Folionomiq Open the app
Folionomiq · for AI assistants

Your AI can explain your portfolio. Now it can ask us to measure it.

Connect ChatGPT, Claude or any assistant that speaks MCP to the same engine the app uses. Ask it about your portfolio; it calls the measurement, then explains the figures in your words. Free, no account, nothing stored.

The address to give your assistant

https://folionomiq.com/mcp

Setting it up

Claude (claude.ai, the desktop app)

  • Open Settings → Connectors → Add custom connector.
  • Paste https://folionomiq.com/mcp as the URL. There is no login and no key.
  • In a chat, turn the connector on and ask, for example: “Measure VOO 60%, BND 40% for me and explain what it says.”

ChatGPT

  • Open Settings → Apps & Connectors (developer mode must be on) and add a new connector with the URL https://folionomiq.com/mcp, no authentication.
  • Then ask about your portfolio in a chat with the connector enabled.

Claude Code, Cursor and other MCP clients

claude mcp add --transport http folionomiq https://folionomiq.com/mcp

Or, in a client's JSON configuration:

{ "mcpServers": { "folionomiq": { "url": "https://folionomiq.com/mcp" } } }

What your assistant can ask for

measure_portfolio
How many independent bets the portfolio really holds, which holding carries risk out of line with its money, what it owns through its funds (and how much of the money that view covers), what past crises did to it in money with the time it took to come back, and the engine's full report, every figure numbered beside the method that produced it.
compare_portfolios
Up to eight portfolios measured over one shared window.
test_a_change
A change measured before it is made: the portfolio now and after, on the same history. The “after” is counterfactual, and nothing says whether to make it.
check_before_you_buy
One addition measured on the portfolio it would join: how much of it is already owned (for a fund, from its SEC filing), whether it adds a bet, and what past crises would have done to the money with it. It never says whether to buy.
living_off_it
Yearly withdrawals, raised with real inflation, replayed on the portfolio's own history from every month it allows as a start: how often the money lasted, when it ran out, and the highest rate that never did. Replayed, not a forecast.
portfolio_structure · look_through · past_crises · portfolio_report
The parts of the measurement one at a time, when only one question was asked: the real bets and who carries the risk, the companies under the funds, the crises replayed, the numbered report.
factor_exposure · currency_effect · health_axes
Sensitivity to the market, size, value and momentum; what exchange rates did to the money; the portfolio read along separate axes, with no overall score.
sweep_holding · reverse_stress · return_history · plain_findings
One holding at every weight; the market moves a given loss would take, with no probability attached; the record month by month; a few findings in plain words.
last_price · asset_info
A latest close, to turn shares into an amount; one fund or share, its cost and what it holds.
simulate_paths · walk_forward
Paths resampled from the portfolio's own record, labelled simulated and refused when the record is too short; and an allocation method tested on prices it had not seen, which returns the record and never a mix to hold.
search_symbols · validate_symbols
Find a ticker by name or ISIN, and check which symbols have usable history.

What it will not do

Every answer is descriptive: measured from end-of-day prices over a stated window. It does not forecast prices, returns or the probability of either, and it does not say what to buy, sell or hold. Each answer carries those instructions for the assistant reading it; we cannot enforce what an assistant then says, and we say so rather than pretend.

The funds' holdings come from what they publish, usually their largest positions, so the companies a portfolio owns through its funds are minimums, and every answer says how much of the money they cover.

Your holdings are measured and never stored, exactly as in the app. The request carries your assistant's address, not yours. See the privacy section and what you may do with the results in section 5.

Questions, or an assistant it does not work with: [email protected]