A collection of autonomous, AI-driven quantitative finance systems that implement Anthropic-aligned harnessing patterns using Claude Code as the customizable orchestration client. We demonstrates how to construct reproducible, auditable agent workflows for financial research and trading by leveraging Claude Code's extension architecture: Skills for role definition, Model Context Protocol (MCP) for tool integration, and structured state files for inter-agent communication.
Why Claude Code as the harness layer? Claude Code provides the most customizable client-side orchestration environment currently available for LLM-based agent systems.
| Project | Description |
|---|---|
| paper-to-factor-pipeline | Discovers arXiv papers, translates research into executable trading factors, backtests with survivorship-aware data, and iteratively refines until validation thresholds are met. |
| tradingagents-cc | Multi-agent trading system (re-implementation of arxiv:2412.20138) with 5 specialized teams — Analysts, Researchers (Bull/Bear debate), Trader, Risk Management, and Portfolio Manager — that analyze a ticker and submit orders to paper/Alpaca/IBKR. |
| moomoo-dashboard | Visual equity research reports powered by MooMoo OpenAPI. Input a ticker → get an interactive HTML dashboard with real-time market data, technicals, fundamentals, sector comparison, and sentiment analysis. |
[DEPRECATED] |
Superseded by moomoo-dashboard. This project is kept for reference only and is no longer actively maintained. |
Most projects share a common design pattern native to Claude Code:
┌──────────────────────────────────────────────────────────┐
│ Claude Code CLI │
│ • Orchestrates agent reasoning, tool routing, state I/O │
│ • Implements context engineering via progressive skill │
│ loading and session-aware compaction strategies │
├──────────────────────────────────────────────────────────┤
│ Skills (.claude/skills/*.md) │
│ └─ Markdown instruction files that define agent roles │
│ │
│ State Bus (session/*.md or sandbox/*.md) │
│ └─ Structured markdown + embedded JSON │
│ └─ All agents read/write this shared file │
│ │
│ MCP Servers (mcp_servers/*/server.py) │
│ └─ Tool providers registered in .claude.json │
│ └─ Market data, news, sentiment, exchange, etc. │
│ │
│ Persistence (data/*.db, data/*.json) │
│ └─ SQLite audit trail, portfolio state, cache │
└──────────────────────────────────────────────────────────┘
| Requirement | Version |
|---|---|
| Python | 3.10+ |
| Claude Code CLI | Latest |
Each project has its own requirements.txt and virtual environment. See individual project READMEs for setup instructions.
# Clone the repo
git clone https://github.com/NatBrian/auto-research-finance.git
cd auto-research-finance
# Pick a project and follow its README
cd paper-to-factor-pipeline/ # or tradingagents-cc/ moomoo-dashboard/ stock-insight/These systems are for research and educational purposes only. They are not financial advice. Past performance does not guarantee future results. Use real-money exchange adapters at your own risk.