Skip to main content
Glama
Sign In
Universal Memory-Vault MCP Server

Universal Memory-Vault MCP Server

by Arnabest
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Nexus Memory Architecture

Standalone, pure-Python (stdlib-only) agent memory architecture extracted from Nexus Harness v1.1 alongside the Universal Memory-Vault MCP Server. Designed for predictable multi-turn recall, deterministic noise gating, cross-session continuity, temporal graph relations, Markdown Wiki compilation, and zero-VRAM offline dreaming consolidation.

Architecture Overview

Module / Component

Role

Layer

nexus/services/memory.py (Vault)

SQLite FTS5 + BM25 + CJK LIKE hybrid search, dual-stream (code vs interaction) routing, temporal relation edges, node deactivation/superseding

L2 Core

nexus/services/retrieval_gate.py (RetrievalGate)

Zero-dependency turn gate skipping trivial confirmations and filtering low-overlap retrieval noise

L2 Core

nexus/services/memory_wiki.py (MemoryWikiCompiler)

Atomic Markdown Linked Wiki compiler (INDEX.md, per-node pages with Outgoing Relations and ## Backlinks, dream digests)

L2 Core

nexus/services/memory_dreaming.py (MemoryDreamingEngine)

5-stage deterministic consolidation (SUPERSEDES dedup, anchor linking & CONTRADICTS, stale decay, session/handoff distillation, Wiki sync) with local_vram_guard

L2 Core

nexus/services/negative_extractor.py (NegativePatternExtractor)

Scoped (global/workspace/session) + TTL negative constraints with quoted/fenced/meta-question false-positive guards

L2 Core

nexus/services/user_profile.py (UserProfileService)

Thread-safe user constraints and preferences compiled into Zone-4 JIT <user_profile> injections

L2 Core

nexus/services/handoff.py & session_profile.py

4-state session isolation and structured cross-session handoff continuity

L2 Core

nexus/services/context/shadow_archive.py

Per-session lossless SQLite FTS5 archive linked to compressed turns and rewind_to support

L2 Core

nexus/branches/plugins/memory_plugin.py

Tool & lifecycle hook orchestrator (vault_*, shadow_recall, vault_wiki, memory_dream, before_turn/after_turn/session_cleared)

L4 Plugin

nexus/branches/server/handlers/memory.py

Framework-agnostic HTTP handlers (/api/memory/wiki, /api/memory/dream)

L4 HTTP

mcp_memory_vault/

Standalone JSON-RPC 2.0 stdio MCP server implementing physical dual-vault persistence, RRF/FAIR-DRI/MMR reranking, SSOT Markdown projection, sanitizer, and watchdog daemon

MCP Service

Related MCP server: mcp-zeromem

Directory Layout

nexus-memory-architecture/
├── nexus/
│   ├── infra/              # Minimal path & JSONL logging utilities
│   ├── contract/           # Plugin DI & event contract primitives
│   ├── services/           # Vault, RetrievalGate, Wiki, Dreaming, NegativeExtractor, UserProfile, Handoff, ShadowArchive
│   └── branches/           # MemoryPlugin and /api/memory/* HTTP handlers
├── mcp_memory_vault/       # Standalone Dual-Vault MCP server, engines, scripts & tests
└── tests/harness/          # Extracted & standalone tests for the Nexus memory subsystem

Quickstart & Verification

No external Python runtime dependencies are required (standard library only; pytest for tests):

pip install -e .[dev]
pytest -q

To launch the standalone Memory-Vault MCP stdio server:

python mcp_memory_vault/server.py

License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    This server cannot be deployed

    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    Resources

    Related MCP Connectors

    Related MCP Servers

    • openmemkit

      Jlnine
      A
      license
      A
      quality
      C
      maintenance
      Enables AI agents to maintain a persistent, queryable memory stored as user-owned Markdown files, with dual-channel retrieval (FTS5 and optional semantic search) and an audited write pipeline.
      11
      MIT

    • mcp-zeromem

      cubicecho
      F
      license
      Not graded
      quality
      B
      maintenance
      Enables agents to store conversational turns in a single SQLite store and recall relevant context through lexical, entity, dense, and recent views without an LLM in the recall loop. It exposes MCP tools for remembering, bulk ingest, recall, stats, and session deletion, with persistent indexes and zero-token maintenance.
      -

    • knowledge-base-server

      Francois0788
      A
      license
      Not graded
      quality
      A
      maintenance
      Gives every AI agent a shared, persistent memory by ingesting Obsidian vaults, code, transcripts, bookmarks and terminal sessions into a three-tier SQLite FTS5 store, then serving token-optimized hybrid retrieval, classification, promotion and cross-source synthesis over MCP and REST. This lets multiple agents read from and write to one compounding brain so each session starts with better context than the last.
      10 npm
      MIT

    • OwnMemory

      lordvorp
      A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to save, recall, and forget persistent memories stored entirely on the local machine in SQLite, using hybrid semantic and keyword search with on-device embeddings, plus spaces and profile context injection.
      MIT