposures. Needs no portfolio.
- Screen stocks
screen_stocksStocks matching sector and style-factor criteria across the universe. Candidates for research, never buy recommendations.
- Compare ETFs
compare_etfsTwo ETFs side by side on look-through factor exposures and holdings coverage.
05 Privacy and control
Your data, your say.
Read-only, by construction
Every tool reads. None can add, edit or delete a portfolio, place a trade, or touch a brokerage connection. The connector is built on the same read paths the dashboard uses.
You approve, you revoke
Access is granted on a consent screen on quantstark.com, never by pasting a key into another app. Revoke any connection from Account, then Connections. Revocation takes effect within about a minute, and changing your password revokes every connection at once.
Two scopes, stated plainly
Portfolio access lets the app read every portfolio on your account, including values in dollars and which broker each one syncs from. Market access covers only public securities data. The consent screen names both.
As-of dates are stated, not hidden
Figures are as of the last nightly data run. A long conversation can cross that refresh, so two answers may carry different as-of dates. Each answer says which date it used.
The data-handling terms for connected apps are in section 6 of our Privacy Policy. Infrastructure and encryption details are on the Security page.
06 Questions
Before you connect.
- Does the assistant see my brokerage login?
- No. Brokerage credentials never touch QuantStark in the first place, and the connector reads only the portfolio data already in your account. Nothing about a brokerage connection is exposed beyond the broker's name.
- Does QuantStark see my conversations?
- No. We record when each connection was created and last used, and nothing else. What you ask and what the assistant says stays between you and the assistant, under that provider's privacy policy.
- Can it trade or change my portfolio?
- No. There is no write path. Anything that changes a portfolio, including the What-if sandbox, still happens in the dashboard with you at the controls.
- How fresh is the data?
- Prices, factor exposures and risk figures refresh once a day after the market close. Brokerage holdings refresh on the same nightly pull. Every answer states its as-of date, so a stale figure is never silent.
- Does it work with assistants other than Claude?
- Yes. MCP is an open standard, and the connector speaks both ways a client can identify itself: the published-identity route Claude uses and the registration route Cursor and ChatGPT use. Claude, Claude Desktop and Claude Code are the ones we use every day. In ChatGPT the connector appears with developer mode on.
- Is it included with my account?
- Yes. There is no separate plan. If you can sign in to the dashboard, you can connect.
Connect once. Ask anything about your own book.
Included with every account. Read-only. Revoke any time.