Jesse Server

Бесплатно

An MCP server that exposes the Jesse algorithmic trading framework's capabilities to LLM agents for backtesting, optimization, and risk analysis. It provides 32

автор: bkuri

GitHub

Описание

An MCP server that exposes the Jesse algorithmic trading framework's capabilities to LLM agents for backtesting, optimization, and risk analysis. It provides 32 specialized tools for managing trading strategies and performing comprehensive market simulations via the Jesse REST API.

README

PyPI version Python 3.10+ License: MIT

An MCP (Model Context Protocol) server that exposes Jesse's algorithmic trading framework capabilities to LLM agents.

Status: Feature Complete ✅

All planned features implemented and tested. 32 tools available (17 core + 15 agent).

Installation

PyPI

pip install jesse-mcp

uvx (recommended for running directly)

uvx jesse-mcp

Arch Linux (AUR)

yay -S jesse-mcp
# or
paru -S jesse-mcp

From Source

git clone https://github.com/bkuri/jesse-mcp.git
cd jesse-mcp
pip install -e .

Usage

# stdio transport (default, for MCP clients)
jesse-mcp

# HTTP transport (for remote access)
jesse-mcp --transport http --port 8100

# Show help
jesse-mcp --help

Environment Variables

Variable	Description	Default
`JESSE_URL`	Jesse REST API URL	`http://server2:9100`
`JESSE_PASSWORD`	Jesse UI password	(required)
`JESSE_API_TOKEN`	Pre-generated API token	(alternative to password)

Features

Backtesting - Single and batch backtest execution via Jesse REST API
Optimization - Hyperparameter tuning with walk-forward validation
Monte Carlo Analysis - Statistical robustness testing
Pairs Trading - Cointegration testing and strategy generation
Strategy Management - CRUD operations for trading strategies
Risk Analysis - VaR, stress testing, comprehensive risk reports
Agent Tools - 15 specialized tools for autonomous trading workflows

Architecture

LLM Agent ←→ MCP Protocol ←→ jesse-mcp ←→ Jesse REST API (localhost:9000)
                                    ↓
                            Mock Fallbacks (when Jesse unavailable)

Available Tools (32 Total)

Core Tools (17)

Phase 1: Backtesting

Tool	Description
`backtest`	Run single backtest with specified parameters
`strategy_list`	List available strategies
`strategy_read`	Read strategy source code
`strategy_validate`	Validate strategy code

Phase 2: Data & Analysis

Tool	Description
`candles_import`	Download candle data from exchanges
`backtest_batch`	Run concurrent multi-asset backtests
`analyze_results`	Extract insights from backtest results
`walk_forward`	Walk-forward analysis for overfitting detection

Phase 3: Optimization

Tool	Description
`optimize`	Optimize hyperparameters using Optuna

Phase 4: Risk Analysis

Tool	Description
`monte_carlo`	Monte Carlo simulations for risk analysis
`var_calculation`	Value at Risk (historical, parametric, Monte Carlo)
`stress_test`	Test under extreme market scenarios
`risk_report`	Comprehensive risk assessment

Phase 5: Pairs Trading

Tool	Description
`correlation_matrix`	Cross-asset correlation analysis
`pairs_backtest`	Backtest pairs trading strategies
`factor_analysis`	Decompose returns into systematic factors
`regime_detector`	Identify market regimes and transitions

Agent Tools (15)

Specialized tools for autonomous trading workflows:

Tool	Description
`strategy_suggest_improvements`	AI-powered strategy enhancement suggestions
`strategy_compare_strategies`	Compare multiple strategies side-by-side
`strategy_optimize_pair_selection`	Optimize pairs trading selection
`strategy_analyze_optimization_impact`	Analyze impact of optimization changes
`risk_analyze_portfolio`	Portfolio-level risk analysis
`risk_stress_test`	Advanced stress testing
`risk_assess_leverage`	Leverage risk assessment
`risk_recommend_hedges`	Hedging recommendations
`risk_analyze_drawdown_recovery`	Drawdown recovery analysis
`backtest_comprehensive`	Full backtest with all metrics
`backtest_compare_timeframes`	Compare performance across timeframes
`backtest_optimize_parameters`	Quick parameter optimization
`backtest_monte_carlo`	Backtest with Monte Carlo analysis
`backtest_analyze_regimes`	Regime-aware backtest analysis
`backtest_validate_significance`	Statistical significance validation

Testing

# Install dev dependencies
pip install jesse-mcp[dev]

# Run all tests
pytest -v

# Run with coverage
pytest --cov=jesse_mcp

Status: 49 tests passing

Local Development

Prerequisites

Python 3.10+
Jesse 1.13.x running on localhost:9000
PostgreSQL on localhost:5432
Redis on localhost:6379

Start Jesse Stack (Podman)

# Start infrastructure
podman run -d --name jesse-postgres --network host \
  -e POSTGRES_USER=jesse_user -e POSTGRES_PASSWORD=password -e POSTGRES_DB=jesse_db \
  docker.io/library/postgres:14-alpine

podman run -d --name jesse-redis --network host \
  docker.io/library/redis:6-alpine redis-server --save "" --appendonly no

# Start Jesse
podman run -d --name jesse --network host \
  -v /path/to/jesse-bot:/home:z \
  docker.io/salehmir/jesse:latest bash -c "cd /home && jesse run"

Start Dev MCP Server

./scripts/start-dev-server.sh   # Start on port 8100
./scripts/stop-dev-server.sh    # Stop server

Add to OpenCode

Add to ~/.config/opencode/opencode.json:

{
  "mcp": {
    "jesse-mcp-dev": {
      "type": "remote",
      "url": "http://localhost:8100/mcp",
      "enabled": true
    }
  }
}

Documentation

Using with LLMs - How to use with MCP-compatible LLMs
Production Deployment - Production deployment guide
Jesse Setup - Jesse integration setup
Agent System - Agent architecture
AGENTS.md - Development guidelines for AI agents

API Reference

Jesse REST Client

The jesse_rest_client.py module provides direct access to Jesse's REST API:

from jesse_mcp.core.jesse_rest_client import get_jesse_rest_client

client = get_jesse_rest_client()

# Run backtest
result = client.backtest(
    strategy="OctopusStrategy",
    symbol="BTC-USDT", 
    timeframe="1h",
    start_date="2024-01-01",
    end_date="2024-01-31"
)

Mock Implementations

When Jesse is unavailable, all tools gracefully fall back to mock implementations that return realistic synthetic data. This enables development and testing without a full Jesse installation.

Key Dependencies

Package	Version	Purpose
fastmcp	>=0.3.0	MCP server framework
numpy	>=1.24.0	Numerical computations
pandas	>=2.0.0	Data manipulation
scipy	>=1.10.0	Statistical functions
scikit-learn	>=1.3.0	ML utilities
optuna	>=3.0.0	Hyperparameter optimization

Project Structure

jesse_mcp/
├── server.py            # FastMCP server with 17 core tools
├── optimizer.py         # Phase 3: Optimization tools
├── risk_analyzer.py     # Phase 4: Risk analysis tools
├── pairs_analyzer.py    # Phase 5: Pairs trading tools
├── agent_tools.py       # 15 agent-specific tools
├── core/
│   ├── integrations.py  # Jesse framework integration
│   ├── jesse_rest_client.py  # REST API client
│   └── mock.py          # Mock implementations
├── agents/
│   ├── base.py          # Base agent class
│   ├── backtester.py    # Backtesting specialist
│   └── risk_manager.py  # Risk management specialist
└── scripts/
    ├── start-dev-server.sh
    └── stop-dev-server.sh

License

MIT License - see LICENSE file for details.

Publishing

This package uses GitHub Actions with PyPI trusted publishing. To release a new version:

Update version in pyproject.toml and jesse_mcp/__init__.py
Create a git tag: git tag v1.x.x
Push tag: git push origin v1.x.x
Create GitHub release - automatically publishes to PyPI

Как установить

Добавь это в claude_desktop_config.json и перезапусти Claude Desktop.

{
  "mcpServers": {
    "jesse-mcp-server": {
      "command": "npx",
      "args": []
    }
  }
}