Claude-trading-agent

🧠 Root hub · Variables · Subfolders: self-learning/ · tradingview-mcp/

This project has for ambition to create a Claude–Warren-Buffett bot that trades automatically.

It brings together three building blocks:

1. A Claude trading-skills library (.claude/skills/) — ~25 progressive-disclosure skills covering market structure, technical analysis, chart patterns, risk management, execution, psychology, regulations and more. These are the skills the agent loads at runtime to reason about a trade.
2. A self-learning trading system (self-learning/) — a Python pipeline that combines technical-analysis strategies with Claude-powered signal validation and a feedback loop that continuously updates strategy weights based on past trade performance. It also keeps a persistent cross-session memory so Claude learns from its own trades over time. (Originally the self-learning-claude-trade repo.)
3. A TradingView MCP server (tradingview-mcp/) — a Model Context Protocol plugin that gives Claude live market data (Twelve Data) and TradingView webhook alerts, directly inside Claude Desktop.

Repository layout

Claude-trading-agent/
├── .claude/
│   └── skills/               # Runtime Claude trading skills (trading-master, claude-trade, …)
├── self-learning/
│   ├── skills/               # Module docs for the self-learning Python system
│   ├── claude_memory.py      # Persistent, self-updating memory of Claude's trades
│   └── memory/               # Where Claude stores what it learned each session
├── tradingview-mcp/          # MCP server: TradingView webhooks + Twelve Data market data
└── README.md

TradingView MCP server (plugin)

tradingview-mcp/ is a TypeScript MCP server that exposes live market data (Twelve Data) and TradingView webhook alerts to Claude via stdio. Tools: get_quote, get_indicator, get_alerts, get_history. Full details in tradingview-mcp/README.md.

Install the plugin in Claude Desktop

1. Get a (free) Twelve Data API key

Register at https://twelvedata.com/register (free tier: 500 requests/day) and copy your key.

2. Build the server

cd tradingview-mcp
npm install
cp .env.example .env        # then put your TWELVE_DATA_API_KEY inside
npm run build               # compiles TypeScript to dist/

3. Register it with Claude Desktop

Open your claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

and add the server (adapt the absolute path and key):

{
  "mcpServers": {
    "tradingview": {
      "command": "node",
      "args": ["/absolute/path/to/tradingview-mcp/dist/index.js"],
      "env": {
        "TWELVE_DATA_API_KEY": "your_key_here",
        "WEBHOOK_PORT": "3001"
      }
    }
  }
}

4. Restart Claude Desktop

Quit and reopen Claude Desktop. The tradingview tools (get_quote, get_indicator, get_alerts, get_history) now appear in the 🔌 tools menu.

5. (Optional) Wire up TradingView alerts

To feed Pine-script alerts into get_alerts, point a TradingView webhook at your running server (use ngrok http 3001 to expose it during development):

• URL: http://your-ip:3001/webhook
• Message:

{"ticker": "{{ticker}}", "price": {{close}}, "action": "BUY", "time": "{{time}}"}

Self-learning system skills

Skill	File	Description
Market Analysis	self-learning/skills/market_analysis.md	Fetch data and compute indicators
Technical Analysis	self-learning/skills/technical_analysis.md	15+ technical indicators (RSI, MACD, BB, …)
Signal Generation	self-learning/skills/signal_generation.md	Generate BUY/SELL/HOLD signals
Claude Advisory	self-learning/skills/claude_advisory.md	AI-powered signal validation
Risk Management	self-learning/skills/risk_management.md	Position sizing and risk checks
Portfolio	self-learning/skills/portfolio.md	Portfolio state, stop/TP management
Backtesting	self-learning/skills/backtesting.md	Historical strategy evaluation
Self Learning	self-learning/skills/self_learning.md	Adaptive strategy weights + persistent cross-session memory

The self-learning system also includes claude_memory.py (ClaudeMemory), a dependency-free module that lets Claude learn from each closed trade and persist those lessons in self-learning/memory/ across sessions. See self-learning/memory/README.md.

Claude trading-skills library

See .claude/skills/README.md and .claude/skills/trading-master/ for the routing entry point and the full skill index.

Disclaimer

This project is for educational purposes only. It is not financial advice. Always conduct your own research before making any investment decisions.