How do I turn on Claude memory?

In Claude.ai, open Settings → Profile and enable the memory feature. Availability depends on your plan and region; Claude Desktop and Claude Code inherit the same account setting.

How do I clear Claude memory?

Open Settings → Profile → Memory in Claude.ai and delete individual entries or clear all. You can also ask Claude in-chat to forget something and confirm it removed the note.

Does Claude memory use more tokens?

Yes. Anything Claude remembers is injected into the system prompt every message, so more remembered facts means more tokens per turn. For large or growing context, an MCP-connected notes workspace is cheaper because Claude only pulls in what it needs.

How does Claude memory work across chats?

Saved memories persist across every chat on your account. New chats start fresh in terms of conversation history but see the same memory profile plus your Custom Instructions.

Back to home

Guide

Claude memory: how it works and how to extend it

Turn it on, clear it, and go beyond its limits

Claude's memory feature lets it remember facts about you across chats — similar to ChatGPT's saved memory. It's useful for stable preferences, but it caps out fast, costs tokens on every message, and lives inside Anthropic's account. For anything that has to grow with you or move between assistants, you need something more durable.

Turning Claude memory on

Open Claude.ai in your browser.
Go to Settings → Profile.
Enable Memory (availability varies by plan).

Claude Desktop and Claude Code read the same account setting — you only need to enable it once.

Clearing Claude memory

Two ways: from Settings → Profile → Memory, delete individual entries or clear the whole list; or ask Claude in-chat to forget something and it will remove that note and confirm.

Where Claude memory falls short

Capped. There's a hard limit on how much Claude will store. Long users hit it.
Token cost every turn. Everything Claude remembers is loaded into the system prompt each message.
Stuck in Anthropic. Memory doesn't move to ChatGPT, Codex, or any other assistant.
Opaque. You can see items in a list, but you can't organise, search, or version them.

The MCP alternative

Muninn is a private notes workspace Claude reads and writes over MCP. Instead of Anthropic remembering things for you, your notes are the memory — and any MCP-aware assistant (Claude Desktop, Claude Code, ChatGPT, Codex) can pull from and add to the same store.

No cap — write as much as you want.
No per-turn token bloat — Claude only pulls in what's relevant.
Portable — the same notes work in every assistant that speaks MCP.
Editable — real markdown files in real folders, not a mystery list.