How I Built an MT5 Trading Bot with Python and Smart Money Concept

#python #trading #machinelearning #tutorial

Automated trading has always fascinated me. After months of building and iterating, I want to share the architecture behind my MT5 trading bot that combines Python, Smart Money Concept (SMC), and machine learning.

Why Python over MQL5?

While MQL5 runs natively inside MetaTrader 5, Python offers a much richer ecosystem:
Machine learning libraries (scikit-learn, PyTorch) for signal scoring
Easy integration with Telegram, databases, and web dashboards
Better debugging with modern IDEs
The official MetaTrader5 Python package lets you connect directly to the terminal: python _ import MetaTrader5 as mt5 if not mt5.initialize(): print("Connection failed:", mt5.last_error()) quit() # Get latest candles rates = mt5.copy_rates_from_pos('XAUUSD', mt5.TIMEFRAME_M15, 0, 100)_

The Core Architecture
My bot has four main components:

Data layer — connects to MT5, fetches real-time prices and historical candles
Signal logic — analyzes market using Smart Money Concept (Order Blocks, FVG, BOS)
Risk management — calculates position size based on account risk percentage
Execution — places orders with proper SL/TP via the API

Smart Money Concept for Entries

Instead of traditional indicators, I use SMC to identify institutional footprints:

Order Blocks — zones where big players placed orders before strong moves
Fair Value Gaps — price imbalances that often get filled
Liquidity Sweeps — stop hunts before real reversals.

The key insight: a valid Order Block requires the impulse candle after it to be significantly larger than average (1.5-1.8x), not just slightly bigger.

Position Sizing Done Right

A common mistake is hardcoding lot calculation for one symbol. The correct approach uses tick_value and tick_size from MT5, which works universally across gold, forex pairs, and indices:

pythoninfo = mt5.symbol_info(symbol)
risk_usd = balance * risk_percent
n_ticks = sl_distance / info.trade_tick_size
lot = risk_usd / (n_ticks * info.trade_tick_value)

Adding a Machine Learning Layer

The most interesting part: I log every trade's context (ATR, RSI, session, entry position in the Order Block) along with its outcome. A simple logistic regression model then learns which setups have higher win probability, filtering out low-quality signals.

With 200+ trades of data, the model starts recognizing patterns like "this setup wins more during London session when RSI is below 40."

Lessons Learned

Rigid entry conditions kill opportunities — I initially required BOS + sweep simultaneously, which made the bot miss most valid setups. Relaxing to weighted signals improved results significantly.
Feature sync matters — ML features must stay identical between the bot's inference and the training pipeline, or you get silent failures.
Backtest properly — walk-forward testing on out-of-sample data is essential to avoid overfitting.

Full Documentation

I've written detailed guides (in Vietnamese) covering each topic in depth:

Complete MT5 Python Bot Guide
Smart Money Concept Explained
Risk Management Fundamentals
AI & Machine Learning for Trading

Disclaimer: Trading forex and CFDs carries high risk. This content is educational, not financial advice.