The Read-Only AI Trading Agent: The First Workflow Every Funded Trader Should Try
The wrong first question is whether AI can trade. The right first question is whether AI can explain what just happened. Read-only AI agents on CrossTrade MCP do that in seconds.
Every AI trading article opens with the same wrong question: can AI place trades? The funded futures answer is, "depends on the firm, depends on the prompt, depends on whether you trust the model on your worst day." Useful for an internal debate. Not useful for tomorrow's session.
The right first question is different. Can AI explain what just happened?
The right first question
You closed yesterday $180 down. Was the day a clean stop-out on three trades, or did a revenge entry between trades two and three turn a manageable session into a real hole? You ran a TradingView webhook strategy on Sim101. Did every alert produce a fill, or did three signals fail silently? You copy fills from your leader account to two follower accounts. Are they actually in sync right now?
These are the questions whose answers improve the next session. None of them require an agent to place orders. All of them require an agent to read history.
Read-only workflows
The CrossTrade MCP read-only scope (mcp:read) gives an agent access to:
- Accounts and connection state.
- Positions and working orders.
- Executions and the matched-trade journal.
- TradingView webhook signal history.
- CrossTrade Add-On activity log.
- Watermarks and account summary.
The OAuth scope makes mcp:read physically incapable of placing an order. Even if the prompt asks for it, the server returns insufficient_scope. The token is a safety net you don't have to think about.
Funded trader examples
Three workflows worth pinning to a daily checklist.
Pre-session brief. "On APEX1234, give me a pre-session brief. ListAccounts, GetConnections, ListPositions, ListOrders, GetAccountSummary, GetWatermarks. Compute remaining trailing drawdown room against $1,500 using the high watermark and current balance. Tell me what to watch today."
The agent returns a paragraph. You read it in 15 seconds. Sometimes it points at a stale working order or an unprotected position you would have missed.
End-of-session audit. "On APEX1234, pull today's matched trades, GetWatermarks, GetAccountSummary, GetActivityLog. Report total realized P&L, the lowest watermark today, drawdown room at the worst moment, and any rejected orders."
This is the report you would build in a spreadsheet if you had the energy. The agent builds it in seconds.
Webhook audit. "Pull TradingView webhook signals for the last 24 hours on Sim101. Group by outcome: filled, rejected, no order created, no fill. For each non-filled group, list signals and the likely cause."
If you run any TradingView-driven strategy, this is the diagnostic you should run weekly. Patterns of silent misses show up here that you would never catch by clicking through alerts in TradingView.
Prompt examples
Did I revenge trade? "Pull today's matched trades on APEX1234. Find any trade whose entry timestamp is within 90 seconds of a prior losing trade exit timestamp in the same instrument. Report them. Tell me how much of today's P&L came from those trades."
How close was I? "Tell me the lowest balance I hit today and the drawdown room I had at that point. Tell me whether any trade pushed me within $150 of the limit."
What was different about my losers? "Pull the last 30 days of matched trades. Bucket losing trades by setup if tagged, otherwise by instrument. Tell me whether any setup has negative expectancy with at least 10 trades."
Why this is safer than autonomous trading
Three reasons.
- The OAuth scope cannot place orders. There is no prompt that bypasses this.
- The conversation is descriptive, not prescriptive. The agent reports; you decide.
- The information has no expiration. A 9 a.m. journal review is just as good at 9 p.m. The agent does not require a fast trigger finger.
The compounding effect is the real win. After two weeks of reading the agent's daily brief, you start to recognize the patterns it flags. After two months, the patterns become a habit. The agent's job becomes "save me 20 minutes a day," and that is enough.
When to consider trade scope
Once read-only is reliable and you have a specific workflow that needs writes (NinjaScript compile loop, scheduled flatten, confirmed order placement), consider granting mcp:trade. Even then, default to confirmation gates in the prompt. Trade-enabled should mean "I am here."
For funded accounts, verify the firm's automation rules before any trade scope. Apex's current prohibited activities list bars automation. Topstep permits with caveats. Other firms vary. Always read the official rule page.
Links
Start read-only. Stay there longer than you think you need to. Upgrade only when a specific workflow earns the upgrade.