One-stop quant-trading AI agent — research · strategy · backtest · paper trade from one prompt. Works in Claude Code, Cursor, and 20+ AI agents via MCP. 60-second install with auto Skill registration.
Research · strategy · backtest · paper trade — ship a complete quant workflow from a single natural-language prompt, inside Claude Code, Cursor, and 20+ AI agents.
First-time install walkthrough for Claude Code, Codex CLI, and ChatGPT.
How you interact
Example Prompts
Copy-paste any of these into openfinclaw deepagent research "…" (or drop them straight into your AI agent). Each one runs the full research → strategy → backtest loop.
📈 Technical analysis
Find RSI divergence signals on NVDA in the last 6 months, then backtest them.
Compare a Bollinger Bands strategy on TSLA vs AAPL over 1 year — which wins?
Screen the S&P 500 for golden-cross signals this month.
📊 Fundamentals & macro
Pull Apple's last 8 quarters of revenue, margins, and guidance. Summarize the trend.
What's driving the NVDA move this quarter — earnings, guidance, or narrative?
Compare AMD / INTC / NVDA on growth, margin, and valuation.
🎯 Strategy generation
Design a momentum strategy on US mega-cap tech. Backtest 2y. Tell me where it breaks.
Write a mean-reversion strategy on BTC and show drawdown behavior through 2022.
A-shares 沪深 300 日内轮动策略，年化目标 15%，最大回撤 < 10%。
🧪 Backtest & stress-test
Backtest a 50/200 SMA crossover on SPY from 2015. Include costs and slippage.
Stress-test my forked strategy against the 2020 and 2022 crashes.
Want a ready-made one? Run openfinclaw leaderboard to browse the community's highest-ranked strategies, then fork any of them.
install runs the interactive wizard, writes MCP configs to every detected AI agent, persists your fch_ key to ~/.openfinclaw/config.json (chmod 600 on Unix), drops a SKILL.md under ~/.claude/skills/openfinclaw/ so Claude Code / Cursor auto-trigger on keywords like quant / backtest / 量化, and finishes with a connectivity check.
Just the wizard, no SKILL.md registration, no doctor: npx @openfinclaw/cli init.
CLI quick reference
A single fch_ key drives both DeepAgent and the strategy group. Resolution order: --api-key → OPENFINCLAW_API_KEY → ~/.openfinclaw/config.json.
| Group | Commands |
|-------|----------|
| DeepAgent | deepagent +research "<query>", deepagent health, deepagent skills, deepagent threads, deepagent messages, deepagent backtests, deepagent packages, deepagent download |
| Strategy | leaderboard, strategy-info, fork, list-strategies, validate, publish, publish-verify |
| Raw | api GET <path> · api POST <path> --json '<body>' — direct Hub Gateway call, auth pre-attached |
| System | install · init · skill-install · serve · doctor · update · examples |
+verb (e.g. deepagent +research) is the human-friendly streaming path; the bare verbs and MCP-only atomic triplet research_submit / research_poll / research_finalize are for agents/scripts. Run openfinclaw --help for the full surface.
OpenFinClaw ships with a community strategy exchange. Browse what others are running, copy any strategy locally, tweak it, and publish back — think of it as a Hugging Face for quant strategies.
openfinclaw leaderboard --limit 20 # Browse top-ranked strategies
openfinclaw strategy-info <id> # See how a strategy performs
openfinclaw fork <id> # Copy to ./strategies/<slug>
# ... edit strategy.py, tweak fep.yaml ...
openfinclaw validate ./strategies/<slug> # Pre-flight FEP v2.0 check
openfinclaw publish ./my-strategy.zip # Ship to the leaderboard
openfinclaw publish-verify --submission-id <id> # Track backtest progress
Every published strategy is backtested server-side and ranked by live-market-equivalent returns — no self-reported numbers.
Supported Platforms
OpenFinClaw works with any MCP-compatible agent platform:
For other platforms (VS Code, Hermes, Windsurf, Zed, OpenClaw, Junie, Trae, …), see configs/ for ready-to-copy templates. The shape is the same — only the host key (servers vs mcpServers vs context_servers) and config path differ.
Tool Groups & Context Optimization
Load only what you need to save tokens: serve --tools=deepagent (~1,400 tk) or serve --tools=strategy (~1,000 tk), or omit --tools for both.
| Variable | Description |
|----------|-------------|
| OPENFINCLAW_API_KEY | Unified fch_ key. Drives both strategy (Hub) and deepagent (Hub Gateway). Falls back to ~/.openfinclaw/config.json if unset. Get a key at hub.openfinclaw.ai. |
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
