A developer going by lucemia has released a Claude Code plugin that ports Tauric Research's TradingAgents framework into a zero-API-cost experience. The project, posted to Hacker News on May 2nd, lets users run a seven-agent stock analysis pipeline using nothing but their existing Claude subscription and yfinance for market data—no OpenAI keys, no Anthropic API bills, no LangGraph overhead.

How It Works

After installing via /plugin marketplace add lucemia/trading-agents-plugin and syncing Python dependencies with uv sync, users trigger analysis with a simple command: /trading-analysis NVDA. The plugin then spins up seven specialized subagents across five distinct phases. Phase 1 runs four analysts in parallel—Technical, News, Fundamentals, and Macro—each pulling specific data points before Phase 2 forces Bull and Bear agents into direct confrontation over the evidence.

The Analyst Breakdown

Each agent covers granular territory: Technical checks EMA10/SMA50/SMA200 trends, RSI14, MACD, Bollinger Bands, ATR, and key support/resistance zones. Fundamentals dives into P/E, forward P/E, PEG, P/B, revenue growth, margins, ROE, FCF, D/E ratio, current ratio, beta, short ratio, and quarterly income statements. The Risk Analyst gets the unenviable job of stress-testing both bull and bear cases for crowding risk, beta drawdown scenarios, and tail risks neither side addressed.

Real Output From NVDA

The sample output for NVIDIA shows the system landed on a BUY signal with an Overweight rating, suggesting entry at $197–$199 for tranche 1 (or $188–190 if macro conditions deteriorate), a stop-loss at $183 below SMA200, and position sizing of 2–3% initially with 2% reserved for a second tranche. The bull case centered on a PEG of 0.18, $58.1B in FCF, and $51B net cash—arguing NVDA's AI dominance is structurally mispriced. The bear countered with a beta of 2.33 in risk-off macro and crowded longs that could unwind 20%+ against a 30.7x P/B backdrop.

Why This Matters

TradingAgents originally required juggling multiple LLM API providers and complex LangGraph orchestration. Lucemia's plugin strips that friction entirely—leveraging Claude Code's native Agent tool instead of external infrastructure. For developers who want multi-agent reasoning on financial data without the operational complexity (or billing surprises), this is a clean workaround using tools they likely already pay for.

Key Takeaways

  • Seven specialized agents run across five phases: Technical/News/Fundamentals/Macro → Bull/Bear confrontation → Risk stress-test → Research synthesis → Portfolio Manager verdict
  • All market data pulls from yfinance—no paid data subscriptions required
  • No OpenAI, Anthropic, or other LLM API keys needed—runs on existing Claude Code subscription
  • Manual install available via git clone for users who prefer not to use the plugin marketplace

The Bottom Line

This is exactly the kind of creative repurposing that makes Claude Code actually useful beyond writing boilerplate. Multi-agent stock analysis without the API tab? That's a flex—and lucemia pulled it off cleanly.