Back to Browse

Tzivia MCP Server

Developer ToolsLow Risk10.0MCP RegistryRemote
Free

Server data from the Official MCP Registry

Persistent memory for AI agents. Hosted cloud connector or local Windows install.

About

Persistent memory for AI agents. Hosted cloud connector or local Windows install.

Remote endpoints: streamable-http: https://mcp.tzivia.dev/mcp

Security Report

10.0
Low Risk10.0Low Risk

Valid MCP server (1 strong, 1 medium validity signals). No known CVEs in dependencies. Imported from the Official MCP Registry.

Endpoint verified · Requires authentication · 1 issue found

Security scores are indicators to help you make informed decisions, not guarantees. Always review permissions before connecting any MCP server.

Permissions Found in Source Code

Found by scanning the linked source code. This listing connects to a hosted endpoint, so none of this runs on your machine: it describes what the server software does where it is hosted.

HTTP Network Access

Connects to external APIs or services over the internet.

How to Connect

Remote Plugin

No local installation needed. Your AI client connects to the remote endpoint directly.

Add this to your MCP configuration to connect:

{
  "mcpServers": {
    "io-github-mmcberry-bit-tzivia": {
      "url": "https://mcp.tzivia.dev/mcp"
    }
  }
}

Documentation

View on GitHub

From the project's GitHub README.

Tzivia

Middleware that makes LLMs faster and cheaper.

Tzivia is a persistent memory and reasoning layer that attaches to any large language model. Instead of re-feeding entire context windows on every call, Tzivia retrieves only the relevant knowledge chains — cutting search space, latency, and token cost by orders of magnitude.

⚡ ~3,500 tokens/sec ingestion: a 1,000,000-token corpus saved in under 5 minutes 🎯 50.1% vs 44.6% against Mem0 on the full LoCoMo long-term memory benchmark, with $0 spent saving memories 🧠 50ms average recall (65ms p95) from a 1,000,000-token memory


The Problem

Every LLM call today pays the same tax: context has to be stuffed into the prompt, tokenized, and attended over from scratch. As knowledge bases grow, this gets slower and more expensive linearly — or worse. Retrieval-augmented approaches help, but most treat memory as a flat vector store with no structure, no reasoning, and no persistence of insight.

The Approach

Tzivia models memory as a typed, decaying knowledge graph with a dedicated routing algorithm on top:

  • ARIA routing algorithm — sub-linear traversal that reaches relevant nodes without scanning the full store.
  • 4D vector scoring (Vec4) — every node is scored across four independent dimensions: frequency, accuracy, connectivity, and inference weight. Retrieval balances all four rather than relying on similarity alone.
  • Typed semantic relationships — 10 distinct edge types (not just "related to"), enabling a formal logic engine to reason over the graph with defined inference rules.
  • Synthesis nodes — a persistent insight layer that doesn't decay. Discovered patterns are promoted into durable knowledge instead of being recomputed every session.
  • MCP server architecture — Tzivia attaches to LLMs through the Model Context Protocol, making it a drop-in memory layer for any compatible model.

Benchmarks

MetricResultCondition
LoCoMo accuracy50.1% (Mem0 44.6%, naive RAG 35.7%, full context 53.3%)10 conversations, 1,540 questions, gpt-4o-mini answer + judge
Recall latency50ms avg / 65ms p951,000,000-token memory, 8 vCPU / 8 GB
Ingestion~3,500 tokens/sec1,000,000-token corpus, same server
Cost to save$0 (no LLM calls)Mem0 calls an LLM on every message

Measured November 2026. LoCoMo: every system got the same questions, answer model and judge, with about 500 tokens of context per question; Mem0 is the open-source mem0ai library run by us, not Mem0's hosted platform or published setup. Full results: tzivia.dev/benchmarks.

Applied Result: DNA Decay Finder

To validate the graph-decay mechanics on real data, Tzivia's decay model was applied to cancer genomics. The DNA decay finder detects mutation hotspots and was validated against the public TCGA cancer genomics dataset — demonstrating that the same routing and decay machinery that accelerates LLM memory generalizes to structured scientific data.

Architecture at a Glance

        ┌─────────────┐
  LLM ──│ MCP Server  │── attachment point
        └──────┬──────┘
               │
        ┌──────▼───────────────────────────┐
        │  ARIA Routing  (sub-linear)       │
        ├──────────────────────────────────┤
        │  Typed Knowledge Graph            │
        │   • Vec4-scored nodes             │
        │   • 10 typed relationship edges   │
        │   • Logic engine (formal rules)   │
        ├──────────────────────────────────┤
        │  Synthesis Layer  (no decay)      │
        └──────────────────────────────────┘
               │
        ┌──────▼──────┐
        │   SQLite    │  local persistence
        └─────────────┘
  • SQLite persistence — zero-configuration local storage; the full graph lives in a single file
  • Single-tenant by design — your memory graph stays on your machine

Roadmap

  • PostgreSQL / Supabase backend for hosted, multi-tenant deployments
  • Zero-knowledge encryption for multi-tenant mode
  • Real-time 3D knowledge-graph visualization (Fibonacci-sphere layout, force-directed)

Install

Tzivia-AI v5.1.0 is out (Windows).

  1. Download Tzivia-AI-Setup-5.1.0.exe from the latest release and run it.
  2. Connect your AI app:
    • Claude Desktop: download tzivia-mcp-5.1.0.mcpb from the same release and double-click it.
    • Any MCP client: Tzivia is listed in the official MCP Registry as io.github.MMcBerry-bit/tzivia.
  3. Ask your AI to remember something, then ask it to recall it.

More at tzivia.dev.

Status

v5.1.0 released November 6, 2026. Free for personal use under the Tzivia-AI EULA. Core components — ARIA routing, Vec4 scoring, the typed edge store, the logic engine, and the validated DNA decay finder — are working and benchmarked.

About

Built by Matthew McBerry, founder & CEO of Dorcas Innovations.

The full engine is under active development and kept private as core IP. This repository is a public overview — for a live demo, benchmark reproduction, or partnership inquiries, reach out.

📧 matthew@tzivia.dev


Tzivia — persistent memory and reasoning for any LLM.

Reviews

No reviews yet

Be the first to review this server!