7 Free GitHub Repos That Turn Claude Into a $500K Quant Trader

This episode on Miles Deutscher Finance starts with a simple experiment from the past few weeks, scanning the finance corner of GitHub for repos that actually move the needle on AI trading. The host filtered four headings, finance, quantitative finance, stock market and Bitcoin trading bots, and distilled seven standouts. The promise is bold, turning Claude into a chart reader, a research floor and finally a trade-executing agent, but the video earns it by showing the setup step by step. Framing GitHub as YouTube for software is not just a metaphor, anyone can publish code, likes accumulate as stars, and the most starred projects rise to the top.

The spine of the video is MCP. Model Context Protocol is a bridge that plugs a language model like Claude into an outside dataset or application. Once that bridge is in place, the model stops merely scraping the web and can pull a price, draw on a chart and even place an order through an order engine. Deutscher says you can either paste a single setup prompt into Claude Code and let it wire all seven repos for you, or visit each GitHub page and ask the model to install them one by one for more control. You do not need to code, because the agent handles installation and you call the accumulated developer knowledge straight from the chat window.

First stop is OpenBB. Also branded as OpenBB Platform, this open data layer aggregates dozens of providers for stocks, crypto, options, macro and economic data in one interface. The video positions it as the free counterpart to a Bloomberg Terminal that costs about 25,000 dollars a year. The repo now sits above 72,000 stars with 268 contributors, making it one of the busiest finance projects on GitHub. The demo prompt asks Claude, via OpenBB, for Bitcoin price action over the past 30 days, the US 10-year yield via the Federal Reserve provider and the dollar index via Yahoo Finance, then to summarize the macro picture for crypto this week. The answer leans on real data, rates are a headwind, the dollar is neutral and Bitcoin momentum is weaker than the 30-day print suggests. Compared with deeper fundamental connectors such as FMP or Quiver, OpenBB wins on breadth across equities, crypto, options and ETF holdings, local execution and open source flexibility, while it is thinner on niche fundamentals like insider trades or full earnings call disclosures.

Data Plumbing: Why Bloomberg Can Be Free

The second MCP bridge targets TradingView. This repo connects Claude to the TradingView desktop app and lets the model draw on your behalf. The request is concrete, open the Bitcoin chart and mark three key support and resistance areas derived from repeated price reactions and notable swing points. In the recording the model draws clear pivot, resistance and support bands, then adds VWAP so the point of control, where the most volume has printed, aligns closely with support. What Deutscher stresses is the daily workflow, scanning a watchlist each morning to flag when price nears a zone, spinning up custom indicators or merging two you already use, and running TradingView backtests via prompt. Run on a faster model in the background, it behaves like a trading partner and even a mentor, you can ask what a line means, whether a setup validates a long, or which indicator would help confirm a trade. The seven prompts he shared in the previous video complete the practical playbook.

Third up is Fincept Terminal. A native C++20 desktop terminal for financial research, it feels like a full research desk for equity research, portfolio management and news. Installation is as short as downloading the DMG for Mac Apple Silicon and dropping it into Applications. Inside you find a screener, coverage for ETFs, indices and forex, a watchlist, portfolio entry and an equity curve alongside news. The standout is the chat pane, ask it to summarize today's major financial news and the key events to watch over the next three weeks as a stock trader and it replies in one interface. The quant lab, unlocked on the paid tier, lets you discover and optimize curated algo strategies, with code you can inspect. Fincept's distinctive edge is geopolitics and global intelligence, maritime tracking, geopolitical analysis and relationship mapping, plus analytics agents that can call a real DCF engine, portfolio optimization and derivatives pricing through a QuantLib suite and a visual node editor that talks to MCP tools. Launched in mid 2024, the project now holds more than 30,000 stars and has exploded in the past year. Deutscher notes he is building a modified version himself, yet still recommends trying the stock build because having everything in one place is the core value.

Terminals and Dashboards: Research on One Desk

The fourth repo is the one the English market often overlooks because it is written in Chinese, ZhuLinsen's daily_stock_analysis. Pitched as a multi-market stock analysis system, it covers A-shares, Hong Kong, US, Japan, Korea and Taiwan. Its superpower is automated notifications, for every ticker on your watchlist or portfolio it pulls quotes, technicals, news, announcements and fundamentals, runs them through a language model and produces a decision dashboard with a buy, watch or sell verdict, a score out of 100 and entry and exit levels. Reports can be pushed to Telegram, Discord, Slack, email or enterprise chat apps. The default dashboard aggregates everything, with extras such as extracting tickers from screenshots, CSV import and backtesting of historical accuracy. It runs on a zero server cost schedule via GitHub Actions and carries more than 62,000 stars with well over a hundred contributors. The Telegram card shown in the video flags, in a single view, what is marking a sell, what is a potential buy and which catalyst stands out. As someone who can vibe code a dashboard himself, Deutscher argues the ready made path simply saves time and benefits from a wide contributor base that irons out glitches faster than a solo build.

Fifth is TradingAgents, a multi-agent LLM financial trading framework. It aims to mirror the dynamics of a real trading firm inside the model. Specialist agents are split by role, a fundamentals analyst parses financials, a sentiment analyst reads news tone, a technical analyst interprets momentum, while a researcher team debates outlook and risks. A trader agent fuses those notes to decide timing and sizing, and a risk management plus portfolio manager layer evaluates volatility and liquidity to approve or reject the proposal. An approved proposal is sent to an exchange for execution, simulated in the demo but connectable to a live venue in the final step. The framework can route to different models, using a strong planner for reasoning and lighter agents for grunt work. The Bitcoin demo shows the market analyst delivering the technical read, with sentiment and fundamentals chiming in, culminating in a final synthesis, verdict hold with medium conviction. The rationale is clear, the 50-day simple moving average sits above the rising 200-day with no death cross, yet MACD histogram and RSI are falling, a six week rally is unwinding and two straight weeks of single day ETF outflows around 591 million dollars add pressure. A daily close below 72,000 dollars would flip the call to sell or underweight. The video notes that running this loop on a schedule across a full book is where it shines, and for single stocks with richer qualitative inputs such as cash flows, the agent debate adds even more value.

Sixth is the AI Hedge Fund by virattt, one of the most starred quant repos at more than 63,000 stars. The idea is to turn 14 famous investors into pre-coded agents, Warren Buffett, Charlie Munger, Bill Ackman, Cathie Wood, Stanley Druckenmiller and others, each with that investor's philosophy encoded as instructions. For any ticker each agent returns a signal, a confidence score and a reasoning paragraph. The repo itself does not ship data, so Deutscher feeds it with an external dataset such as FMP or a similar provider and shares discounted access links. The demo build is ambitious, five names, Nvidia, Coinbase, MicroStrategy, Palantir and Tesla, with 100,000 dollars in notional capital and every investor strategy enabled. The model renders JSON into a table and lists the three sharpest divergences where one holder is strongly bullish while another is deeply bearish on the same name. Palantir is striking, Druckenmiller is bull at 80 on revenue and margin expansion while Graham is bear at 85 arguing there is no margin of safety at this price. On Nvidia, Lynch is bull at 88, Graham bear at 72 saying future cash flows do not justify the balance sheet at this valuation. The weighted final call comes out as buy for Nvidia and sell for the other four. Going deeper than 10 years of history towards 30 years requires a paid data tier. Deutscher also frames it as a thought experiment, describe your own investing style to Claude and see which investor persona matches you best.

From Research Floor to Trading Desk

The seventh and final piece is the Alpaca MCP server. Alpaca is a brokerage that supports stocks, options, crypto and real time market data and offers both paper and live accounts. The official MCP server translates plain English commands into its trading API. The demo starts from a 100,000 dollar paper account that can be toggled to live with one switch. After grabbing an API key and plugging it into Claude, a single sentence suffices, buy 1,000 dollars worth of Nvidia via Alpaca MCP, the model confirms account access and submits the order, queuing it for the next open if the market is closed. The compelling part is conditional instructions, for example, if Nvidia drops 10 percent below my fill, sell 30 percent of the position, all writable in natural language. For that to run around the clock the brain, Claude, needs a way to call the API 24/7, which implies a virtual server. Deutscher notes you can wire other brokers such as Interactive Brokers in a similar way but describes Alpaca as the easiest on ramp and urges paper trading first before risking real capital. The signal forged on the research floor now becomes a one sentence execution on the desk.

Together the seven pieces form a layered stack. OpenBB collects the data, TradingView interprets the chart, Fincept unifies research on one desk, the China-born daily system pushes a decision dashboard to your pocket each morning, TradingAgents debates like a research floor to produce a single hold signal, AI Hedge Fund looks at the same name through 14 masters' eyes and surfaces disagreements, and Alpaca presses the button. Deutscher's suggestion to install them with one prompt encourages thinking of the chain as a system rather than parts. GitHub works here like a free app store, each repo is an app and stars are user ratings. If you dislike one you remove it and try another while the model orchestrates all the apps on your behalf.

On the practical side the checklist is clear. You need Claude Code or a comparable agent shell, Python 3.10 plus, uv with uvx, the TradingView desktop app and Alpaca API keys. Most of the open source repos are free and run locally, even with open models. The only paid gate is deep history for the AI Hedge Fund, 10 years is accessible on a lighter plan while 30 years needs a subscription. In the school community Deutscher shares a single prompt that wires everything and includes each repo's connection guide. The alternative path is to visit each GitHub page and ask Claude to install them one by one, more hands-on but also more control. The video repeatedly flags API key and quota management for data heavy repos.

Setup, Cost and Practical Roadmap

No demo in the video offers live profitability proof, which is a crucial boundary. The TradingAgents example stays on a simulated exchange, the AI Hedge Fund output is a synthetic reading of historical philosophies, and while Fincept's quant lab shows curated strategies, its backtests do not spell out transaction costs, slippage or liquidity assumptions. Numbers such as the 591 million dollar ETF outflow and the 72,000 dollar threshold are tied to a moment in market time and age quickly. Hallucination risk remains, especially for verifiable items like financials or insider trades, and needs an independent cross check. That is why the repeated advice to start with paper trading and the warning about needing a virtual server are not throwaway lines, they mark the distance between automation's appeal and real market friction.

The ecosystem signals are strong nonetheless. OpenBB beyond 72,000 stars, TradingAgents near 101,000 stars with more than 19,000 forks, AI Hedge Fund at 63,000 stars under MIT, Fincept above 30,000 stars with heavy C++ contributions, daily_stock_analysis at 62,000 stars under MIT, each has been polished by a broad contributor pool. Stars are not a guarantee of alpha, but they do signal faster debugging and richer documentation than a solo effort. Permissive licenses make forking and tailoring to your own needs straightforward, and Deutscher's own modified Fincept variant is a case in point. Paired with external data providers, these repos let a retail investor carry what once required a payroll of hundreds of thousands of dollars for a research department in their pocket.

Deutscher closes with excitement about where the tech will be in six months, a year and two years, predicting that eventually everyone will trade with AI. For early followers on the channel that is a window of advantage. The takeaway is concrete, join the community to access the repos and summaries in one place, test each in a paper account first, and then go deep on the favorites with dedicated dive videos. The seven repos do not promise magic returns on their own, but together they shorten the road from data to decision, automate repetitive work and keep the investor from trading without a mentor. Building your own edge on top of that skeleton is now as close as a GitHub folder.