
Memory MCP Server: Setup for Claude, Cursor and Other Apps
Gives Claude, Cursor and other assistants a memory that lasts between chats, stored as a small graph of people, things and facts.
The Memory server lets your assistant remember people, projects and facts between chats. It keeps them as a graph of items, notes and links, saved in a file on your computer.
It runs locally from the Model Context Protocol project's reference servers, and it needs no account or API key. Your assistant decides what to save, so tell it in a custom instruction when to remember and when to look things up. If you do nothing, the file lands in the server's own folder, so set a file path to keep your memory somewhere you control.
Good for
- Claude users who are tired of repeating preferences and background in every chat
- Developers who want a coding assistant to remember project decisions
- Anyone who wants memory kept in a local file they can open, not in a vendor's cloud
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install memory MCP
Claude Code
- 1
Copy this prompt.
Add the memory MCP server to Claude Code. Run this command: claude mcp add memory -- npx -y @modelcontextprotocol/server-memory Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js.
Built from the package data we hold. Not yet tested by us.
Tools
Finds remembered people, things and facts that match a word or phrase
Searches names, types and notes in the memory graph and returns the matches with the links between them.
Try: “What do you remember about my Q3 launch plan?”
Files ·
search_nodesOpens specific remembered items by name
Returns the named items and the links between them. Names that do not exist are skipped without an error.
Try: “Open what you stored about Maria and her team.”
Files ·
open_nodesReads everything the assistant has remembered
Returns the whole memory graph: every item, note and link. Use it for a full review, since large graphs fill the chat quickly.
Try: “Show me everything you have stored about me.”
Files ·
read_graphRemembers new people, places, projects or things
Adds new items to the memory, each with a name, a type and a list of notes. An item whose name already exists is skipped and the response says so.
Try: “Remember that Maria is a designer who prefers morning meetings.”
Write ·
create_entitiesRemembers how two things are connected
Adds links such as 'Maria works at Acme'. Both items must already exist, and duplicate links are skipped.
Try: “Note that Maria works at Acme.”
Write ·
create_relationsAdds new facts to something already remembered
Appends notes to items that exist in the memory. It fails if the item is not there, so the assistant creates it first.
Try: “Add that Maria moved to Lisbon in 2026.”
Write ·
add_observationsForgets single facts about something
Removes chosen notes from an item and leaves the item and its links in place. Missing notes are ignored and the response reports how many were removed.
Try: “Forget that Maria prefers morning meetings.”
Delete ·
delete_observationsForgets how two things were connected
Removes the named links between items and keeps the items. The response reports how many links were removed.
Try: “Remove the link saying Maria works at Acme.”
Delete ·
delete_relationsForgets whole people or things along with their links
Removes items and every link attached to them, notes included. Names that do not exist are reported, not treated as errors.
Try: “Delete everything you know about the Acme project.”
Delete ·
delete_entities
Read from the source code
Permissions
medium risk · 3 things to watch
Medium risk
The assistant can read, add and delete everything in your memory file, and it saves what it chooses from your chats. Set a memory file path you back up, and check the file now and then for details you did not mean to keep.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It can delete what it has remembered
Three tools remove facts, links or whole items, and there is no read-only mode. Back up the memory file now and then if the contents matter. github.com
The assistant decides what gets saved
Anything you say in a chat can end up in the memory file, personal details included. Open the file to check, and tell the assistant what not to store. github.com
Several apps share one file safely
Each app starts its own server, and writes take turns through a lock file next to the memory file. The folder must be writable, and network drives may not honour the lock. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Sign-in and keys | No credentials. The memory graph is stored in a local JSONL file. |
|---|---|
| Data retention | Everything stays in one file on your computer until the assistant or you delete it. |
Questions
Is the Memory MCP server official?
It is a reference server in the Model Context Protocol project's own repository, so it is maintained by the protocol's maintainers. It is a small example-grade server, not a product from Anthropic's apps.
Where does it store my memory?
In a local file named memory.jsonl. By default it sits in the server's own folder, which can be a temporary cache when you start it with npx. Set the MEMORY_FILE_PATH environment variable in your config to a fixed path to keep it.
Does it need an API key or account?
No. It runs on your computer and the README lists no credentials. It needs Node.js 24 or later when started with npx, or Docker if you use the Docker route.
Can I use it with Docker instead of npx?
Yes. The README gives a Docker setup that mounts a named volume so the memory survives restarts. Without that volume, the memory is lost when the container stops.
Can it run read-only?
No. The README documents no read-only switch. Three of its nine tools delete data.
Does it share memory between Claude Desktop and Cursor?
Yes, if both point to the same MEMORY_FILE_PATH. Each app runs its own server and writes take turns through a lock file, so none is lost.
How is it different from the built-in memory in ChatGPT or Claude?
This one is a file you own and can open, and any MCP app can use it. The built-in features live in the vendor's service and work only in that app.
Information
updated Nov 3, 2026
- Publisher
- Community
- Package
@modelcontextprotocol/server-memory- Registry
- npm
- Transport
- stdio
- Last commit
- Oct 29, 2026
- Open issues
- 26
- Listed in
- Docker MCP Catalog
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- github.com · github.com
- github.com · github.com