A Self-Hosted Quant Desk That Keeps AI Agents on a Leash

What it does

QuantDinger is a full-stack quantitative trading platform that wires multi-LLM research, Python-native strategy engines, backtesting, and live execution into a single web interface. It targets crypto, stocks, and forex through unified broker accounts—supporting venues like Binance, Interactive Brokers, MetaTrader 5, and Alpaca—while keeping exchange API keys inside your own infrastructure.

The interesting bit

Most “AI trading” projects are chatbots that dump code into a void; QuantDinger actually runs the loop. Its Agent Gateway exposes market data, backtests, and order execution to Cursor, Claude Code, or Codex via an MCP server, but defaults every agent token to paper-only trading and demands an explicit server-side unlock before live money is at risk.

Key highlights

• Dual strategy runtimes: IndicatorStrategy for vectorized dataframe research and ScriptStrategy for event-driven on_bar production trading
• Multi-venue execution via CCXT, IBKR, MT5, and Alpaca, managed through isolated multi-tenant broker sessions
• Agent-native API with scoped tokens, append-only audit logs, and a PyPI MCP package (quantdinger-mcp)
• Self-hosted Docker Compose stack with prebuilt multi-arch images, PostgreSQL 16, and Redis 7
• Closed-loop workflow: AI research → strategy code → backtest → paper/live execution → monitoring

Caveats

• The default admin password is literally 123456; the README warns you to change it immediately after first login
• The frontend ships prebuilt and the standard deploy omits the Vue source tree, so customizing the UI likely means rebuilding the image yourself

Verdict

A strong fit for developers who want to self-host their entire quant pipeline and keep exchange credentials off SaaS servers. Avoid if you are looking for a lightweight pip-installable library—this wants a full container stack and a PostgreSQL instance.