# Mt5

> Knowledge base for the official API, its polling model, and Windows-side production concerns. TRIGGER WHEN: building, implementing, writing, coding, creating, optimizing, or debugging MT5 or MetaTrader 5 trading systems with Python, including the aiomql and ZeroMQ bridge alternatives. DO NOT TRIGGER WHEN: the broker is Interactive Brokers (use ibkr), or the question is about strategy design rather than the terminal and its API.

- Skill: `acaprino/mt5` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add acaprino/mt5`
- Raw SKILL.md: https://api.skillmd.com/api/skills/acaprino/mt5/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: acaprino (https://skillmd.com/u/acaprino)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/acaprino/mt5

---


# MetaTrader 5 Python Algotrading

Knowledge base for building production-grade algorithmic trading systems with MetaTrader 5 Python API.

## What varies per broker

MetaTrader 5 is one piece of software; the broker behind a given login decides most of what an
integration actually has to handle. Nothing below is restated in full here, each item is
detailed in the file already named for it, because a reader chasing the symptom will already be
looking there:

- **Order execution parameters.** Fill mode, stops level, freeze level, execution mode and
  volume limits are broker- and symbol-specific and must be queried at runtime rather than
  assumed. See `order-execution.md`, and the Fill mode and Account mode rows of Key Decision
  Points below.
- **Data quality and coverage.** Tick volume, tick history depth and tick fidelity differ
  between ECN brokers and market makers for the same instrument, which makes cross-broker
  comparison invalid. See `data-feed-historical.md`.
- **Account and margin mode.** Netting, hedging and exchange margin modes are a broker setting,
  not a code choice, read at startup from `account_info().margin_mode`. See the Account mode
  row of Key Decision Points below.

None of this is optional hardening. A value correct for one broker is routinely wrong for the
next, which is the reason this plugin's scope is `multi-broker-platform` rather than
`single-broker`: nothing in this skill is measured against one broker's behaviour and then
offered as true of MetaTrader 5 in general.

## When to Use

- Connecting to MT5 terminal via the official Python API
- Building polling-based event systems (on_tick, on_new_candle, on_position)
- Executing orders with correct fill modes (FOK, IOC, Return)
- Downloading historical data (copy_rates, copy_ticks)
- Handling MT5 disconnections and terminal restarts
- Deploying MT5 bots on Windows with process monitoring
- Choosing between official API, aiomql, and ZeroMQ bridge

## Quick Start

For 80% of use cases, start with:
1. **Library**: `pip install MetaTrader5` (official) or `pip install aiomql` (async wrapper)
2. **Connection**: `mt5.initialize(path=..., login=..., server=..., password=...)`
3. **Event system**: polling loop with candle/tick/position change detection
4. **Orders**: `order_check()` before `order_send()`, always detect fill mode dynamically
5. **Risk**: server-side SL/TP on every position (non-negotiable)
6. **Resilience**: health check every 30-60s, psutil process monitoring, exponential backoff

Then harden incrementally:
- Silent errors - wrap every API call with None check + last_error()
- Fill mode rejections (10030) - dynamic filling_mode detection per symbol
- Terminal crashes - psutil watchdog + subprocess restart
- Weekend handling - datetime.weekday() sleep mode

## Reference Materials

- `api-architecture.md` - MT5 Python API architecture, 32 functions, named pipes IPC, MQL5 EA vs Python, library comparison
- `event-system-polling.md` - polling patterns, new candle detection, tick monitoring, position tracking, concurrency rules
- `order-execution.md` - order_send, fill modes (FOK/IOC/Return), hedging vs netting, retcodes, risk checks, magic numbers
- `data-feed-historical.md` - copy_rates, copy_ticks, depth, timezone caveats, caching, data quality, broker differences
- `production-resilience.md` - disconnection handling, reconnection, weekend management, Windows deployment, community resources

## Symptoms to entry points

| Symptom | Read |
|---|---|
| An API call returned `None` and nothing was logged | `api-architecture.md` |
| Order rejected with retcode 10030 | `order-execution.md` |
| Positions netted when hedging was expected | `order-execution.md` |
| A tick or candle handler never fires | `event-system-polling.md` |
| Bars disagree with the chart, or timestamps look shifted | `data-feed-historical.md` |
| The terminal is running but the API answers nothing | `production-resilience.md` |
| Everything stops on Saturday and resumes wrong on Monday | `production-resilience.md` |

## Key Decision Points

| Decision | Default | Upgrade When |
|----------|---------|-------------|
| Library | Official MetaTrader5 | Need async: aiomql. Need true streaming: ZeroMQ bridge |
| Event model | Polling (1-5s interval) | Tick-sensitive: poll 100-250ms. True events: ZeroMQ EA bridge |
| Fill mode | Detect dynamically per symbol | Never hardcode - changes between brokers/symbols |
| Account mode | Hedging (most forex brokers) | Check account_info().margin_mode at startup |
| Data caching | Parquet + Zstandard | Tick data - partition by day/month |
| Concurrency | asyncio (single thread) | Multi-account - separate processes per terminal |
| Backtesting | Python framework (Backtrader, Backtesting.py) | Need MT5 tester - MQL5 EA wrapper |
| SL/TP | Server-side always | Python trailing only as supplement, never as sole protection |