Getting Started
You need:
a QuantPilot account with access to the app
an AI client that supports remote MCP servers: Claude, Claude Code, ChatGPT or Codex
the QuantPilot MCP server URL:
https://mcp.quantpilot.com/mcp
đĄ Using QuantPilot via MCP does not currently consume QP tokens. Your connected AI client's own usage limits and charges still apply.
âLearn more about QP tokens
Connect Claude (web and desktop)
Custom connectors are available on every Claude plan. The Free plan allows one custom connector.
Open Customize, then Connectors.
Click + Add, then Add custom connector.
Enter
QuantPilotas the name, paste the MCP server URL and click Continue.Under Authentication, select Sign in now. Under OAuth client, keep Use Claudeâs published identity (Recommended).
Click Add, then sign in with your QuantPilot account and allow access.
In a chat, click + in the lower left, open Connectors and turn on QuantPilot.
On a Team or Enterprise plan, an Owner adds the connector first in Organization settings > Connectors. Then each member opens Customize > Connectors, finds QuantPilot and clicks Connect.
Connect Claude Code
Run:
claude mcp add --transport http --scope user quantpilot https://mcp.quantpilot.com/mcp
--scope usermakes QuantPilot available in all your projects. Without it, QuantPilot is added to the current project only.Start Claude Code and type
/mcp.Select quantpilot and follow the steps in your browser to sign in with your QuantPilot account.
Connect ChatGPT
Custom MCP servers are not available on every ChatGPT plan, and workspace admins can turn them off.
Open Plugins in ChatGPT.
Click the plus button, then Add custom MCP server.
Enter
QuantPilotas the name and a short description.Under Connection, enter the MCP server URL.
Choose OAuth for authentication, review the warning and select I understand and want to continue.
Select Create as a plugin. When ChatGPT asks, sign in with your QuantPilot account and allow access.
In a new conversation, type
@and select QuantPilot.
ChatGPT asks you to confirm before it starts a backtest, because starting one is not a read-only action.
Connect Codex
Add the server:
codex mcp add quantpilot --url https://mcp.quantpilot.com/mcp
Sign in:
codex mcp login quantpilot
A browser window opens. Sign in with your QuantPilot account.
In Codex, type
/mcpto check that quantpilot is connected.
In the Codex IDE extension, open the gear menu, select MCP servers, then Add server. Enter the URL, save, select Restart extension, then Authenticate.
How to use it
For the full list of what a connected client can do, see What the QuantPilot MCP server does.
To connect your AI client, see Use QuantPilot from Claude, ChatGPT and other AI clients.
Describe the strategy in plain words. Include the symbol, timeframe, entry and exit rules, the backtest period and your starting capital. For example:
Write a QuantScript strategy that buys BTC on Hyperliquid when the 1h RSI(14) drops below 30 and sells when it rises above 70. Backtest it from January 1 to June 30, 2026 with 1,000 USDC.
Show my last five backtests and compare their returns.
Why did this backtest place no orders? Check its orders and fix the strategy.
The client validates the code, starts the backtest, checks its status until it finishes, then reports the results. If your starting capital is not in the prompt, it asks for it.
What to expect
Exchanges. Backtests use Hyperliquid by default, with capital in USDC. As of November 2026, Binance is available for backtests only, when you ask for it, with capital in the symbolâs quote currency, such as USDT.
Data. Strategies can read candles from 1m to 1w and symbol specs for spot and perpetual markets. As of November 2026, funding rates, open interest and order book data are not available.
Backtest period. The end date must be at least one hour in the past. If the start date is too early for the indicators to warm up, the client suggests the earliest start date that works.
Live trading. As of November 2026, the MCP server runs backtests only. It cannot start paper or live strategies.
Note: Backtests run from an AI client are not attached to a strategy, so they do not appear in the strategy lists in the QuantPilot app. You see their results in your AI client.
What a connected client can and cannot do
A connected client acts as your QuantPilot account. Every connected client gets the same rights: there is no read-only connection, and no way to allow some tools and not others.
A connected client can:
read the QuantScript documentation and market data: markets, symbols and the dates they have data for
validate strategy code
run backtests under your account, on the same engine as backtests in the QuantPilot app
read the results, orders and code of the backtests it ran, and of your other backtests that are not attached to a strategy or a chat
A connected client cannot:
place real orders, or start a paper or live strategy
move funds, or read or change your exchange connections, API keys or wallets
read or change the strategies, chats, research projects or settings in the QuantPilot app
read backtests that belong to a strategy or a chat in the app
see your conversation with the AI client: QuantPilot receives only what the client sends to a tool
Important: Anyone who can use your AI client with this connection can run backtests as you. Do not share access to that client.
To stop a client from using your account, remove QuantPilot from that clientâs connectors, plugins or MCP servers.
FAQ
How do I get the latest version of the QuantPilot tools?
Most updates reach you automatically. When an update changes the tools themselves, your client may keep the old list until you refresh it:
Claude Code: type
/mcp, select quantpilot, then Reconnect. A new session also picks up the changes.ChatGPT: open Plugins, select QuantPilot, click Refresh, then start a new conversation.
Claude: start a new chat. If the tools still look outdated, disconnect QuantPilot in Customize > Connectors and connect it again.
Codex: start a new Codex session. In the IDE extension, select Restart extension.
The client asks me to sign in again
Your sign-in has expired. In Claude Code, type
/mcp, select quantpilot, then Re-authenticate. In other clients, reconnect QuantPilot in the connector or plugin settings.The client says my account has no access to QuantPilot
The connection only works for accounts with access to the QuantPilot app. If access was granted recently, disconnect QuantPilot in your client and sign in again.
A backtest has been running for a long time
Backtests run in the background. Ask the client to check the backtest again in a minute.
The backtest stopped with a platform error
If the error says the run was stopped or interrupted on our side, run it again without changing the code.
The backtest finished but placed no orders
Ask your AI app to inspect the strategy code, backtest orders and any returned warnings. A finished test with no simulated trades may have no trading statistics. Common causes are listed in When a strategy does not work.
Where can I find the results of my backtest?
Backtests run through QuantPilot MCP are saved and can be retrieved through your connected AI assistant. MCP backtests do not currently appear in the QuantPilot web app. You can still access it through your AI assistant.
To retrieve a previous result, try asking:
"Show my saved QuantPilot backtests"
"Show the results for backtest.."
You can also retrieve saved results from a new conversation using the same QuantPilot account.
What usage limits apply?
The following limits apply per QuantPilot account when using MCP:
⢠60 tool calls per minute
⢠30 backtests started per hour
⢠200 backtests started per 24 hours
⢠3 backtests running at the same time
These limits are shared across all AI clients and conversations connected to the same QuantPilot account. Starting a new conversation or switching clients does not reset them.
A tool call is a request your AI client makes to QuantPilot, such as validating a strategy, starting a backtest or checking its status. One chat message may trigger several tool calls.
If you reach a tool-call or backtest-start limit, follow the wait time returned to your AI client before retrying. These allowances replenish gradually.
If you already have three backtests running, wait for one to finish before starting another.
Backtesting capacity is also shared across MCP users. During busy periods, you may need to try again later even if you are within your account's limits.
Need help?
Contact support as described in How to Contact the QuantPilot Support Team, and include:
the AI client you use and its version
the prompt you sent
the backtest ID, if there is one
the error message you see
the date and time when the issue happened
