QuantDinger (OpenByteInc)
by OpenByteInc
QuantDinger is a self-hosted AI trading OS whose MCP server lets Cursor, Claude, and Codex read markets, backtest strategies, and place paper or live trades without ever touching your broker keys.
Install the MCP server
pip install "quantdinger-mcp==0.5.0"
About
QuantDinger is a self-hosted trading platform built around a Python Strategy API: you write indicators and strategies in Python, backtest them against historical data, run them in paper mode, then flip them live against crypto exchanges (Binance, OKX, Bybit, Gate, HTX, Bitget) or traditional brokers (IBKR, Alpaca). Everything, including market data, strategy code, audit logs, and credentials, stays on infrastructure you control, with Prometheus/Grafana monitoring built in.
The bundled quantdinger-mcp server sits in front of QuantDinger's Agent Gateway rather than talking to exchanges directly, so an agent gets scoped tools for markets, watchlists, indicators, strategy versions, backtests, and read-only broker state, but never sees your exchange API keys or admin JWT. Anything that mutates state (placing an order, cancelling a job, restoring a strategy version) requires an idempotency key and, for live trading, an explicit agent-token permission your server operator has to turn on first.
Key features
- Python Strategy API V2 for writing and versioning custom indicators and strategies
- Server-side backtesting and paper-trading before any capital goes live
- Live execution across multiple crypto exchanges and IBKR/Alpaca
- Agent Gateway that exposes trading operations as MCP tools without exposing credentials
- PostgreSQL audit logs plus Prometheus/Grafana monitoring and alerting
- Self-hosted install via a single Docker-based install script
Use cases
- Backtesting a Python strategy idea and getting an agent's read on the results before risking capital
- Letting an agent monitor a watchlist and surface signal alerts without granting it order-placing rights
- Running a fully agent-driven paper-trading loop to validate a strategy over weeks before going live
- Auditing a live strategy's version history and rolling back a bad deploy through the agent gateway
Available tools
market_and_watchlist_tools
Look up tradable symbols, pull OHLCV history, and manage universes/watchlists used for research.
indicator_tools
Compute and return charting-ready indicator series for a symbol and timeframe.
strategy_tools
Compile, save, and browse version history for Strategy API V2 code.
backtest_and_job_tools
Kick off backtests and poll or stream their progress and results.
deployment_tools
Create new strategy deployments, which start in a stopped state pending manual activation.
broker_and_alert_tools
Read-only observation of broker/execution state, plus scoped signal-alert notifications.
Frequently asked questions
Can an agent connected through quantdinger-mcp see my exchange API keys?
No. The MCP server only talks to QuantDinger's own Agent Gateway, which is scoped by an agent token. Your exchange credentials and admin JWT never pass through it.
Will the agent place real trades on its own?
Not by default. Live trading has to be explicitly enabled on the server and authorized on the agent token; out of the box, execution tools operate in paper mode and mutating actions require an idempotency key.