Product

Ask your AI assistant about your portfolio

Your dashboard knows what your portfolio did. Your AI assistant knows how to explain things, compare them and put them in a table. You can connect the two: create a token in BullBenchmark, paste it into Claude Desktop, Cursor or ChatGPT, and ask your questions there instead of reading them off a chart.

The connection is read-only by design. There is no tool that changes a setting, no tool that uploads a file, no tool that places an order, and no way for the assistant to reach somebody else's account.

This article explains what the connection can see, what it cannot do, and how to set it up in each of the three assistants.

What the connection is

BullBenchmark speaks MCP, the Model Context Protocol, an open standard for giving a language model a set of tools it may call. Our endpoint offers six tools, all of them read-only, and all of them scoped to the one account your token belongs to.

The verdict. Your return against the S&P 500 and the Nasdaq 100 over the same window, per year and in total.
The summary. Value, money deposited, price result, dividends received, costs, total return, base currency.
The positions. Name, ticker or ISIN, shares, value, weight, return, how each one did against the S&P 500, plus your own group and label.
The dividends. Received per month, expected over the next twelve months, yield on cost, growth against the year before.
The performance. Time-weighted return against the index over 1m, 3m, 6m, 1y, 3y and the full history, and your drawdown.
The uploads. The names, dates and row counts of the files you imported.

Those six tools are the complete surface. The numbers are the same ones your dashboard shows, calculated by the same code, so a figure your assistant reads out is the figure on your screen.

What to know before you create a token

The model runs on your own AI subscription, not on ours, so the connection costs nothing extra at BullBenchmark. The other side of that is that whatever you ask about travels to your own AI provider and stays with them under their terms.

If you would rather not hand your portfolio to a third party, the dashboard works as before without a token. The feature is off until you switch it on, and revoking a token on the Account page stops the connection immediately.

Two smaller limits apply. The connection needs an active subscription or trial, the same as the dashboard, and a tool call returns a plain "subscription needed" message once that runs out.

There is also a rate limit of sixty calls a minute per account and thirty per token, which a normal conversation does not reach but a script in a loop does.

How to create a token

Open the dashboard and click Account.
Find "Connect an AI assistant".
Give the token a name you will recognise later, such as the machine it will live on, and create it.
The value starts with "bbm_" and is shown once. We store only a hash of it, so if you lose it you create a new one rather than looking the old one up.

An account can have up to five active tokens, and each one can be revoked on its own.

Claude Desktop

Open the configuration file from the developer settings in Claude Desktop.
Add the block below, with your own token in place of the placeholder.
Restart the app.

This setup needs Node.js installed, because mcp-remote is the piece that carries the authorization header for you.

{
  "mcpServers": {
    "bullbenchmark": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://bullbenchmark.com/api/mcp",
               "--header", "Authorization: Bearer bbm_your_token_here"]
    }
  }
}

Claude in the browser works differently for now. A connector added there expects to sign you in through OAuth, which this endpoint does not offer yet.

A token alone therefore does not get you in through the browser. OAuth support is on the list, and Claude Desktop works today.

Cursor

Cursor speaks the same protocol and can carry the header itself, so no extra tooling is needed. Put the block below in .cursor/mcp.json in a project, or in ~/.cursor/mcp.json to have it available everywhere.

{
  "mcpServers": {
    "bullbenchmark": {
      "url": "https://bullbenchmark.com/api/mcp",
      "headers": { "Authorization": "Bearer bbm_your_token_here" }
    }
  }
}

ChatGPT

In ChatGPT, connectors you add yourself sit behind developer mode, and which plans include it has changed more than once, so check what your own account offers before you start.

When you can add one, it needs the address of the endpoint and an authorization header that carries your token.

https://bullbenchmark.com/api/mcp
Authorization: Bearer bbm_your_token_here

What to ask once it works

The useful questions are the ones that are tedious to answer by clicking. Four examples of the kind of question the connection is built for.

Which of my positions beat the S&P 500 over the money they held, and which did not?
Group my holdings by sector and tell me what each group contributed.
How much dividend came in per month last year, and what do the next twelve months look like at the current share counts?
How deep was my worst drawdown, when did it happen, and how long did it take to recover?

The connection does not tell you what to buy, and neither do we. It reads what your portfolio did, and the conclusions stay yours. If your assistant volunteers a recommendation anyway, that comes from the model and not from your data, and the model knows nothing about your circumstances.

A note on the numbers

Everything the assistant reads comes out of the same engine as the dashboard, so the same limits apply. Returns are time-weighted, which means deposits and withdrawals never count as performance and received dividends do.

Prices refresh every hour while markets are open and settle on the official closing prices after the US close, so a figure quoted mid-session is as fresh as the last refresh.

The series are thinned before they leave. A chart of twenty years is thousands of daily points, and the connection sends at most twenty-four per series, enough to describe the shape without filling your assistant's context.

In short

The connection is read-only and offers six tools, all scoped to the account your token belongs to.
The model runs on your own AI subscription, and what you ask about goes to your own AI provider.
A token is created on the Account page, is shown once, and can be revoked there at any time.
The numbers are the same as on the dashboard, with the same hourly prices and the same time-weighted method.

BullBenchmark reads the transaction export from your broker and shows your return next to the S&P 500 and the Nasdaq 100, fed with the same deposits on the same dates, in your own currency. The first two weeks are free and no card is needed.