CONNECT YOUR AI

Give your AI a door to your desk.

Your AI assistant gets a direct connection to your paper trading desk — it can read what you hold, run honest backtests and manage bots, using a key you scope yourself and can replace at any time.

FIVE STEPS

Set it up in about two minutes.

Step 1What you are actually connecting

Your AI gets a direct connection to your trading desk. Instead of you copying numbers back and forth, it reads the desk itself and does the work where the work lives.

Under the hood this is one address plus two secret headers. Everything your AI does through it goes through the same checks the website uses — same honest fills, same rules, same limits.

Step 2Make your key pair

Go to your desks and open Connect Claude on the desk you want your AI to work on. Pick how far the key may reach, then generate it. You get a key id and a secret.

The secret is shown once. Copy it before you close that window — it can never be shown again, only replaced.

Open my desks and make a key

Step 3Paste it into your AI

The address is the same for everyone. Swap <key id> and <secret> for the two values you just copied.

Addresshttps://agent.quantradin.com/mcp

Claude Code (in your terminal)

Claude Code carries the two headers itself, so one command is the whole setup. Run it, then start a session and ask it to list your desks.

Claude Code

claude mcp add --transport http quantradin https://agent.quantradin.com/mcp \
  --header "X-Vega-Account-Key-Id: <key id>" \
  --header "X-Vega-Account-Secret: <secret>"

Claude Desktop (claude_desktop_config.json)

Claude Desktop's config file only accepts local programs, so it goes through the standard mcp-remote bridge. Paste this into the file and restart the app. Keep the headers written with no space after the colon — that form survives Windows.

Claude Desktop

{
  "mcpServers": {
    "quantradin": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://agent.quantradin.com/mcp",
        "--header",
        "X-Vega-Account-Key-Id:<key id>",
        "--header",
        "X-Vega-Account-Secret:<secret>"
      ]
    }
  }
}

Step 4What your AI can do once it's in

Twenty-two tools reach your desk through that connection. Grouped into what they actually let your AI do, in plain words — each one is a real action on the desk, not a chat about one.

Read your desk

Cash, open positions, every bot you have deployed and the engine's read on each one. Nothing is hidden from it that isn't hidden from you.

Draft a strategy from a description

You describe an idea in English; it turns that into a proper strategy and tells you which parts didn't fit. The draft is only a draft — no number comes from the AI.

Run an honest backtest and read the verdict

The engine grades it on real prices with no look-ahead, and hands back LIVE / SHADOW / SHELF with the numbers behind it. The verdict comes from the engine, never from the AI.

Propose a bot — you decide what it does

It can deploy onto your paper desk and set each bot to live, shadow or shelf, pause it, resume it or retire it. Which of those it may do at all is set by you when you make the key.

Explain what happened

Ask why a bot traded — or didn't — on a given day and it walks the real conditions for that day, one by one, then says it in plain words.

Check itself against reality

It can read how a live bot is doing against what the backtest promised, and where the bot sits on the ladder from backtested to paper-proven.

Step 5What keeps it safe

One key, one desk
The pair opens exactly the desk you made it on. Pointed at any other desk it is refused, the same way the website refuses it.
You choose what it may do
When you make the key you pick its reach: read only, backtest only (the default — it can test but not trade), or allowed to deploy. The limit is enforced on our side, not by asking the AI nicely.
Paper money only
Every desk here trades simulated money on real prices. There is no route from this key to real funds.
One switch stops everything
The agent kill switch in Settings pauses all agent activity at once. There is deliberately no tool for the AI to turn it back on — only you can, from your own signed-in session.
Replace it any time
The secret is shown once. Generate a new pair whenever you want and the old one stops working immediately — your desk, money, positions and bots are untouched.

WHERE IT WORKS

Which apps work today — and which don't.

This connection is carried by two secret headers, so it only works in an app that can send them. The claude.ai website connector signs in with OAuth and the Anthropic API's MCP connector only sends a bearer token — neither can carry these headers today, so neither works yet. We would rather say that than let you spend an hour finding out.

App	Works today?	How
Claude Code	Yes	Sends the two headers itself. One command, above.
Claude Desktop	Yes	Through the mcp-remote bridge, above. The app's own connector screen is OAuth-based and can't carry headers.
Other apps that can pin two fixed headers	Yes	Should work with the same address and headers. We have not tested each one, so we don't claim them.
claude.ai website connectors	Not yet	Signs in with OAuth; there is no field for a custom header. It needs an OAuth door we have not built yet.
Anthropic API MCP connector	Not yet	Only sends a bearer token, never custom headers. Use our normal API from an API-driven agent instead.

Make the key, then say hello.

A good first ask: read my desk and tell me in plain English what I hold and what is deployed.

Go to my desks