Back to MCP Servers

Mem MCP Server

Official remote MCP server for Mem, the AI-powered self-organizing notebook. Lets AI agents create, search, and organize your notes and collections.

Collaboration by Mem (mem.ai) OAuth2 active
Overview

Mem is an AI-powered notebook that automatically organizes notes, meeting transcripts, ideas, and research. The official Mem MCP server exposes that workspace to MCP-compatible AI clients over a hosted HTTP endpoint at https://mcp.mem.ai/mcp, with OAuth-based authentication. Agents can read, write, search, and reorganize content natively, so users do not have to copy text between their assistant and Mem.

The server exposes 22 tools across four areas: Notes (create, update, search, trash, restore, find related, list, get, set creation time), Collections (create, update, delete, list, search, add/remove notes, move notes between collections, fetch metadata), Attachments (read structured content, fetch signed download URLs, ask focused questions about a single attachment), and Audio Recordings (get transcript and metadata). Because the server is provider-hosted, there is no local install or self-hosted binary, and there are no API keys to manage beyond completing the browser OAuth flow.

Mem markets first-class integration with Claude (web and desktop), ChatGPT, Claude Code CLI, Codex CLI, and Gemini CLI. Authentication is handled by Mem with guided setup in app clients and a one-line install for CLI clients.

Tools

Tool Description
create_note Create a new note, optionally with a specific ID and timestamp.
update_note Replace the full markdown body of an existing note.
get_note Retrieve the full current state of a note by ID.
list_notes List notes chronologically with cursor pagination.
search_notes Relevance-ranked search across notes using a natural-language or keyword query.
find_related_notes Find notes semantically related to the current persisted content of a given note.
set_note_created_at Modify the visible creation timestamp of a note without altering its content.
trash_note Soft-delete a note by moving it to trash.
restore_note Reverse a soft-delete on a trashed note.
read_attachment Access structured extracted content of an attachment (PDF text, OCR, transcript, etc.).
get_note_attachment_download_url Generate a temporary signed download URL for a note attachment.
answer_question_about_attachment Ask a focused question about a single attachment.
get_audio_recording Retrieve transcript and metadata for an audio recording.
create_collection Create a new collection, optionally with a specific ID.
update_collection Update title or description of an existing collection.
delete_collection Permanently remove a collection resource.
get_collection Fetch metadata for a collection by ID.
list_collections List collections with cursor pagination.
search_collections Relevance-ranked search across collections.
add_note_to_collection Link an existing note to an existing collection.
remove_note_from_collection Unlink a note from a collection while preserving both resources.
move_note Transfer a note from one collection to another.
Setup Guide

Prerequisites

  • An active Mem workspace account (sign up at mem.ai)
  • An MCP-compatible client (Claude, ChatGPT, Claude Code CLI, Codex CLI, or Gemini CLI)
  • A browser available for the OAuth consent flow

The server endpoint is hosted by Mem:

https://mcp.mem.ai/mcp

Claude (Web / Desktop)

  1. Open Settings, then Connectors
  2. Click "+" and either choose Mem from the directory or pick "Add custom connector" and paste https://mcp.mem.ai/mcp
  3. Complete the browser OAuth flow to authorize your Mem workspace

ChatGPT

Install the Mem app from the ChatGPT app store and click Connect.

Claude Code CLI

claude mcp add --transport http mem https://mcp.mem.ai/mcp

Then run /mcp inside Claude Code and complete the browser OAuth.

Codex CLI

codex mcp add mem --url https://mcp.mem.ai/mcp

Trigger any Mem tool call to launch the OAuth prompt.

Gemini CLI

gemini mcp add -s user -t http mem https://mcp.mem.ai/mcp

Then start gemini and run /mcp auth mem to complete the OAuth flow.

Generic MCP client config

For clients that accept raw HTTP MCP server config:

{
  "mcpServers": {
    "mem": {
      "transport": "http",
      "url": "https://mcp.mem.ai/mcp"
    }
  }
}

Authentication is handled by Mem via browser OAuth on first use.

Use Cases
  • Capture a meeting summary directly from a chat session: have the agent call create_note with action items and add_note_to_collection to file it under the right project
  • Ask "what did I write about the Q3 launch?" and let the agent run search_notes and find_related_notes to assemble a briefing across past notes
  • Build a research workspace: agent uses create_collection, drops in linked notes with add_note_to_collection, and reorganizes with move_note
  • Pull insights from voice memos and PDFs by chaining get_audio_recording, read_attachment, and answer_question_about_attachment
  • Inbox cleanup: list recent notes with list_notes, then update_note, trash_note, or restore_note based on what the agent finds
Example Prompts
  • "Search my Mem notes for anything about onboarding and summarize the top five points."
  • "Create a new note titled 'Q3 launch retro' with the action items below and add it to my Marketing collection."
  • "Find notes related to my last meeting note about pricing, then draft a follow-up email."
  • "Read the attached PDF in note abc123 and answer: what is the proposed timeline?"
  • "List my collections and move every note tagged 'archived' out of the Active Projects collection."
Pros
  • Official, provider-hosted remote MCP server, no local install or self-hosting needed
  • Broad tool coverage (22 tools) across notes, collections, attachments, and audio recordings
  • OAuth authentication, so there are no long-lived API keys to manage in client config
  • First-class setup paths documented for Claude, ChatGPT, Claude Code, Codex, and Gemini
Limitations
  • Requires a Mem account and is only useful if you actively use Mem as your notebook
  • Server is closed-source and provider-hosted, so behavior and rate limits are not user-configurable
  • All requests go through Mem's servers; not suitable for users who need a fully local or self-hosted memory store
Alternatives