Mem0 MCP Server

The Mem0 MCP server gives AI assistants a long-term memory. Normally every session starts with a blank sheet: the assistant no longer knows which programming language you prefer, how your project is structured, or what you decided yesterday. Through Mem0's server it can store such information deliberately and retrieve it later, across clients. What you told Claude Code can be looked up by Cursor the next day, provided both are connected to the same Mem0 account. The server is operated by Mem0 itself and described in the official documentation.

What is the Mem0 MCP server?

The server implements the Model Context Protocol (MCP), an open standard through which AI applications talk to tools. It runs hosted at https://mcp.mem0.ai/mcp; nothing runs on your machine, and the memories live in your Mem0 account. The earlier open-source repository mem0ai/mem0-mcp (Apache-2.0) has been archived on GitHub since March 24, 2026 and itself points to the hosted server. The variant described in the documentation at docs.mem0.ai is therefore the authoritative one. Mem0 itself is also available as an open-source project and as a platform; according to the documentation the MCP server requires an account on the platform.

Which tools the server provides

The documentation groups the tools into four sets. add_memory is for storing. For searching and reading there are search_memories, get_memories, and get_memory, where search works by meaning and not only by keywords. With update_memory you correct an existing entry. For deleting there are delete_memory, delete_all_memories, and delete_entities. For looking things up there are list_entities, list_events, and get_event_status; the latter show whether an asynchronously processed request has finished. Memories can be assigned to individual people or agents, so an assistant only sees the memory of the respective user.

Setup and sign-in

You need a Mem0 platform account, Node.js 18 or later for the helper tool, and an MCP-capable client such as Claude, Claude Code, Codex, Cursor, Windsurf, VS Code, or OpenCode. The quickest route is npx mcp-add with the name mem0-mcp, the type http, and the server address; the tool writes the server into the configuration of the chosen clients, after which you restart the clients. In Claude Desktop you instead add a custom connector with the address under Settings > Connectors. On first tool use the client opens a browser window in which you authorize access to your account. For clients without browser sign-in or for automated workflows you send the API key from the dashboard as a bearer token. Without either, the server answers with a 401 error.

Security and privacy

A memory is only as good as its upkeep, and it is sensitive: what the assistant stores stays with Mem0 permanently and can flow into later conversations. Do not store passwords, keys, or third-party personal data, and occasionally check with search_memories or get_memories what has actually been saved. delete_all_memories, which removes the entire stock, calls for particular care; approve such calls only after your own review. Keep the API key out of committed files and supply it through an environment variable. Because memories enter the language model's context, they also reach your client's model provider.

Costs

The documentation names no separate fee for the MCP server; usage counts toward your Mem0 account. According to the pricing page there is a free Hobby plan with limited monthly write and retrieval requests and one project, an inexpensive Starter plan, a Pro plan for production use with multiple projects, graph memory, and additional analytics, and Enterprise with custom terms, audit logs, and single sign-on. Current amounts and limits are on the official pricing page.

Who is the Mem0 MCP server for?

It is worthwhile for developers who use several assistants and do not want to re-explain project rules, style guides, or decisions every time. It also suits personal assistants that should retain preferences over weeks. Anyone who only chats occasionally hardly needs it, and anyone with strict data-handling requirements should clarify beforehand whether hosted storage is permitted. The advantage over a simple notes file is search by meaning and a shared store for several clients.

A Mem0 platform account, Node.js 18 or later for the mcp-add helper, and an MCP client such as Claude, Claude Code, Codex, Cursor, Windsurf, VS Code, or OpenCode.

The Mem0 MCP server runs hosted by Mem0 and gives your assistants a shared long-term memory. This guide walks through setup, sign-in, and the first calls.

Step 1: Prerequisites

You need a Mem0 platform account, Node.js 18 or later, and an MCP client such as Claude, Claude Code, Codex, Cursor, Windsurf, VS Code, or OpenCode.

Step 2: Connect with mcp-add

Run npx mcp-add --name mem0-mcp --type http --url https://mcp.mem0.ai/mcp and choose your clients. The tool writes the server into their configuration. Restart the clients afterwards.

npx mcp-add --name mem0-mcp --type http --url https://mcp.mem0.ai/mcp

Claude Desktop and Codex

mcp-add does not support Claude Desktop: add a custom connector with the address under Settings > Connectors. For Codex, add a block [mcp_servers.mem0] with the address and bearer_token_env_var = "MEM0_API_KEY" to ~/.codex/config.toml. Do not additionally use the Mem0 plugin, otherwise the server is registered twice.

Step 3: Sign in

On first tool use the client opens a browser window in which you authorize access. For clients without browser sign-in or for CI, send the API key from the dashboard as a bearer token and keep it out of committed files.

First calls

Ask the assistant to remember a preference and retrieve it in a new session with a search. Afterwards check with get_memories what was stored, and approve delete_all_memories only after your own review.

Browser sign-in (OAuth) on first tool use, or a Mem0 API key as a bearer token; without either the server answers with 401.

Store, search, read, update, and delete memories in the Mem0 account and list entities and events.

The server runs hosted by Mem0, and memories are stored in the Mem0 account. Retrieved memories enter the AI client context and therefore reach the model provider.

delete_all_memories removes the entire stock; accidentally stored secrets or personal data remain retrievable permanently; an API key in the configuration can leak.