TradeMemory Protocol MCP Server: Setup for Claude, Cursor and Other Apps

• Records a trade in memory

  Saves a trade with its reasoning and outcome into several memory layers at once, and updates running stats such as confidence, streaks and typical hold time.

  Try: “Record my AAPL long at $195: earnings beat, high confidence.”

  Write · remember_trade

• Finds past trades that matter now

  Ranks stored trades by outcome, how similar the conditions are, recency and your current mood. With losses first it shows losing trades in similar conditions before you act.

  Try: “Before I buy, show me the losses I took in similar conditions.”

  Write · recall_memories

• Shows how each strategy is doing

  Returns totals per strategy and symbol so you can see which strategies work and which do not.

  Try: “How is my VolBreakout strategy performing on XAUUSD?”

  Read · get_strategy_performance

• Shows one trade in full

  Returns the full context and reflection recorded for a single trade ID.

  Try: “Show the reflection for trade MT5-7047640363.”

  Read · get_trade_reflection

• Describes your trading habits

  Reports hold times, how you sell winners versus losers, position size spread and a Kelly comparison, from stored trades.

  Try: “What do my trading habits look like for XAUUSD?”

  Read · get_behavioral_analysis

• Shows confidence and drawdown

  Returns current confidence, risk appetite, drawdown, win and loss streaks, and a suggested action based on drawdown.

  Try: “What state am I in right now?”

  Read · get_agent_state

• Saves a conditional plan

  Stores a rule-based plan that waits for a trigger, such as a market regime or drawdown level, until it fires, expires or you cancel it.

  Try: “Plan: if the market turns ranging, reduce size by half.”

  Write · create_trading_plan

• Checks which plans fire now

  Compares your waiting plans with the market context you give, expires old ones and returns the matches.

  Try: “Do any of my plans trigger in a volatile regime?”

  Write · check_active_plans

• Gives a go, reduced or skip rating

  Scores whether you have enough data and confidence to trade, from sample size, memory quality, regime experience, streaks and drawdown. It advises only and places no orders.

  Try: “Do I have the track record to trade VolBreakout at full size?”

  Write · check_trade_legitimacy

• Scores the quality of a decision

  Rates the decision process, not the outcome, on regime match, sizing against Kelly, similarity to past trades, risk state and historical pattern.

  Try: “Score my planned long on XAUUSD before I enter.”

  Write · compute_dqs

• Tests a strategy for overfitting

  Runs four statistical tests on a trade log or returns file you supply. The tool says it is for education and research and is not financial advice.

  Try: “Validate the strategy in this trade log CSV.”

  Read · validate_strategy

• Downloads price history from Binance

  Fetches OHLCV bars for a pair such as BTCUSDT and timeframe from Binance, to feed pattern discovery and backtests.

  Try: “Fetch 90 days of 1h BTCUSDT bars.”

  Write · evolution_fetch_market_data

• Asks Claude to propose trading patterns

  Sends price data to Claude, which returns candidate entry and exit patterns you can backtest. Uses your Anthropic API key.

  Try: “Discover 5 candidate patterns on BTCUSDT 1h.”

  Write · evolution_discover_patterns

• Backtests one candidate pattern

  Runs a vectorized backtest of a pattern on historical data and returns Sharpe ratio, win rate, trade count, drawdown and total profit and loss.

  Try: “Backtest the first pattern.”

  Write · evolution_run_backtest

• Runs the full strategy evolution loop

  Generates patterns with an LLM, backtests them, keeps those that hold up on unseen data and drops the rest across several rounds.

  Try: “Evolve a strategy for ETHUSDT on 4h bars.”

  Write · evolution_evolve_strategy

• Lists past evolution runs

  Returns the evolution runs from this session with graduated strategies, rejected ones, token use and backtest counts. It resets when the server restarts.

  Try: “Show what evolution has run so far.”

  Read · evolution_get_log

• Exports tamper-evident decision records

  Returns the saved Trading Decision Records with the memory context behind each decision, for one trade or filtered by strategy.

  Try: “Export the audit trail for VolBreakout.”

  Read · export_audit_trail

• Checks one record for tampering

  Recomputes a record's SHA-256 hash and compares it with the hash stored at decision time.

  Try: “Verify trade MT5-7047640363 was not altered.”

  Read · verify_audit_hash

• Checks the whole record chain

  Walks the hash chain from one record to another and reports whether every link matches and where it first breaks.

  Try: “Verify the whole audit chain.”

  Read · verify_audit_chain

• Gets a day's combined audit hash

  Returns the daily Merkle root over that day's records, or rebuilds and overwrites the stored one. A rebuild can submit it to an outside timestamp service.

  Try: “Get the audit root for 2026-10-01.”

  Write · get_daily_root

Watch out for

• Its ratings can steer real trades

  check_trade_legitimacy and compute_dqs return go, reduced or skip style advice from your own history. They place no orders, but do not treat them as financial advice. github.com

• A hash is sent to a timestamp service

  README: daily audit roots are timestamped through RFC 3161 by default (a 32-byte hash, no trade data). Set TRADEMEMORY_TSA=off to stop it. github.com

• Evolution tools use outside services

  They read public Binance data. If you set ANTHROPIC_API_KEY they also send price data to Claude and use your key. github.com

• The broker proxy is separate and optional

  The optional proxy extra can forward orders to a broker through Mnemox Control. That is not part of the default install. github.com

Security advisories