Master Guide: Connecting AI Harnesses, Claude, Cursor & Custom Agents

TM Pro is built from the ground up to work not only for human clicks, but also as a governed execution harness for AI coding and trading agents (Claude Desktop, Claude Code, Cursor, Antigravity, local Ollama, and custom Python / Node / Webhook scripts).

Most importantly, no AI harness can bypass your safety rules:

• Every AI request goes through the exact same Order Planner and Risk Guard (Max Risk %, Max Daily Loss %, Max Lots, Stop Loss Required) as the manual Order Ticket.
• When "Ask me before trading" is enabled on an agent token (the default), any trade proposed by an AI harness pauses and surfaces an Approve / Reject card at the top of your TM Pro window—so you always have the final click.

---

Part 1: Creating a Scoped Agent Token (Agents Tab)

Before connecting any external AI harness (Claude, Cursor, or a script), issue it a dedicated token so you can control its permissions and revoke it at any time:

1. Open the TM Pro desktop app and click the Agents tab (Bot icon).
2. In the New agent token card:
   • Name: Give the harness a clear name (for example, Claude Code, Cursor IDE, or My Python Bot).
   • Scopes:
     • read: Allows the AI harness to inspect your connected terminals, live prices, account balance/equity, open positions, and AI Quant Regime signals (Hurst, ATR, Quarter-Kelly, VaR). Cannot place or modify trades.
     • trade: Allows the AI harness to preview orders, place orders, modify stops, and close positions (subject to your Risk Guard and Approval setting).
     • admin: Full administrative scope.
   • Ask me before trading Checkbox:
     • Keep this CHECKED (ON) (Recommended): Whenever the AI harness calls place_order, modify_position, or close_position, TM Pro pauses the command and shows an Approval Card at the top of your window for you to click Approve or Reject.
     • Uncheck (OFF): Allows the AI harness to execute trades autonomously without waiting for a click (still strictly bounded by your Risk Guard limits in Settings!).
3. Click Create.
4. Copy the yellow token banner immediately (it starts with tmpro_...). For security, TM Pro stores only the SHA-256 hash on disk and never displays the raw token again.

---

Part 2: Connecting Claude Code, Claude Desktop & Cursor via MCP

TM Pro ships with a native Model Context Protocol (MCP) stdio server located at app/dist/mcp/tmpro-mcp.js. Connecting it gives your AI harness direct, typed access to your live MetaTrader terminals.

Option A — Connect to Claude Code (1 Command in Terminal)

Run this single command in PowerShell or Command Prompt (replace tmpro_YOUR_TOKEN with the token you copied from the Agents tab):

claude mcp add tmpro -e TMPRO_TOKEN=tmpro_YOUR_TOKEN -- node "D:/itrad/repos/itradeaims-tmpro/app/dist/mcp/tmpro-mcp.js"

Verify it is connected by typing /mcp inside Claude Code.

Option B — Connect to Claude Desktop, Cursor, or Antigravity (mcp.json)

Add the tmpro block to your MCP configuration file:

• Claude Desktop: %APPDATA%\Claude\claude_desktop_config.json
• Cursor: .cursor/mcp.json (or Cursor Settings -> MCP -> Add new global MCP server)

{
  "mcpServers": {
    "tmpro": {
      "command": "node",
      "args": [
        "D:/itrad/repos/itradeaims-tmpro/app/dist/mcp/tmpro-mcp.js"
      ],
      "env": {
        "TMPRO_TOKEN": "tmpro_YOUR_TOKEN",
        "TMPRO_API_URL": "http://127.0.0.1:17845"
      }
    }
  }
}

All 6 MCP Tools Available to Your AI Harness

Once connected, your AI harness can invoke these six tools automatically when you chat with it:

MCP Tool Name	Required Scope	What Your AI Harness Can Do With It
get_state	read	Reads all connected MT4/MT5 sessions, live Bid/Ask, balance/equity, open positions, risk limits, and the on-device AI Quant Regime signal (Hurst, ATR, Quarter-Kelly, VaR).
preview_order	read	Dry-runs an order (market, limit, or stop) through the lot-size calculator and Risk Guard—returning exact lots, SL / TP prices, dollar risk, and any risk errors without placing a trade.
place_order	trade	Submits a buy or sell order sized by risk_percent, risk_money, or lots, with optional slPips, tpPips, breakEven, trailing, and partialClose rules.
modify_position	trade	Updates the Stop Loss (sl) and/or Take Profit (tp) price on an open position ticket.
close_position	trade	Closes 100% of an open position, or a partial volume (e.g., 50%).
set_protection	trade	Updates the autonomous EA protection rules (breakEven, trailing, partialClose) on a connected chart.

Example Prompts to Give Your Connected AI Harness

Try typing any of these prompts into Claude Code, Claude Desktop, or Cursor once tmpro MCP is connected:

• "Inspect my connected MT5 terminal via TM Pro. What is the current EURUSD Hurst regime, Yang-Zhang volatility, and Quarter-Kelly recommendation?"
• "Preview a BUY market order on EURUSD using the Quarter-Kelly risk percentage with a 20-pip stop loss and 40-pip take profit. Tell me the exact lot size and dollar risk."
• "Place that EURUSD BUY order with a 15-pip break-even trigger and 20-pip trailing stop so I can approve it in TM Pro."
• "Check all my open positions across my connected terminals—are any positions missing a stop loss? If any position is up more than 15 pips, move its stop loss to break-even."

---

Part 3: Connecting the Built-In AI Assistant (AI Tab)

In addition to external MCP harnesses, TM Pro has a built-in Account-Aware AI Assistant right inside the AI tab. Whenever you ask it a question, TM Pro automatically injects a live JSON summary of your connected terminals, open positions, risk limits, and Quant Regime signals into the system context.

To enable the chat assistant, open the Settings tab → AI assistant card and configure any OpenAI-compatible provider:

Option 1 — Free Local AI with Ollama (100% Private on Your PC)

1. Install Ollama and pull a model: ollama run llama3.2 (or qwen2.5:14b).
2. In TM Pro Settings → AI assistant:
   • Check: Enable AI assistant
   • Base URL: http://127.0.0.1:11434/v1
   • Model: llama3.2
   • API Key: ollama (any non-empty string works for local Ollama)
3. Click Save AI settings.

Option 2 — Cloud Frontier Models (OpenAI / Groq / OpenRouter)

• OpenAI:
  • Base URL: https://api.openai.com/v1
  • Model: gpt-4o-mini or gpt-4o
  • API Key: Your sk-... key (stored locally on your PC, never committed or shared).

---

Part 4: Connecting Custom Python, Node.js, or Webhook Harnesses via REST (127.0.0.1:17845)

Building your own systematic Python strategy, local TradingView webhook relay, or custom execution script? TM Pro runs a fast, Bearer-authenticated HTTP REST server bound strictly to 127.0.0.1:17845 (local machine only).

REST Endpoint Reference

Method & Path	Scope	Description
GET /v1/health	None	Returns {"ok": true, "version": "2.0"} to verify TM Pro is running.
GET /v1/state	read	Returns full JSON snapshot of all sessions, accounts, positions, and signal (Quant Regime telemetry).
POST /v1/orders/preview	read	Validates and sizes an order request, returning the calculated plan and errors.
POST /v1/orders	trade	Submits an OrderIntent (side, orderType, sizingMode, sizingValue, slPips, tpPips, protection).
POST /v1/commands	trade	Submits a general HubRequest (place_order, modify_position, close_position, close_all, set_protection).

Ready-to-Run Python Harness Example

import requests

TMPRO_URL = "http://127.0.0.1:17845"
HEADERS = {"Authorization": "Bearer tmpro_YOUR_TOKEN"}

# 1. Fetch live multi-terminal state and on-device Quant Regime telemetry
state = requests.get(f"{TMPRO_URL}/v1/state", headers=HEADERS).json()
session = state["sessions"][0]
sid = session["sessionId"]
signal = session.get("signal")

print(f"Connected: {session['hello']['chart']['symbol']} | Equity: ${session['account']['equity']}")
if signal:
    print(f"Quant Regime: {signal['regime']} | Hurst: {signal['metrics']['hurst']} | 1/4-Kelly: {signal['recommendation']['kellyRiskPercent']}%")

# 2. Submit a risk-checked trade using the Quarter-Kelly recommendation
order_payload = {
    "sessionId": sid,
    "side": "buy",
    "orderType": "market",
    "sizingMode": "risk_percent",
    "sizingValue": signal["recommendation"]["kellyRiskPercent"] if signal else 0.5,
    "slPips": signal["recommendation"]["suggestedSlPips"] if signal else 20,
    "tpPips": signal["recommendation"]["suggestedTpPips"] if signal else 40,
}
resp = requests.post(f"{TMPRO_URL}/v1/orders", json=order_payload, headers=HEADERS).json()
print("Order Outcome:", resp)

---

Part 5: How the Human-in-the-Loop Approval Bar Works

Whenever an AI harness (Claude, Cursor, or Python) submits a trade using a token that has Ask me before trading enabled:

1. The AI call receives {"status": "pending_approval", "approvalId": "..."}.
2. An amber Agent approval needed banner immediately appears at the very top of your TM Pro desktop window showing:
   • Which agent requested the trade (e.g., Claude Code),
   • A plain-English summary (e.g., BUY 0.21 EURUSD (market)), and
   • The calculated dollar risk (e.g., $42.50 risk).
3. Click Approve to dispatch the trade immediately to MetaTrader, or Reject to cancel it. If you are away from your desk, the request automatically expires after 5 minutes so stale orders never fire late.

Every trade—whether clicked by you, allocated by the Fund Manager, or submitted by an AI harness—is recorded in the Activity tab with its exact broker ticket and fill price.