MARS-QUANT MCP server
Commit Nov 14, 2026Provides a quantitative research foundation for modeling and forecasting intraday Gold (XAU/USD) session dynamics using machine learning.
Also known as dela-99/MARS-QUANT
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- MIT
What is MARS-QUANT MCP?
MARS-QUANT provides a quantitative research foundation for modeling and forecasting intraday Gold (XAU/USD) session dynamics using machine learning.
Summary generated from the project's README
About MARS-QUANT MCP server
Modeling intraday Gold session dynamics through machine learning is the primary function of MARS-QUANT. This research foundation allows developers to build workflows that transform XAUUSD market data into predictive features and labels. It focuses on time-series correctness, specifically testing hypotheses like predicting London session direction based on Asian session returns and ranges.
Users interact with the system via a Python environment, utilizing specific command-line entry points like the xauusd_baseline_pipeline to execute end-to-end research. The platform supports loading H1 parquet files, engineering features through the mars.libs.features module, and training models using XGBoost or PyTorch architectures such as LSTMs and Transformers. While it includes legacy support for MetaTrader 5 bots and demo account connections, the core utility lies in its structured approach to chronological training, validation, and vectorized session-level backtesting.
What you can do with it
- 1Execute end-to-end research pipelinesRun a complete workflow from data loading to XGBoost model reporting for a specific year.
- 2Perform vectorized backtestingTest a trained model against processed feature files using a session-level backtest.
Written from the project's README. Every use case is checked against a quote from it.
Configuration
DEMO_ACCOUNT_NUMBER
Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachabledela-99/mars-quant is public and reachablegithub.com/dela-99/mars-quant ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-04github.com/dela-99/mars-quant ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubM.A.R.S. — Mathematical Algorithm Risk System
M.A.R.S. is a quantitative trading research and execution foundation focused on time-series correctness, modular strategy/risk interfaces, and reproducible experiment workflows.
This repository originated as a XAUUSD session-dynamics research project (Hypothesis A: Asian session → London session prediction) and has been reorganized into a maintainable research platform without discarding prior work.
What M.A.R.S. is (and is not)
Is:
- A research layer for data → features → labels → models → evaluation
- A trading-core foundation: strategy setups, risk policy stubs, backtest hooks
- Explicit about leakage risks and incomplete components
Is not (yet):
- An autonomous multi-strategy hedge fund
- A “fully AI trader” product UI
- A claim that legacy model metrics are production-ready
What the repo currently supports
| Capability | Status |
|---|---|
| Load / validate XAUUSD H1 parquet | Supported |
| Hyp-A Asia→London feature engineering | Supported (mars.libs.features) |
| Direction + return labels | Supported |
| Chronological train/val/test split | Supported |
| Baseline XGBoost train + OOS metrics report | Supported |
| Simple session-level backtest | Supported (vectorized) |
| PyTorch LSTM / Transformer architectures | Present (use chronological splits only) |
| Live MT5 bots | Legacy only (legacy/src/bots) |
| Hypotheses B & C | Research plan only |
| Regime engine / full risk engine | Skeleton / backlog |
Historical notebooks, models, and reports remain for research continuity.
Repository layout (short)
mars/ # Clean package (apps + libs)
data/ # raw + processed market data
models/ # trained artifacts
reports/ # metrics & plots
notebooks/ # exploratory (experimental)
docs/ # architecture, audit, backlog
legacy/ # original src/ and research artifacts
tests/ # smoke / unit tests
See docs/architecture.md for full detail.
Setup
# From repo root
python -m venv .venv
# Windows
.\.venv\Scripts\activate
# Unix
source .venv/bin/activate
pip install -r requirements.txt
Optional: copy .env.example → .env for MetaTrader 5 downloads / live bots.
DEMO_ACCOUNT_NUMBER=YOUR_ACCOUNT_NUMBER
PASSWORD=YOUR_PASSWORD
SERVER=MetaQuotes-Demo
Run the baseline XAUUSD workflow
Requires existing raw data, e.g. data/raw/xauusd_h1_2018_present.parquet
(already present in this repo for several start years).
# End-to-end: load → features → labels → time split → XGBoost → report
python -m mars.apps.research_lab.xauusd_baseline_pipeline --year 2018
# Equivalent trainer entry point
python -m mars.apps.trainer.train_xgb_baseline --year 2018
Outputs:
- Report:
reports/mars_hyp_a_xgb_baseline_*.txt - Models:
models/mars_hyp_a_xgb_classifier_*.joblib,models/mars_hyp_a_xgb_regressor_*.joblib - Features:
data/processed/mars_hyp_a_features_*.parquet
Optional download (MT5):
python -m mars.apps.research_lab.download_data --symbol XAUUSD --year 2018 --timeframe H1
Optional session backtest on a processed feature file:
python -m mars.apps.backtester.run_hyp_a_backtest \
--model models/mars_hyp_a_xgb_classifier_XXXX.joblib \
--features data/processed/mars_hyp_a_features_XXXX.parquet
What is still experimental
- All notebooks/ and legacy/ training scripts
- PyTorch metrics from the original repo (random train/test split — not trustworthy OOS)
- Paper-tuned advanced-feature XGBoost models (artifacts kept; pipeline not fully ported)
- Live trading bots (duplicated feature logic; not wired to
mars.libs) - Monte Carlo notebooks (useful diagnostics; not integrated into apps)
Read docs/audit_report.md and docs/research_backlog.md.
Research context (Hypothesis A)
- Features: Asian session return/range + indicators at Asia close
- Labels: London session direction (close ≷ open) and return
- Decision time: After Asia, before London open
Original research plan: docs/RESEARCH_PLAN.md.
Author
RIDGE DELA TORJAGBO
LinkedIn
MARS-QUANT MCP questions
What assets does this system focus on?
The system is specifically designed for XAU/USD (Gold) session-dynamics research.
How do I run the baseline research pipeline?
You can run the end-to-end workflow using the command: python -m mars.apps.research_lab.xauusd_baseline_pipeline --year [YEAR].