TradeLab-MCP

MCP server for quantitative research. The AI defines the strategy. MetaTrader 5 Strategy Tester runs the backtest.

IA (Codex / GPT / Grok / Gemini / Claude)
        │ MCP
        ▼
Trading Research MCP
        │
        ▼
terminal64.exe  →  Strategy Tester oficial  →  runs/<run_id>/

Fork of PHUICMT/mcp-mt5 (MIT). It does not reimplement the tester in Python.

Principle

MT5 is the authority for orders, fills, ticks, spread, SL/TP, margin, commissions, and P&L.

Python only organizes experiments (run_id, manifest, hashes, parser, derived statistics).

Each backtest generates a unique identifier. Nothing is retrieved by the “most recent file”.

runs/20260817_073412_a8f231/
  tester.ini
  strategy.mq5
  strategy.ex5
  report.htm
  tester.log
  trades.csv
  metrics.json
  manifest.json

Related MCP server: OpenFinClaw CLI

MCP Tools (V1)

Installation

Windows + MetaTrader 5 + Python 3.10+.

cd C:\Users\caioa\TradeLab-MCP
python -m pip install -e ".[dev]"

MCP Client:

{
  "mcpServers": {
    "tradelab": {
      "command": "tradelab-mcp",
      "env": {
        "MT5_INSTALL": "C:\\Program Files\\MetaTrader 5",
        "TRADE_LAB_ROOT": "C:\\Users\\caioa\\TradeLab-MCP\\research"
      }
    }
  }
}

Variables: MT5_INSTALL, MT5_DATA, MT5_TERMINAL_HASH, TRADE_LAB_ROOT.

Example

Compile este EA e faça um backtest de PETR4 D1,
entre 2015 e 2025, usando real ticks.

The agent calls run_backtest(strategy="experts/RSI2.mq5", symbol="PETR4", timeframe="D1", from_date="2015.01.01", to_date="2025.08.01", model=4) and receives:

{
  "run_id": "20260817_073412_a8f231",
  "status": "completed",
  "symbol": "PETR4",
  "period": "D1",
  "model": "real_ticks",
  "metrics": {
    "total_trades": 134,
    "net_profit": 18342.21,
    "profit_factor": 1.48
  }
}

get_run("20260817_073412_a8f231") returns exactly that experiment.

Where EAs live in MT5

Every automated test EA is published to:

<MQL5>\Experts\TradeLab MCP\

On this PC this resolves to:

C:\Users\caioa\AppData\Roaming\MetaQuotes\Terminal\D0E8209F77C8CF37AD8BF550E51FF075\MQL5\Experts\TradeLab MCP

In the Strategy Tester the expert appears as TradeLab MCP\RSI2. The folder is always relative to the active terminal (layout.experts_dir), not a fixed hash.

The source in git remains in experts/. The MCP copies source + .ex5 + local includes to the terminal folder.

The lab's base EA (CTrade, ArraySetAsSeries, one position) has not been standardized yet — it will come later, without rushing.

Acceptance EA

experts/RSI2.mq5

• RSI(2)

• buys if RSI < 10

• exits if RSI > 70

• one position

• no optimization

Acceptance test #1 is: the same EA / asset / dates / inputs / capital / model in manual Strategy Tester and via MCP must produce the same metrics and the same trades.

Tests

pytest

Integration tests (Windows + real MT5):

$env:TRADE_LAB_INTEGRATION = "1"
$env:TRADE_LAB_SYMBOL = "EURUSD"
pytest tests/test_integration_mt5.py -v

Priority

reproducibility > reliability > simplicity > speed > features

V1 is sequential (lock mt5.lock on the data directory). No parallel Python backtester. No .opt parser as official source.