doobidoo/mcp-memory-service v8.46.0

v8.46.0 - Quality System + Hooks Integration

on GitHub

Quality System + Hooks Integration

Complete 3-phase integration of AI quality scoring into memory awareness hooks.

✨ New Features

Phase 1: Backend Quality Scoring

• Hooks read backendQuality from memory metadata (20% weight in scoring)
• calculateBackendQuality() in memory-scorer.js extracts quality_score

Phase 2: Quality Evaluation Endpoint

• New POST /api/quality/memories/{hash}/evaluate endpoint
• Uses multi-tier system (ONNX local → Groq → Gemini → Implicit)
• Returns quality_score, quality_provider, ai_score, evaluation_time_ms
• Performance: ~355ms with ONNX ranker
• triggerQualityEvaluation() in session-end.js for async scoring

Phase 3: Quality-Boosted Search

• Added quality_boost and quality_weight parameters to /api/search
• Over-fetches 3x results, reranks with composite score
• Formula: (1-weight)*semantic + weight*quality
• Returns search_type: "semantic_quality_boost" with score breakdown
• queryMemories() in memory-client.js supports qualityBoost option

🔧 Technical Details

• Hook evaluation: Non-blocking with 10s timeout, graceful fallback on failure
• Updated hook scoring weights: timeDecay (20%), tagRelevance (30%), contentRelevance (10%), contentQuality (20%), backendQuality (20%)
• Requires Memory Quality System (v8.45.0+) to be enabled

📦 Installation

pip install --upgrade mcp-memory-service

🔗 Related

• Builds on Memory Quality System (v8.45.0)
• Compatible with Natural Memory Triggers v7.1.3+

Full Changelog: v8.45.3...v8.46.0