Open-Source AI Trading and Finance Tools
AI assistants are starting to connect to brokers, market data and backtesting engines. The tools here range from an official broker MCP server to self-hosted trading platforms and a memory layer for trading agents.
This is a list of software, not advice. Most of these projects say so themselves: backtests do not predict live results, and an assistant that can place orders can lose real money.
AgentIndex · Updated 2026-10-14
Not financial advice. Backtest results do not guarantee future returns, and trading carries a risk of loss. Keep paper trading on until you have added limits and human confirmation.
Quick picks
- alpaca-mcp-server — official MCP server for Alpaca's trading API covering stocks, options and crypto; paper trading is the default.
- tradingview-mcp — market screening and a multi-agent technical analysis debate; its README describes research and analysis, not order placement.
- QuantDinger — self-hosted trading platform with strategies, backtests, paper and live execution.
- tradememory-protocol — memory layer for trading agents that records trades with context and recalls patterns.
- FinnewsHunter — multi-agent platform that turns financial news into research.
Side-by-side comparison
| Tool | What it is | Stars | License | Last push | Install |
|---|---|---|---|---|---|
| alpaca-mcp-server | Official MIT MCP server for Alpaca's trading API: quotes, options, crypto, positions and orders, with paper trading on by default | 1.0k | MIT | 2026-10-12 | uvx alpaca-mcp-server |
| tradingview-mcp | MCP server that screens markets and runs a three-agent technical analysis debate | 5.0k | MIT | 2026-10-13 | git clone https://github.com/atilaahmettaner/tradingview-mcp.git |
| QuantDinger | Self-hosted AI trading OS: strategies, backtests, paper and live execution | 12.7k | Apache-2.0 | 2026-10-09 | curl -fsSL https://raw.githubusercontent.com/OpenByteInc/QuantDinger/main/ins… |
| tradememory-protocol | MIT MCP server that gives trading agents persistent memory of trades, with outcome-weighted recall and an LLM-driven strategy evolution engine | 1.4k | MIT | 2026-10-09 | pip install tradememory-protocol |
| FinnewsHunter | Multi-agent platform turning financial news into research | 1.5k | Apache-2.0 | 2026-10-01 | pip install -e . |
Tools in this list with an in-depth AgentIndex review: 5/5
How to choose
- Start in paper trading. Prefer tools where paper mode is the default and live trading needs an explicit switch.
- Separate analysis tools from execution tools. A market screener cannot lose you money; an order-placing server can.
- Treat published backtest numbers as the authors' own results, not as evidence of future performance.
- Check data sources and rate limits, and whether your broker's terms allow automated trading.
- Add position limits and human approval before an agent can place real orders.
FAQ
Can an AI agent trade for me safely?
Not without safeguards. Start with paper trading, set position and order limits, and require your confirmation before live orders. Models can misread a prompt and trade the wrong symbol or size.
Are backtest results from these projects reliable?
Treat them as the authors' own results. Backtests can be overfitted and do not guarantee live performance. Re-run them on your own data and add out-of-sample checks.
What is the difference between an analysis tool and an execution tool?
Analysis tools read data and produce signals or reports. Execution tools can place, change or cancel orders with real money. Give your assistant analysis tools first and add execution only with limits.
How we compiled this list
Tools are chosen by hand from the projects in our index, favouring those we have reviewed in depth. Stars, licence, last-push date and install command come from each repository and are refreshed with the site data. We read each project's README and docs; we have not run every tool ourselves, so test before you depend on one.