Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/backtest-kit/ai-trading-mcp/llms.txt

Use this file to discover all available pages before exploring further.

Paper mode lets you run the complete AI Trading MCP pipeline — Telegram feed, Claude decision loop, engine validation, position lifecycle — against live Binance market data without risking a single cent. Price feeds (OHLCV candles and order book) are fetched in real time from Binance via ccxt, but every fill, take-profit hit, and stop-loss trigger is simulated inside the backtest-kit engine. No Binance API keys are required.

​
Launch Command

npm start -- --paper --entry ./content/manual.strategy/manual.strategy.ts --ui

--paper

Activates the paper broker. The engine runs all fill logic internally — no exchange credentials needed.

--entry

Points to the strategy file. The bundled manual.strategy.ts registers no signal logic — entries come exclusively from the Claude agent.

--ui

Starts the dashboard server on port 60050. View portfolio state, open positions, PnL history, and the full MCP audit trail from any browser.

​
What Happens at Startup

1

Engine initialises persistence

Live state directories under dump/ are created. Signals, positions, and session data survive process restarts.
2

13 background loops start

One Live.background() call is issued for every symbol in CC_SYMBOL_LIST. The default whitelist is 13 pairs: BTCUSDT, POLUSDT, ZECUSDT, HYPEUSDT, DOGEUSDT, SOLUSDT, PENGUUSDT, TRXUSDT, HBARUSDT, NEARUSDT, FARTCOINUSDT, ETHUSDT, PUMPUSDT.
for (const symbol of CC_SYMBOL_LIST) {
  Live.background(symbol, {
    exchangeName: exchangeSchema.exchangeName,
    strategyName: strategySchema.strategyName,
  });
}
3

MCP HTTP bridge starts on :60051

serve() (called in config/setup.config.ts) opens the HTTP bridge on 127.0.0.1:60051. The stdio MCP server (npx @backtest-kit/mcp) forwards every get_status, open_position, and close_position call here.
4

Dashboard ready on :60050

The web UI serves the portfolio state, position history, and the markdown dump viewer so you can inspect every get_status response the agent saw.

​
The Paper Module

modules/paper.module.ts registers the ccxt-exchange schema with a public-only Binance client — no apiKey or secret fields, no authenticated endpoints: 
const getExchange = singleshot(async () => {
  const exchange = new ccxt.binance({
    options: {
      defaultType: "spot",
      adjustForTimeDifference: true,
      recvWindow: 60000,
    },
    enableRateLimit: true,
  });
  await exchange.loadMarkets();
  return exchange;
});
This single singleshot-wrapped client powers all four exchange schema methods used by the engine:
MethodWhat it calls
getCandlesexchange.fetchOHLCV — real Binance OHLCV candles
getOrderBookexchange.fetchOrderBook — live bid/ask depth
formatPriceroundTicks(price, tickSize) when tick size is available; falls back to exchange.priceToPrecision
formatQuantityroundTicks(quantity, stepSize) when step size is available; falls back to exchange.amountToPrecision
Because no credentials are configured, any attempt to call an authenticated endpoint (placing or cancelling a real order) would throw immediately — the architecture makes accidental live execution in paper mode impossible.

​
How Paper Fills Work

When the agent calls open_position, the engine computes entry cost, take-profit, and stop-loss engine-side. modules/paper.module.ts registers no broker adapter — it only supplies an addExchangeSchema with the public ccxt Binance client. All fill logic is handled entirely by the backtest-kit engine’s internal simulation using the live candle and order-book data the exchange schema provides. No orders are placed on Binance. The position, its unrealized PnL, and every state transition are written to dump/data/ so the full lifecycle survives a restart.

​
Dashboard: Port 60050

Portfolio State

Per-symbol cards showing current price, open position direction, entry price, unrealized PnL, and queued signals.

Audit Trail Viewer

Every get_status response is dumped to dump/mcp/<minute-stamp>.md with attached chart images. The UI renders these dumps so you can replay exactly what the agent saw before each decision.

​
Why Use Paper Mode

Paper mode is the right starting point for every new setup because it exercises the entire system end-to-end:
  • The Telegram scraper connects, authenticates, and begins fetching posts.
  • The MCP feed assembles portfolio state + last 15 posts (text and images) into a single get_status response.
  • The engine validates open_position calls against the symbol whitelist and existing positions.
  • The audit trail writes every byte the agent saw to disk.
  • The dashboard lets you verify positions and history visually.
Nothing about this pipeline changes when you switch to live — only the broker adapter is swapped out.
Paper mode shares the identical code path with live mode all the way down to the broker boundary. A strategy that opens, manages, and closes positions correctly in paper will behave identically in live — except fills are real and slippage is real. Run paper long enough to observe a full position lifecycle (open → TP or SL hit → close) before switching.

Build docs developers (and LLMs) love

Get started for freeTalk to us