What it does

QuantDinger is a full-stack, self-hosted quantitative trading platform. You write Python strategies, run server-side backtests, paper-trade, then flip to live execution across Binance, IBKR, Alpaca, MT5, and others. It bundles charting, an indicator IDE, multi-LLM research assistants, and a web UI into one Docker Compose stack backed by PostgreSQL 16 and Redis 7.

The interesting bit

The project treats AI agents as first-class citizens, not bolt-ons. It exposes an Agent Gateway (/api/agent/v1) plus a PyPI MCP server (quantdinger-mcp), so Cursor, Claude Code, or Codex can read market data, run backtests, and place trades — paper-only by default, with every call audit-logged. That safety model is the rare part: live trading requires explicit server-side unlock, and exchange keys never leave your own deployment.

Key highlights

• Dual strategy runtimes: vectorized IndicatorStrategy for research and event-driven ScriptStrategy for production, both in the same codebase
• One-command install: curl | bash pulls prebuilt GHCR images; no Node.js or local build step required
• Multi-venue execution: CCXT crypto venues, Interactive Brokers, MetaTrader 5, and Alpaca unified under one broker accounts page
• Agent-native by design: scoped tokens, append-only audit trails, and MCP compatibility out of the box
• Commercial adjacent: Apache 2.0 core with a hosted SaaS variant and AWS Marketplace AMI available

Caveats

• The README is heavy on feature lists and light on actual performance numbers or strategy examples; backtest quality claims are unverified in the source
• Default admin credentials (quantdinger / 123456) ship in the quick-start — the docs remind you to change them, but it is a foot-gun
• The “AI research” layer is described as multi-LLM, yet no specifics on which models, costs, or quantization are disclosed

Verdict

Worth a look if you want a unified, self-controlled quant environment and are skeptical of SaaS platforms holding your exchange keys. Skip it if you need battle-tested, audited strategy performance data before committing capital — this is infrastructure, not a proven alpha generator.