A MyClawn project
Your context — rules, facts, preferences — distilled into a vault you own. Any AI tool reads it over HTTP or MCP. Agents can propose new memories, but nothing gets in without your approval.
No passwords, no email. A passkey from your device is the only way in.
You already wrote your context down somewhere — a CLAUDE.md, custom instructions, a bio, chat exports. MyMemory turns it into memory every tool can share.
Paste text or drop a file. It is distilled into atomic entries — directives, facts, preferences, notes — instead of one blob every tool parses differently.
Imports and agent proposals land in a review queue. You promote what is true and reject what is not. The vault only contains what you approved.
One URL, one key. Claude Code reads it through MCP, ChatGPT through a paste or an action, anything else with a plain HTTPS GET.
Create a vault, mint a key, and hand any tool the same memory. This is what a connected tool sees.
curl -H "Authorization: Bearer mm_28dbde38035867722fed80de7a99a48f" \ https://www.myclawn.com/api/mymemory/context
https://www.myclawn.com/api/mymemory/context
curl -H "Authorization: Bearer mm_…" \ https://www.myclawn.com/api/mymemory/context
{
"mcpServers": {
"mymemory": {
"command": "npx",
"args": [
"-y",
"mymemory-mcp"
],
"env": {
"MYMEMORY_API_KEY": "mm_…"
}
}
}
}Passkey sign-in, revocable keys, delete any entry any time. Your memory belongs to you, not to a lab.
Plain HTTP and MCP — no SDK, no lock-in. Switch models, keep your memory. Neutral ground across AI tools.
After a conversation, an agent can propose what it learned. Proposals wait in your queue until you say yes.
The questions people ask AI assistants about memory, answered straight.
Yes — ChatGPT remembers things about you, and Claude keeps project instructions and chat context. But that memory lives inside each lab's product. It does not follow you to other tools, you cannot inspect it as plain entries, and it is gone if you switch. MyMemory is a separate vault that you own and any tool can read.
Three ways. Ownership: your context lives in a vault you control, not in a lab's silo. Portability: every tool reads the same vault over HTTP or MCP, so switching models does not mean starting over. Control: memories are explicit entries — directives, facts, preferences, notes — that you can read, edit, and delete, and new ones enter only through a review queue you approve.
Anything that can make an HTTPS request or speak MCP. Claude Code and other MCP clients connect natively, ChatGPT works through a paste or a custom GPT action, the iOS app reads it directly, and anything else can fetch your memory with a plain GET request and your key.
Create a vault, mint a key, and point the tool at it. For Claude Code and other MCP clients, add MyMemory to your MCP config (npx mymemory-mcp with your key in the environment). Tools without MCP support read the same memory with a single authenticated HTTPS GET.
Agents can propose, not write. After a conversation, an agent may suggest what it learned about you, and those proposals land in a pending-review queue. Your vault changes only when you approve an entry — reject anything that is wrong or unwanted.
Sign-in uses passkeys, so there are no passwords to leak. Access keys are shown once and stored hashed — the raw key never sits in the database. You can delete any entry at any time, and your memory is served only to tools that present your key.
Nothing. MyMemory is free: create a vault, import your context, mint keys, and connect as many tools as you like.