Example workflow

You do not need to be a developer to get value from Jesse MCP. Think of it as chatting with a helper that is allowed to press the same buttons you would in Jesse—through a safe channel—instead of you clicking everything yourself.

What your assistant can help with

In plain language, a connected assistant can usually help you run and read backtests, run optimization sessions, fine-tune strategy parameters, run Monte Carlo analysis, spot overfitting risks, tweak or draft a strategy, and adjust Jesse settings (like routes), always through Jesse rather than by hand-editing random files on your computer. The goal is to handle research work that would otherwise take hours of manual testing and comparison.

You stay in the driver’s seat: if your editor asks you to approve a step, that is normal. If something is not possible through Jesse, the assistant should say so instead of guessing.

TIP

Jesse writes and maintains AGENTS.md at your project root automatically on every jesse run, so your assistant can load the MCP rules without any setup. See MCP agent rules if you want to customize it.

Example — “Get my candles ready”

Example message:

Import Binance Perpetual Futures candles for ETH-USDT on the 1h chart from 2023-01-01 through today, and tell me when it is done.

The assistant works with Jesse’s import flow (the same idea as the Import Candles screen). When it finishes, your project has the history it needs for backtests on that symbol and timeframe. You can still watch progress in the Jesse dashboard and logs like you always do.

More detail on candles: Import Candles.

Example — “Try a strategy idea and show me what happened”

Example message:

I trade DOGE on Binance Perpetual Futures. Sketch a simple long-only trend idea, test it over the last couple of years, try a few small improvements, and leave me a short write-up of what you tried.

You might end up with strategy code in your strategies/ folder and a short markdown write-up next to it (for example under a reports/ folder) so you can reread the story later—not only the chat tab.

Here is one possible shape of the strategy code (example only—not trading advice):

from jesse.strategies import Strategy
import jesse.indicators as ta
from jesse import utils


class DOGEUSDTSharpeHunter2026(Strategy):
    def _signals(self):
        fast = ta.ema(self.candles, period=21)
        slow = ta.ema(self.candles, period=55)
        adx = ta.adx(self.candles, period=14)
        rsi = ta.rsi(self.candles, period=14)
        atr = ta.atr(self.candles, period=14)
        return fast, slow, adx, rsi, atr

    def should_long(self) -> bool:
        if len(self.candles) < 60:
            return False
        fast, slow, adx, rsi, _ = self._signals()
        return fast > slow and adx > 20 and rsi > 52 and rsi < 72

    def should_short(self) -> bool:
        return False

    def go_long(self):
        _, _, _, _, atr = self._signals()
        qty = utils.size_to_qty(self.available_margin * 0.06, self.price, fee_rate=self.fee_rate)
        self.buy = qty, self.price
        self.stop_loss = qty, self.price - (1.3 * atr)
        self.take_profit = qty, self.price + (3.5 * atr)

    def go_short(self):
        pass

    def should_cancel_entry(self) -> bool:
        return False

    def update_position(self):
        fast, slow, _, _, _ = self._signals()
        if self.is_long and fast < slow:
            self.liquidate()

The write-up might simply answer: what was the goal, what changed each try, which version looked best, and what you could try next—even when the “perfect” result never shows up. That is still useful.

Example — “Is this entry signal any good?”

Example message:

Before we build the full strategy, just test the entry rule for me:
"Buy BTC-USDT 4h on Binance Perpetual Futures when RSI(14) crosses below 30."
Run a Rule Significance Test from 2022 to 2024 with 2,000 simulations and tell
me whether the p-value is low enough to be worth pursuing.

The assistant wraps that rule in a minimal strategy (entry only, no exit tuning), stages a Rule Significance Test through Jesse, fires it, polls until it’s done, and reports back the p_value, observed_mean, and annualized_return. Anything below 0.05 is statistically meaningful; above that you can decide together whether to refine the idea or move on—before sinking time into a full backtest pass.

By default, when you describe a brand-new strategy idea, the assistant will run this validation step first and only proceed to a full strategy build-out if the entry rule shows a real edge.

Before you connect

Hook up MCP first (MCP server setup), then your assistant (Connect in Codex, Connect in Cursor, Connect in VS Code, or Connect in Zed). Before you copy a prompt, make sure your assistant can read AGENTS.md.