Jesse MCP Server

# Jesse Bare Metal Integration Investigation **Date**: 2026-02-19 **Status**: Investigation Complete - Mock Data Recommended **Outcome**: Jesse multiprocessing issues prevent reliable API-based backtest result retrieval --- ## Executive Summary Attempted to fix Jesse backtest integration by switching from Docker to bare metal installation. While workers complete successfully, they crash before persisting results to the database, making API-based result retrieval unreliable. **Recommendation**: Use mock data for paper trading E2E tests. Jesse integration requires deeper investigation into multiprocessing and database connection handling. --- ## Problem Statement ### Original Issue - Jesse Docker container spawns backtest workers via multiprocessing - Workers fail with circular import error when using spawn mode - Backtest sessions never complete or persist results ### Attempted Solution - Install Jesse bare metal with Python 3.11 via pyenv - Patch multiprocessing to use fork instead of spawn - Run Jesse directly on host (no Docker isolation) --- ## Root Cause Analysis ### Docker Issue Circular import error when spawning workers with spawn mode. Jesse forces spawn mode which creates fresh Python processes that hit circular imports when loading Jesse modules. ### Bare Metal Issue Even with fork mode, database connections don't survive fork. Peewee ORM connection pool corrupted in child process. Workers crash before calling store_backtest_session(). --- ## What Works 1. Direct Python backtest calls (same process) 2. Jesse API starts and accepts requests 3. Worker processes start 4. Removed finished worker messages appear ## What Doesn't Work 1. API-based backtest result retrieval 2. Session persistence after worker completes 3. Backtest log files (all 0 bytes) 4. Result caching --- ## Evidence ### Log Files (All 0 bytes) All backtest log files are empty, indicating workers crash before writing. ### Database No new sessions persisted after API calls. Only direct Python calls succeed. ### Worker Logs Workers start and complete but session not found in database. --- ## Current State **Working**: - Jesse API on port 9200 - Strategy loading (4 strategies) - Candle data (Binance Spot, BTC-USDT) - Direct Python backtest calls - mcproxy connects to Jesse **Not Working**: - API-based result retrieval - Session persistence - Log file writing --- ## Recommendations 1. Use mock data for paper trading E2E tests 2. For future Jesse integration, consider: - Reinitializing database in worker processes - Queue-based architecture with separate API/worker - Thread pool instead of process pool --- *Full investigation details in JESSE_INVESTIGATION_INDEX.md*