Skip to main content
The Model Context Protocol (MCP) lets AI clients call external tools. MemoryPlugin ships an MCP server, so any MCP client can store and recall your memory and chat history through it. This page covers the local server, which runs on your machine via Node.js.
Most people should use the hosted Remote MCP server instead. It needs no Node.js and no token in a config file: you connect by URL and approve access in your browser. Use this local server only when you specifically want a process on your own machine, for example to keep the connection fully under your control or when your client cannot do the browser sign-in flow. The tools and data are the same either way.

​
What is MCP?

MCP is a standard from Anthropic for how AI apps talk to external tools and data sources. There are two halves:
  • MCP clients: the app you chat in (Claude Desktop, Cursor, Windsurf, and others).
  • MCP servers: tools the client can call. MemoryPlugin is one, and it adds memory and chat-history recall.
Claude Desktop is a common MCP client because Anthropic created the protocol, but 100+ other tools support MCP.

​
Requirements

You need three things:
  1. Node.js to run the server. Download it from nodejs.org.
  2. An MCP client such as Claude Desktop, Cursor, or Windsurf.
  3. A MemoryPlugin account to get your auth token.
Local MCP setup is more technical than the other integrations. You edit a JSON config file by hand and paste in a token. If that is not what you want, the Remote MCP server is the easier path.

​
Setup

You do not install the server like an app. You add a config entry to your MCP client that tells it to run the MemoryPlugin server on demand.
1

Copy your auth token

In the MemoryPlugin dashboard, open the account menu and go to Settings → Integrations. Under Authentication Token, click Copy to copy Your Auth Token.
Settings modal with the Integrations tab active, showing the masked Authentication Token field with Copy and Regenerate buttons
This token is like a password to your account. Paste it into your config, but never share it. If it leaks, click Regenerate on the same screen to invalidate the old one.
2

Add the server to your client config

Add the MemoryPlugin entry to your client’s MCP config file (see the example below) and paste your token in as MEMORY_PLUGIN_TOKEN.
3

Restart and use it

Restart your client so it picks up the new server. The AI then calls the MemoryPlugin tools when they are relevant.
Account email independence: Your MemoryPlugin sign‑in email does not need to match the email you use on ChatGPT, Claude, TypingMind, or any other tool. Access and permissions are tied to your MemoryPlugin account (and API key when applicable), not third‑party logins.
For client-specific steps, see the official MCP quickstart and the @memoryplugin/mcp-server npm page.

​
Configuration example

Every MCP client has its own config format, but they follow the same shape. Here is an example for Claude Desktop: 
{
  "mcpServers": {
    "memoryplugin": {
      "command": "npx",
      "args": ["@memoryplugin/mcp-server"],
      "env": {
        "MEMORY_PLUGIN_TOKEN": "your-auth-token-here"
      }
    }
  }
}
This runs npx @memoryplugin/mcp-server when the client needs it, which downloads and starts the server automatically. There is nothing to install ahead of time.

​
Supported MCP clients

Claude Code

Anthropic’s official CLI tool

Cursor

AI-powered code editor

Windsurf

AI development environment

Cline

AI assistant for developers

Sage

AI assistant application

Roo-Code

AI coding assistant

VS Code Copilot

GitHub Copilot Chat in VS Code

And many more...

Any client that speaks MCP works

​
Available tools

The server exposes 12 tools, grouped by feature. Each tool ships with a description that tells the AI when and how to use it, so most of the time the client calls the right one on its own.

​
Memory tools

ToolDescription
store_memorySave information for future recall. The AI is prompted to proactively save preferences, decisions, and important context.
get_memories_and_bucketsRetrieve saved memories, optionally filtered by bucket. Also returns the available buckets.
search_memoriesSemantic search across memories, ranked by relevance.
list_bucketsList your memory buckets.
create_bucketCreate a new bucket to organize memories.
update_or_move_memoriesEdit a memory’s text, move it to another bucket, or bulk-move up to 100 memories at once.

​
Smart Memory tools

For buckets with Smart Memory enabled:
ToolDescription
list_bucket_categoriesGet the Smart Categories for a bucket, including summaries and topic information.
list_category_memoriesLoad every memory in a specific category.

​
Chat History tools

Search and pull from your imported and synced conversations:
ToolDescription
recall_chat_historySearch past conversations and get synthesized context back. Supports parallel queries for complex topics.
get_conversation_summaryGet one conversation’s details: the full transcript for short chats, an AI summary for long ones.
get_full_conversationRetrieve the complete transcript of a past conversation.

​
File tools

Query your uploaded documents:
ToolDescription
search_uploaded_filesSearch documents in your file buckets. Returns relevant passages with file and page info.
Chat History recall returns about 600 tokens of context by default, up to a hard cap of 2,000. That keeps recall useful without flooding the conversation.

​
What you get through MCP

  • Memories stored and recalled across conversations
  • Buckets to keep separate contexts apart
  • Smart Memory categories and summaries
  • Chat History search across your imported conversations
  • File search over your uploaded documents
  • The same data across every tool you connect, because it all reads one account

​
Troubleshooting

Confirm Node.js is installed and on your PATH. Run node --version in a terminal; you should see a version like v20.10.0.
Recopy the token from Settings → Integrations and make sure you pasted the whole string. If in doubt, click Regenerate and update your config with the new token.
  1. Check the config file for typos.
  2. Confirm the config is in the location your client expects.
  3. Run npx @memoryplugin/mcp-server directly in a terminal to see if it starts.
Restart your client fully. Some clients only detect new servers on a clean restart.

​
Next steps

Remote MCP server

The easier, no-install path: connect by URL

Memory Buckets

Organize your memories into separate contexts

Smart Memory

AI-organized categories and summaries

File Buckets

Upload documents to query via MCP