QuantConnect

QuantConnect Strategy Development (Component-Based)

Develop modular strategies using reusable components: ./component

When to Load This Skill

• Creating new QuantConnect strategy
• Need to add indicators, signals, or risk management
• Troubleshooting QC-specific issues

Component Library (Progressive Disclosure)

Use components instead of writing from scratch. Load only what you need.

Discovery

# List all components
./component list

# List by category
./component list indicators
./component list signals
./component list risk_management

# Search by keyword
./component search momentum
./component search stop

Integration

# View component code
./component show add_rsi

# Get integration guide
./component explain add_rsi

IMPORTANT: Do not read component source files directly. Use --help and explain commands.

Available Components

Indicators (indicators/)

• add_rsi - RSI indicator for overbought/oversold
• add_sma - Simple Moving Average for trend detection

Signals (signals/)

• mean_reversion - RSI-based mean reversion signals
• momentum_breakout - SMA crossover momentum signals

Risk Management (risk_management/)

• stop_loss - Fixed or trailing stop loss

Sentiment (sentiment/)

• Future: Kalshi prediction market integration

Strategy Development Workflow

1. Plan Strategy

1. Choose hypothesis (mean reversion, momentum, etc.)
2. Select components needed:
   - Indicators: RSI, SMA, MACD?
   - Signals: Mean reversion, breakout?
   - Risk: Stop loss, position sizing?

2. Browse Components

./component list
./component explain COMPONENT

3. Build Strategy

from AlgorithmImports import *
from strategy_components.indicators.add_rsi import add_rsi
from strategy_components.signals.mean_reversion import MeanReversionSignal

class MyStrategy(QCAlgorithm):
    def Initialize(self):
        # Add components
        self.rsi = add_rsi(self, symbol="SPY", period=14)
        self.signal = MeanReversionSignal(oversold=30, overbought=70)
        
    def OnData(self, data):
        # Use components
        if self.rsi.IsReady:
            signal = self.signal.get_signal(self.rsi.Current.Value, self.is_long)
            # Execute trades...

4. Test Strategy

# Local test first (if possible)
# Then: ./qc_backtest run --strategy strategy.py

Common Patterns

Mean Reversion

# Components: add_rsi, mean_reversion
self.rsi = add_rsi(self, "SPY", period=14)
self.signal = MeanReversionSignal(oversold=30, overbought=70)

Momentum Breakout

# Components: add_sma, momentum_breakout
self.sma = add_sma(self, "SPY", period=20)
self.signal = MomentumBreakoutSignal(volume_confirmation=True)

With Stop Loss

# Add: stop_loss component
self.stop_loss = StopLossManager(stop_loss_pct=0.05, trailing=False)
# Call: self.stop_loss.set_entry_price(price) after entry
# Check: if self.stop_loss.should_exit(current_price): ...

Beyond MCP Principles

1. Use component CLI, not source code

   • ./component list - browse available
   • ./component explain COMPONENT - integration guide
   • Don't read .py files directly

2. Progressive Disclosure

   • Load only components you need
   • Don't load entire 955-line skill

3. Modular Architecture

   • Components are independent
   • Mix and match as needed
   • Reuse across strategies

Critical QC Errors (Still Important)

Error 1: SMA NoneType

Problem: self.sma.Current is None Fix: Check if self.sma.IsReady before using

Error 2: Data Key Missing

Problem: KeyError on data[self.symbol] Fix: Check if data.ContainsKey(self.symbol) first

Error 3: Warmup Issues

Problem: Strategy trades during warmup Fix: if self.IsWarmingUp: return

Authoritative Documentation

When confused about QC API or architecture:

• QC Docs: https://www.quantconnect.com/docs
• Component README: SCRIPTS/strategy_components/README.md

Do not guess. Use component CLI and QC docs as source of truth.

Context Savings: 120 lines (vs 955 lines in old skill) = 87% reduction

Progressive Disclosure: Use component CLI to load only what you need

Trifecta: CLI works for humans, teams, AND agents

Beyond MCP Pattern: Use --help and explain, not source code