Memory Engine MCP v1.7.0

v1.7.0 — Test Suite, Auth/Hardening, Backup/Restore, Benchmark, SQLite Concurrency

Major engineering hardening release. 135 tests, token auth, full backup/restore, recall benchmarking, and SQLite concurrency improvements.

Testing (P1)

• 135 tests across 9 files, all passing in ~14s
• Covers: CRUD, bonds, graph, merge, decay, errors, contradictions, FTS, migrations, ranking, recall, auth, backup, concurrency

Auth & Hardening (P2)

• auth.py — API token (constant-time hmac.compare_digest), Bearer extraction
• Secure bind (127.0.0.1 default), input validation, rate limiting (120 req/min)
• Web UI auth on all /api/* endpoints

Backup / Restore / Export (P3)

• backup.py — SQLite snapshots, JSON export/import, verify, cleanup
• 4 new MCP tools: backup_database, restore_database, export_all, import_data

Benchmark (P4)

• benchmark.py — CLI recall quality suite (Precision@K, Recall, MRR, latency p50/p95)

SQLite Concurrency (P5)

• PRAGMAs: synchronous=NORMAL, temp_store=MEMORY, mmap_size=256MB
• Retry on SQLITE_BUSY with exponential backoff

Backward Compatible

No breaking changes. Auth disabled by default — existing deployments work unchanged.

Full Changelog: https://github.com/SimoneB79/memory-engine-mcp/blob/main/CHANGELOG.md

Memory Engine MCP v1.6.0

What's New

Contradiction Management & 3-Tier Memory

This release introduces explicit contradiction/supersession handling and a 3-tier memory classification system, plus a new impact analysis tool and an optional web UI for graph exploration.

New Tools

• memory_contradict — Supersede an old atom with a newer contradictory one. Old atom is kept with status='superseded', linked via contradicts and supersedes bonds.
• list_contradictions — List explicit contradiction/supersession records.
• classify_memory_tier — Infer the 3-tier class (episodic/semantic/procedural) without creating an atom.
• memory_impact — Impact analysis: traverses bonds up to N hops, returns connected nodes, edges, dependents, and contradiction chains.

3-Tier Memory Model

Every atom is now classified as:

• episodic — logs, events, session messages, daily entries
• semantic — facts, decisions, consolidated knowledge
• procedural — preferences, procedures, standing rules

Inferred automatically from type/domain. Migration backfills existing atoms idempotently.

Web UI (web_ui.py)

Optional HTTP server with:

• Interactive graph visualization (circle layout, status/tier colors)
• Atom detail panel with impact analysis
• Contradiction browser
• Stats dashboard (atoms, bonds, tiers, domains)
• Full-text search

Ranking Improvements

• tier_boosts and status_penalties in ranking config
• Superseded atoms receive a -0.25 score penalty
• Semantic tier gets +0.04, procedural +0.03 boost

Infrastructure

• Pin mcp>=1.3.0,<2.0.0 (MCP 2.x removes fastmcp)
• FTS trigger fix: fire only on title/body/tags changes
• Dockerfile supports dual process (MCP server + optional UI)

Full Changelog: https://github.com/SimoneB79/memory-engine-mcp/blob/main/CHANGELOG.md

Memory Engine MCP v1.5.3

Registry compatibility release.

• Fix MCP Registry server namespace casing to match GitHub ownership (io.github.SimoneB79/memory-engine-mcp).
• Fix OCI image annotation io.modelcontextprotocol.server.name for Official MCP Registry validation.
• Keep SSE transport metadata and GHCR package metadata aligned.

Memory Engine MCP v1.5.2

Memory Engine MCP v1.5.2

Polished public release for MCP discovery and registry publication.

Highlights

• Cognitive curator for conservative memory maintenance
• Graph-aware recall with bidirectional bond expansion
• Error memory and structured preference search
• Isolated atom classification
• Session watcher with TTL raw messages and permanent digests
• Repository prepared for MCP Registry publication (server.json, OCI label, GHCR workflow)

Distribution

• Docker/GHCR workflow is configured to publish images from this tag.
• MCP Registry name: io.github.simoneb79/memory-engine-mcp