AI & MLOfficial

Memory

Persistent knowledge graph that lets your AI assistant remember facts, entities, and relationships across conversations.

Docs & setup

Overview

By default, an AI assistant forgets everything when a conversation ends. The Memory server fixes that: it gives the assistant a small local notebook, stored as a knowledge graph: a structured map of things (people, projects, preferences) and how they relate. You might tell it once that you run a bakery, prefer short answers, and that your busy season is December; in a fresh conversation weeks later, it can recall all of that.

This is an official reference server from the Model Context Protocol project. Everything is saved in a single file on your own computer (you choose where with the MEMORY_FILE_PATH setting), so nothing is sent to a separate service. The assistant can read, add to, and delete entries in that file, and only that file. If you ever want a clean slate, you delete the file and the memory is gone.

Instructions

npx -y @modelcontextprotocol/server-memory

Install and runYou need Node.js installed (nodejs.org). The command npx -y @modelcontextprotocol/server-memory downloads and runs the server in one step. There are no accounts, API keys, or sign-ups involved.
Choose where memory livesOptionally set the MEMORY_FILE_PATH environment variable to a file path like C:\Users\you\claude-memory.json. If you skip this, the server uses a default file inside its install folder, which can be lost on updates. Setting your own path is safer.
Connect to your clientClaude Desktop: add command "npx", args ["-y", "@modelcontextprotocol/server-memory"], and MEMORY_FILE_PATH in env to claude_desktop_config.json. Claude Code: claude mcp add memory -e MEMORY_FILE_PATH=... -- npx -y @modelcontextprotocol/server-memory
Teach it, then test recallSay: "Remember that I prefer bullet-point answers and my project is called Sparrow." Start a brand-new conversation and ask: "What do you remember about me?" If it answers correctly, persistence is working.

Starter recipes

Real tasks you can ask an agent once this server is connected.

A standing personal profile

Store your role, tools, writing preferences, and ongoing projects once, so every new conversation starts already knowing your context instead of you re-explaining it.

Try

Track people and relationships

Keep notes like "Dana is my accountant, prefers email, handles quarterly taxes" and later ask who to contact about a tax question. The graph connects the dots.

Try

Carry a project across sessions

At the end of a work session, have the assistant save decisions and next steps; the next day, ask it to recall where you left off and continue.

Try

Install

npx -y @modelcontextprotocol/server-memory

Runs on your computer. Your AI client starts it with npx when it needs it.

From the project README · checked 2 Oct 2026

Owner

modelcontextprotocol

github.com/modelcontextprotocol/servers

Downloads

4.9M

npm · all time since Nov 2024 · as of 2 Oct 2026

Checked 2 Oct 2026

License

MIT

Official server

Access

Read and write, but only to one local memory file on your machine. It cannot touch other files, your network, or any online account. Deleting the file (or pointing MEMORY_FILE_PATH elsewhere) wipes or swaps the memory entirely.

Memory

Starter recipes

A standing personal profile

Track people and relationships

Carry a project across sessions

One AI recipe you can use.Every Monday.

One AI recipe you can use.
Every Monday.