You can let an AI read your NinjaTrader 8 accounts, history and strategies without giving it any way to send an order. It isn't a checkbox you have to remember to untick: the code to trade simply doesn't exist.
Why connect an AI to NinjaTrader
Questions like "what time of day do I lose the most money?", "how far am I from the trailing drawdown on each account?" or "what exactly does this strategy I downloaded do?" have answers in your NinjaTrader data. The hard part is getting it out: exporting, cleaning, cross-referencing. A connected AI can do it for you in seconds, in plain language.
The connection uses MCP (Model Context Protocol), an open standard that lets AI apps use external tools. Claude, Copilot in VS Code, Cursor, Gemini CLI, Codex CLI or a local model with Ollama all speak MCP.
The reasonable fear
An AI with access to your trading platform should not be able to trade. Models make mistakes, misread a sentence or follow instructions hidden in a file. If a tool can send orders, sooner or later someone will use it when they shouldn't. That's why we designed NT8-MCP Lite the other way around: first, make trading impossible; then, everything else.
How it's built
- Your AI app talks to a small MCP server running on your PC.
- The server reads the NinjaTrader database read-only (your history works even with NinjaTrader closed) and, for live data, talks to an AddOn inside NinjaTrader.
- Everything happens on
127.0.0.1, your own computer: no cloud, no telemetry and no account passwords asked.
The three layers that make trading impossible
- The MCP server has no order tools. The AI can't ask for what doesn't exist.
- The local client rejects any order command before it leaves the PC.
- The AddOn inside NinjaTrader contains no order code; legacy commands answer "BLOCKED".
All three would have to fail at once for anything to reach the market, and none of them has the code needed to do it.
What your AI can read (37 tools)
- Live accounts: balances, positions, orders and the session's executions, plus the trailing drawdown values prop firms publish.
- Full history: round-trip trades with net P&L, win rate with a confidence interval, profit factor, expectancy, drawdown, streaks, by day and by hour.
- Market and price history: instrument details, trading hours, quotes, bars of any type and statistics such as ATR.
- Your NinjaScript code: read, search and explain your strategies and indicators.
Install in 5 minutes
- You need Windows 10/11 with NinjaTrader 8 and Python 3.10 or later.
- Download NT8-MCP Lite from GitHub ("Code" → "Download ZIP"), unzip it and double-click Instalar.cmd: it creates its own Python environment, writes a random local token, installs the AddOn and registers the server in every AI app it finds.
- In NinjaTrader: New → NinjaScript Editor → F5 (or restart NinjaTrader). If NinjaTrader is open with positions, the installer doesn't touch the AddOn until you're flat.
- Restart your AI app and ask: "Run nt8_status".
What to ask it
- "Analyze my last 30 days on account X grouped by hour. Where am I losing money?"
- "List my open positions on all accounts and the distance to each one's trailing drawdown."
- "Read my MyBreakout.cs strategy and explain the entry logic in plain English."
- "Compare MNQ's 1-minute ATR in the 30 minutes after the open with the rest of the session."
Lite and Pro
| What's included | Lite (free) | Pro |
|---|---|---|
| Tools | 37, all read-only | 88 (they also write code with backups and trade in simulation) |
| Accounts, history, prices, code | Yes | Yes |
| Monte Carlo of your trades and prop firm account simulation | — | Yes |
| Timeline of every order and the reason for each rejection | — | Yes |
| Backtest vault and fast rule-based engine | — | Yes |
| Orders on live accounts | Impossible | Impossible |
Pro also has a practice mode that can only trade NinjaTrader simulation accounts; on live accounts it's still impossible.
Good to know
NT8-MCP is a tool, not an advisor: the AI can misread your data, so double-check anything important in NinjaTrader. Your data doesn't leave your PC through NT8-MCP, but whatever you write to your AI app follows that app's terms: if you use a cloud service, your questions and its answers travel to it; with a local model, they don't.
Educational content based on our own experience. Not investment advice. Firm rules change: check them on each firm's website. This article contains no affiliate links.