Generic MT5 data and execution infrastructure for Python applications. Export from the CLI or import a small, stable Python API in downstream packages.
The [Public API Contract](docs/api/public-contract.md) lists stable SDK exports (`mt5cli.STABLE_SDK_EXPORTS`), CLI commands, internal helpers, and responsibilities that remain out of scope (strategy logic, backtests, optimization).
Schema contracts live in `mt5cli.schemas` (`DataKind`, `validate_schema`, `normalize_dataframe`). Export and storage helpers are in `mt5cli.utils` (`Dataset`, `export_dataframe`) and `mt5cli.history`.
`MT5Client.order_send()` is a live execution primitive: it can place real trades on the connected account. mt5cli does not implement strategy logic, signal generation, backtesting, or optimization — downstream applications must gate live execution explicitly.
| `collect-history` | Collect rates, history-orders, and history-deals for one or more symbols into a single SQLite database (ticks opt-in via `--dataset ticks`) |
Collect several historical datasets per symbol into one SQLite database in a single MT5 session. Pick datasets with repeatable `--dataset` (default: `rates`, `history-orders`, `history-deals`; add `--dataset ticks` when tick-level history is required — tick data can grow the SQLite database quickly), choose conflict behavior with `--if-exists append|replace|fail` (default: `fail`), and optionally derive `cash_events` / `positions_reconstructed` views from `history_deals` via `--with-views`.
History orders and deals are fetched per symbol and concatenated, so the symbol filter is applied consistently across all datasets. The `cash_events` view is derived from symbol-filtered `history_deals`, so account-level cash events with empty or non-matching symbols may be excluded. The `rates` table records the requested `timeframe` so appended runs at different timeframes remain distinguishable. The `positions_reconstructed` view aggregates trade deals by `position_id`, excludes positions without closing-side entries, and uses volume-weighted open/close prices; reversal deals (`DEAL_ENTRY_INOUT`) are reported via `volume_reversal` / `reversal_count` columns.
### Incremental history SDK
For automated pipelines, use the importable incremental API instead of re-fetching fixed date ranges:
timeframes=["M1","H1"],# default: all fixed MT5 timeframes
lookback_hours=24,
create_rate_views=True,
with_views=True,
include_account_events=True,
)
finally:
client.shutdown()
# Standalone wrapper that opens and closes MT5 for you
update_history_with_config(
output="history.db",
symbols=["EURUSD"],
config=Mt5Config(login=12345),
)
```
- **`collect-history`**: explicit date-range export into SQLite.
- **`update_history`**: incremental append based on existing SQLite `MAX(time)` per symbol (and timeframe for rates); account-level deals use a separate cursor when `include_account_events=True`.
- **`rates` table**: normalized storage with `symbol` and `timeframe` columns.
- **Rate compatibility views**: mt5cli manages all `rate_*` views. Naming is `rate_<symbol>__<timeframe>` when a symbol has one timeframe, otherwise `rate_<symbol>__<granularity>_<timeframe>` (for example `rate_EURUSD__M1_1`). Stale `rate_*` views are dropped and recreated when rates change for offline downstream tools.
- **Rate view resolution**: use `resolve_rate_view_name()` / `resolve_rate_view_names()` to map symbols and granularities to existing SQLite compatibility views without creating databases. Both accept `None` (or a missing path) and return deterministic default names unless `require_existing=True`.
- **Rate view loading**: use `load_rate_data()` / `load_rate_data_from_connection()` to load a SQLite rate table or view into a `DatetimeIndex` DataFrame.
- **Multi-series rate loading**: use `build_rate_targets()` to build neutral `RateTarget(symbol, timeframe)` pairs, `resolve_rate_tables()` to map them to table/view names (pass `require_existing=True` for strict resolution), and `load_rate_series_from_sqlite()` to load them into a mapping keyed by `(symbol, integer timeframe)`. The loader requires existing managed views unless `explicit_tables` is supplied, and rejects duplicate `(symbol, timeframe)` targets.
- **Multi-account latest rates**: use `collect_latest_rates_for_accounts()` with `AccountSpec` to read the latest bars for several account groups, merged into a `(symbol, integer timeframe)` mapping. For long-running pollers, `collect_latest_rates_for_accounts_with_retries()` adds bounded exponential backoff that retries only recoverable MT5 errors and re-raises once `retry_count` is exhausted.
- **Latest closed bars**: use `collect_latest_closed_rates_for_accounts()` when downstream logic must exclude the still-forming current bar. It fetches `count + 1` bars at `start_pos=0`, drops the last row with `drop_forming_rate_bar()`, and validates each series is non-empty. `collect_latest_closed_rates_by_granularity()` returns the same data keyed by `(symbol, granularity_name)` such as `("EURUSD", "M1")`.
- **Credential resolution**: use `resolve_account_spec()` / `resolve_account_specs()` to merge explicit override values over `AccountSpec` fields and expand `${ENV_VAR}` placeholders (via `substitute_env_placeholders()`), raising `ValueError` for missing variables. This keeps secrets out of plan/config files without coupling to any strategy code. For config dicts or nested structures loaded from YAML/TOML, use `substitute_mapping_values(data, keys={"login", "password"})` to expand placeholders only for caller-specified keys — key names are never hard-coded in mt5cli.
- **Throttled history updates**: use `ThrottledHistoryUpdater` to wrap `update_history()` with a minimum `interval_seconds` between successful runs (monotonic clock). Call `should_update()` / `update(client, symbols)` from an application loop; errors propagate by default, or pass `suppress_errors=True` to swallow recoverable `Mt5*Error`, `sqlite3.Error`, `ValueError`, `OSError`, and MT5 client capability errors for history API methods without advancing the throttle (other `AttributeError` / `TypeError` values always propagate). Pass `update_backend` to inject a custom history update callable (same keyword arguments as `update_history`) instead of monkey-patching `mt5cli.sdk.update_history`.
- **Trading session helpers**: use `mt5_trading_session()` for a trading-capable client that initializes/logs in via `Mt5Config.path` and always shuts down safely. Pair with `detect_position_side()`, `calculate_margin_and_volume()`, and `determine_order_limits()` for generic position and sizing utilities. Keep read-only collection on `mt5_session()` / `MT5Client`.
- **Granularity-keyed rate loading**: `load_rate_series_by_granularity()` builds targets with `build_rate_targets()`, loads them with `load_rate_series_from_sqlite()`, and returns a mapping keyed by `(symbol | None, granularity_name)` such as `("EURUSD", "M1")` to reduce downstream boilerplate.
- **MT5 session helper**: use the `mt5_session()` context manager to attach to (or, when `Mt5Config.path` is set, launch) an MT5 terminal, log in, and yield a connected `MT5Client` that shuts down on exit.
