An AI-powered trading system that connects Claude to MetaTrader 5 — covering strategy definition, backtesting, walk-forward validation, demo execution, supervised live trading, and autonomous live trading with safeguards.
This system is built as a companion to Lewis Jackson's TradingView MCP connector:
- GitHub: github.com/LewisWJackson/tradingview-mcp-jackson
- YouTube: @LewisWJackson
- Community: Zero One Systems
Lewis built and open-sourced the TradingView MCP connector that allows Claude to read live chart data in real time. This repo extends that foundation by adding MetaTrader 5 connectivity, backtesting infrastructure, and a full execution pipeline.
If you haven't set up the TradingView connector first, start there:
https://github.com/LewisWJackson/tradingview-mcp-jackson
| Phase | What it builds |
|---|---|
| Strategy | Defines your rules in rules.json — any strategy, any market |
| MT5 Bridge | Python MCP server + Expert Advisor connects Claude to MetaTrader 5 |
| Backtest | Walk-forward test: 70% train, 30% unseen validation |
| Demo | Fires real signals on a demo account — real conditions, no risk |
| Supervised live | Real money — Claude signals, you approve before each trade fires |
| Autonomous live | Full auto within your risk limits — kill switch always on |
- It is not financial advice. Every strategy is the user's own.
- It does not place trades without your explicit consent (until you enable autonomous mode — which requires multiple deliberate steps).
- It is not a get-rich-quick system. It automates the mechanical execution of rules you define. The edge comes from your strategy, not from Claude.
Claude Code
│
├── tradingview-mcp-jackson/ ← Lewis Jackson's connector (TradingView)
│ rules.json ← shared strategy config (lives here)
│ .env ← shared secrets
│
└── claude-mt5-trader/ ← this repo (MT5)
mt5-bridge/
mt5_server.py ← Python MCP server (reads MT5 data)
backtest.py ← walk-forward backtester
ClaudeSignalEA.mq5 ← MT5 Expert Advisor (executes signals)
signals/ ← signal handoff between Python and EA
scripts/
notify.py ← Telegram / Gmail delivery
logs/ ← trade logs (gitignored)
PROMPT_MT5_TRADING_SYSTEM.md ← paste into Claude Code to run
The rules.json strategy file lives in tradingview-mcp-jackson so both systems share a single source of truth. The path is configurable via RULES_PATH in .env.
- Set up Lewis's TradingView connector first (link above)
- Open a fresh Claude Code session
- Paste the contents of
PROMPT_MT5_TRADING_SYSTEM.md— Claude walks you through the rest
That's it. The prompt handles Python installation, MT5 connection, strategy building, backtesting, and execution setup.
claude-mt5-trader/
├── README.md
├── PROMPT_MT5_TRADING_SYSTEM.md ← the onboarding prompt
├── .env.example ← copy to .env and fill in
├── .gitignore
├── mt5-bridge/
│ ├── mt5_server.py ← MCP server (Python)
│ ├── backtest.py ← walk-forward backtester
│ ├── ClaudeSignalEA.mq5 ← MT5 Expert Advisor (MQL5)
│ └── signals/ ← signal files (gitignored)
├── scripts/
│ └── notify.py ← Telegram / Gmail notifications
├── pine/ ← Pine Script storage (if using Branch B)
└── logs/ ← trade logs (gitignored)
| What | Who | Where |
|---|---|---|
| TradingView MCP connector | Lewis Jackson | github.com/LewisWJackson |
| TTFM strategy methodology | TTrades | youtube.com/@TTrades_edu |
| MT5 Python library | MetaQuotes | pypi.org/project/MetaTrader5 |
| AI engine | Anthropic Claude | anthropic.com |
This software is provided for educational and research purposes. Trading financial instruments carries significant risk. Past performance (including backtest results) does not guarantee future results. You are solely responsible for any trades placed using this system. Always test on a demo account before using real money.