Skip to content

GETTING_STARTED.md

Latest commit

 

History

History
214 lines (151 loc) · 5.48 KB

GETTING_STARTED.md

File metadata and controls

214 lines (151 loc) · 5.48 KB

Getting Started with TradeMemory

Choose your path:

Trader Track

For traders using Claude Desktop or Claude Code. No coding required.

1. Install (30 seconds)

pip install tradememory-protocol

Add to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "tradememory": {
      "command": "uvx",
      "args": ["tradememory-protocol"]
    }
  }
}

Restart Claude Desktop. TradeMemory is now connected.

Other platforms (Claude Code, Cursor, Windsurf) 
# Claude Code
claude mcp add tradememory -- uvx tradememory-protocol

# Cursor — add to .cursor/mcp.json
# Windsurf — add to Windsurf MCP config
# All platforms: run `tradememory config` for your exact snippet

2. Your First Trade Memory (2 minutes)

Before trading — ask what happened last time:

"I'm thinking about buying AAPL. Have I traded AAPL before? What happened?"

Claude checks your memory and returns past trades in similar conditions — what you did, why, and whether it worked.

Check your state:

"How's my trading state right now? Am I on a losing streak?"

Claude returns your confidence level, current drawdown, and a recommendation (normal / reduce size / stop trading).

After a completed trade — record it:

"Record this trade: I bought 100 shares of AAPL at $195 and sold at $205 for a $1,000 profit. Reason: earnings beat expectations and institutional buying volume was high."

One call. Five memory layers update automatically:

  • Episodic — the full event with context
  • Semantic — updates your AAPL strategy win rate belief
  • Procedural — updates average hold time and position sizing
  • Affective — updates confidence, tracks the win streak
  • Audit — SHA-256 hashed record of the decision

3. Your Pre-Flight Checklist

Based on how real users run TradeMemory in production:

Before every trade:
  1. "What happened in similar conditions?"
  2. "What's my current trading state?"
  3. "Should I take this trade?"
     → The system returns: full size / reduced size / skip

After every trade:
  4. "Record this trade with full context"

Daily:
  5. "Run my daily trading review"

Weekly:
  6. "Give me my weekly strategy breakdown"
Technical: which MCP tools power each step
Step MCP Tool / REST Endpoint
1 recall_memories
2 get_agent_state
3 check_trade_legitimacy
4 remember_trade
5 REST: /reflect/run_daily
6 REST: /reflect/run_weekly

If any pre-flight check returns a red flag — high drawdown, bad streak, low legitimacy score — pause and review before trading.

4. Tips

  • Be specific with context. "Bought AAPL because of earnings beat" is better than "Bought AAPL." The more context you give, the better recall works next time.
  • Record losses too. The system learns more from losses than wins.
  • Check memory before trading, not after. The biggest value is preventing repeat mistakes.

Developer Track

For developers integrating TradeMemory into trading bots or AI agents.

1. Install + Configure

pip install tradememory-protocol

# Start MCP server
python -m tradememory

# Or via uvx (no install needed)
uvx tradememory-protocol

MCP SSE endpoint: http://localhost:8001/sse REST API: http://localhost:8000

Docker 
git clone https://github.com/mnemox-ai/tradememory-protocol.git
cd tradememory-protocol
docker compose up -d

2. Core Pattern (3 tools)

Write — record a completed trade:

# MCP tool: remember_trade
{
  "symbol": "AAPL",
  "direction": "long",
  "entry_price": 195.0,
  "exit_price": 205.0,
  "pnl": 1000.0,
  "strategy_name": "EarningsBreakout",
  "market_context": "Post-earnings gap up, institutional volume spike, RSI 62"
}
# → Writes to all 5 memory layers automatically

Read — recall similar past trades:

# MCP tool: recall_memories
{
  "symbol": "AAPL",
  "market_context": "Pre-earnings, IV rising, support at 190"
}
# → Returns ranked memories weighted by outcome quality + context similarity

State — check agent health:

# MCP tool: get_agent_state
# → { "confidence": 0.72, "drawdown_pct": 3.2, "recommended_action": "normal" }

3. REST API Integration

# Record a trade decision
curl -X POST http://localhost:8000/trade/record_decision \
  -H "Content-Type: application/json" \
  -d '{"symbol": "AAPL", "direction": "long", "entry_price": 195.0, "strategy_name": "EarningsBreakout", "market_context": "Post-earnings gap up"}'

# Record outcome
curl -X POST http://localhost:8000/trade/record_outcome \
  -H "Content-Type: application/json" \
  -d '{"trade_id": "...", "exit_price": 205.0, "pnl": 1000.0}'

# Daily reflection
curl -X POST http://localhost:8000/reflect/run_daily

# Weekly reflection
curl -X POST http://localhost:8000/reflect/run_weekly

4. Full Reference

Back to README · Use Cases · API Reference