What is an MCP memory server? A clear explanation
An MCP memory server lets AI assistants like Claude and Cursor save and search a long-term memory store. How it works, and what MyMemory's exposes.
· 2 min read
Quick answer
An MCP memory server is a program that uses the Model Context Protocol to give an AI assistant tools for saving and searching a long-term memory store. The assistant decides when to call those tools mid-conversation. MyMemory's MCP server exposes five tools (save, search, list spaces, fetch and delete) over your MyMemory library.
Large language models forget everything between conversations. An MCP memory server is one of the cleanest ways to fix that, without retraining a model or pasting context by hand.
First, what is MCP?
The Model Context Protocol (MCP) is an open standard for connecting AI applications to outside tools and data. An AI app (the client, such as Claude Desktop or Cursor) connects to one or more servers. Each server advertises a list of tools with names, descriptions and input schemas. During a conversation, the model can decide to call a tool, and the client runs it and passes the result back.
So what is a memory server?
A memory server is an MCP server whose tools read and write a memory store. Typically:
- a save tool, to write something down,
- a search tool, to retrieve relevant memories for the current task,
- fetch and delete tools, to read or remove a specific item.
Because the memory lives in the server's store rather than in the chat, it survives across conversations and across different AI clients.
The tools MyMemory exposes
| Tool | What it does |
|---|---|
save_memory | Saves text, a note or a URL, optionally into spaces |
search_memories | Hybrid semantic and full-text search, up to 100 results |
list_spaces | Lists the spaces your account can access |
get_memory_by_id | Returns one memory's full content |
delete_memory | Previews first; deletes only when called with confirm: true |
The two-step delete is deliberate: a model can't permanently remove something in one call by mistake.
Why use a hosted memory store behind it
A memory server is only as good as its retrieval. MyMemory's library already contains what you've saved everywhere else (articles, X bookmarks, Notion pages, transcripts), so Claude or Cursor can draw on your whole knowledge base, not just facts saved during past chats.
Local or remote
MyMemory's MCP server is available two ways:
- Locally over stdio: your MCP client starts
npx @everstash/mcpas a Node.js process on your machine. Runnpx @everstash/mcp setupto configure every installed client in one step, or copy a config for each client in the app under Add memory → MCP. - Remote, over Streamable HTTP: point a client at
https://api.mymemory.dev/v1/mcp. Connector-style clients such as claude.ai and ChatGPT sign you in with your MyMemory account (OAuth); clients that accept custom headers can send your API key as a Bearer token instead. No install, no local process — this is the option for clients that only support remote MCP servers.
Ready to try it? See how to give Claude long-term memory or the setup pages for Claude and Cursor.