doobidoo/mcp-memory-service v10.45.0

v10.45.0 — OpenAI-Compatible Quality Scoring + Soft-Delete UPDATE Guards

on GitHub

What's New

feat(quality): OpenAI-compatible quality scoring provider (PR #790)

Point quality scoring at any OpenAI /v1/chat/completions-compatible endpoint — Ollama, LiteLLM, MLX-LM server, or vLLM — without a cloud API key or the ONNX model.

Quick start:

MCP_QUALITY_AI_PROVIDER=openai-compatible
MCP_QUALITY_AI_BASE_URL=http://localhost:11434/v1   # Ollama
MCP_QUALITY_AI_MODEL=qwen2.5:7b-instruct
# MCP_QUALITY_AI_API_KEY=ollama                     # optional

Recommended models: qwen2.5:7b-instruct (Ollama), mlx-community/Qwen2.5-7B-Instruct-4bit (MLX), or any instruct model via LiteLLM proxy.

New fallback chain: local ONNX → openai-compatible → Groq → Gemini → implicit signals. Endpoint failures fall through silently — no exception bubbles to the storage path.

18 new tests in tests/test_openai_compat_quality.py.

fix(storage): Soft-delete UPDATE guards (PR #783)

Seven UPDATE memories SET ... statements in sqlite_vec.py were missing the AND deleted_at IS NULL guard, meaning they could operate on soft-deleted (tombstoned) rows. All seven have been patched. No behavioral change for live rows. Continues the series from PRs #557, #558, #562.

Upgrade

pip install --upgrade mcp-memory-service

Full Changelog

See CHANGELOG.md for complete details.

🙏 Special Thanks

• @filhocf — PR #783 (soft-delete UPDATE guards). Consistent contributor improving SQLite storage correctness across multiple releases.