Jesse MCP Server

Jesse MCP Server

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

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
• Community Browsing - Browse, compare, and inspect jesse.trade community strategies

Architecture

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

Available Tools

Core Tools (17)

Phase 1: Backtesting

Phase 2: Data & Analysis

Phase 3: Optimization

Phase 4: Risk Analysis

Phase 5: Pairs Trading

Agent Tools (15)

Specialized tools for autonomous trading workflows:

Community Tools (5)

Browse, compare, and inspect community strategies from jesse.trade:

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

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:

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