TradeMemory Protocol

What if your trading bot could learn from every mistake — and invent better strategies by itself?

200+ trading MCP servers execute trades. None of them remember what happened.

TradeMemory is the memory layer that changes that.

Why TradeMemory?

"Why does my bot keep making the same mistakes?"

Persistent memory records every trade with full context — entry reasoning, market regime, confidence level, outcome. Pattern discovery finds what you can't see manually.

"My strategy worked for months, then suddenly stopped."

Outcome-weighted recall auto-downweights patterns from old regimes. Your bot adapts without you rewriting a single rule.

"How do I know it's not just overfitting?"

Every pattern carries Bayesian confidence + sample size. Built-in out-of-sample validation. Suspicious patterns get flagged, not blindly followed.

"I just want it to figure out what works."

Evolution Engine: feed raw price data. No indicators, no hand-written rules. It discovers, backtests, eliminates, and evolves — autonomously.

22 months of BTC data. Sharpe 3.84. 477 trades. 91% positive months. Zero human strategy input.

Quick Start

pip install tradememory-protocol

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

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

Then say to Claude:

"Record my BTCUSDT long at 71,000 — momentum breakout, high confidence."

claude mcp add tradememory -- uvx tradememory-protocol

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

git clone https://github.com/mnemox-ai/tradememory-protocol.git
cd tradememory-protocol
pip install -e .
python -m tradememory
# Server runs on http://localhost:8000

docker compose up -d

git clone https://github.com/mnemox-ai/tradememory-plugin.git
claude --plugin-dir ./tradememory-plugin

Use Cases

Crypto — Feed BTC/ETH trades from Binance. The Evolution Engine discovers timing patterns you'd never find manually. Auto-adapts when market regime shifts.

Forex + MT5 — Auto-sync every closed trade from MetaTrader 5. Persistent memory across sessions means your EA remembers that Asian session breakouts have a 10% win rate — and stops taking them.

Developers — Build memory-aware trading agents with 15 MCP tools + 30 REST endpoints. Your agent starts each session knowing its confidence level, active plans, and which strategies are working.

Architecture

Every memory is scored by five factors when recalled:

Based on ACT-R (Anderson 2007), Kelly criterion (1956), Tulving's memory taxonomy (1972), and Damasio's somatic markers (1994). Full spec: OWM_FRAMEWORK.md

MCP Tools

Core Memory (4 tools)

OWM Cognitive Memory (6 tools)

Evolution Engine (5 tools)

Star History

Contributing

See CONTRIBUTING.md for guidelines.

• Star the repo to follow progress
• Report bugs via GitHub Issues
• Submit PRs for bug fixes or new features

Documentation

License

MIT — see LICENSE.

Disclaimer: This software is for educational and research purposes only. It does not constitute financial advice. Trading involves substantial risk of loss.