Cortex Memory Engine
Give Claude Code or Cursor a local, encrypted memory layer so agents recall people, facts, and beliefs across sessions without sending context to the cloud.
Overview
Cortex Memory Engine is a Build-phase MCP server that provides local-first, encrypted four-tier memory, a people graph, and Bayesian beliefs for fast agent recall.
What is this MCP server?
- Four-tier memory model for structured short- and long-term recall
- People graph linking entities and relationships for agent context
- Bayesian belief updates for uncertain or evolving facts
- Local-first storage with encryption and ~62µs access latency
- Ships as Docker OCI image (ghcr.io/gambletan/cortex:v1.8.0) over stdio MCP
- Server version 1.8.0 in MCP manifest
- Advertised ~62µs memory access latency
- Four-tier memory architecture plus people graph and Bayesian beliefs
What problem does it solve?
Agents forget prior decisions and relationships every session unless you paste huge logs or pay for cloud memory you do not control.
Who is it for?
Solo builders shipping agent-first tools who need on-device memory, entity graphs, and belief tracking under Docker stdio MCP.
Skip if: Teams that want a managed multi-user knowledge base with no local ops or Docker runtime on the workstation.
What do I get? / Deliverables
After you register Cortex over MCP, your agent can read and write structured local memory with encrypted storage and sub-millisecond tool latency.
- Running Cortex MCP server connected to your agent
- Tool access to tiered memory, people graph, and belief updates
- Local encrypted memory store under your control
Recommended MCP Servers
Journey fit
Persistent agent memory is wired during product build when you connect MCP tooling to your assistant stack, even though recalled context helps every later phase. agent-tooling is the canonical shelf for MCP servers that extend what the agent remembers and reasons about, not a single app feature integration.
How it compares
MCP memory engine for agents, not a generic vector database admin skill or hosted chat memory product.
Common Questions / FAQ
Who is Cortex Memory Engine for?
Indie developers and agent builders who run Claude Code, Cursor, or similar clients and want encrypted, local memory exposed as MCP tools.
When should I use Cortex Memory Engine?
Use it while building or extending an agent workflow when session continuity, people/entity context, and uncertain facts must persist on your machine.
How do I add Cortex Memory Engine to my agent?
Add the MCP server entry pointing at the ghcr.io/gambletan/cortex Docker image with stdio transport, per the server manifest in the Cortex repository.