MCP memory server

An open-source MCP memory server, plus the runtime most people end up needing

If you are searching for an MCP memory server, the core need is usually local persistent memory for AI agents. NEXO does that, but it also adds the parts that become necessary in real use: guard checks, workflows, doctor diagnostics, scripts, and shared-brain coordination across clients.

MCP-compatible Local persistence Runtime, not endpoint only
Install from npm Open docs
Local by default Memory stores, vectors, workflows, and operational state live on your machine instead of a hosted control plane.
Protocol compatible NEXO exposes its cognitive capabilities through MCP so multiple clients can attach to the same brain.
Operationally broader Doctor, followups, workflows, scripts, and Evolution exist because real agent operations need more than retrieval alone.

What you get at the memory layer

Persistent memory, retrieval, consolidation, and a shared brain that more than one MCP-capable client can use.

What you get around it

Pre-action guard checks, runtime doctor diagnostics, reminders, workflows, and background tasks that make the memory layer operationally useful.

Why the libre model matters

You can inspect the full stack, fork it, and run it locally under AGPL-3.0 instead of depending on a closed memory control plane.

Best fit
1

People evaluating the category

This page is for the person who knows they want MCP memory, but has not yet chosen between a thin endpoint and a fuller local runtime.

2

Teams avoiding hosted dependency

NEXO is strongest when local-first, inspectable, and self-managed behavior matters as much as the retrieval API itself.

3

Operators with more than one client

The shared-brain model is especially useful when Claude Code, Codex, and Claude Desktop should not each build isolated memory silos.

Need the shorter proof points?

The demos and use-case pages show the same story in faster, lighter form.

See demos Use cases Back to solutions