Claude Code memory on every machine you use
Claude Code remembers now. Its auto memory is on by default: Claude writes notes to a MEMORY.md under ~/.claude/projects, one folder per repository, and the first 200 lines load into every new session. Anthropic’s docs also say where it stops: "Auto memory is machine-local."
Connect Pando once and Claude Code gets one bullet of your outline as its memory, which is still there in a new session, on a different machine, and in a different repository.
Set it up
- In a terminal: claude mcp add --transport http --scope user pando https://pando.ink/mcp
- Start claude, run /mcp, select pando and choose Authenticate.
That is the whole setup. There is no API key to paste: Pando registers the client itself over OAuth, and you approve it once in the browser.
What each place is for
- CLAUDE.md, for instructions you write and the team shares. It loads at the start of every session.
- Auto memory, for what Claude learns in one repository on one machine. It is plain Markdown, and /memory opens it.
- A branch of your outline, for decisions and their reasons that have to reach a second machine, a second repository, another agent, or you on your phone.
What to put in it, and what not to
The test is whether you would otherwise re-derive it next week by reading three files and asking one question. If yes, it belongs in the memory. If you would find it in ten seconds by looking, it does not.
- A decision, and why it was made. Not what was done this afternoon.
- A constraint that is not visible in the code.
- A correction you gave it, so you do not give it twice.
- A pointer to something external: a dashboard, a ticket, where a key lives.
Never a secret. You can read this bullet, share it, and export it, and every other connector you approve starts with your whole outline, this bullet included.
What it looks like after a week
A branch of your own outline, in the same tree as everything else you think about. You can open it, correct a line that went wrong, move a decision somewhere it belongs better, or share that branch with a colleague. It is a bullet, so everything you can do to a bullet you can do to it.
Why not auto memory alone
If one repository on one machine is your whole world, auto memory is enough and you should use it. What it learned in one repository is not there in the next, and none of it reaches your other laptop, a cloud session, Codex or ChatGPT. The moment any of those is part of your week, the memory has to live where all of them reach. This is a URL.
What it costs
Nothing, for 1,000 bullets, with every feature included and no payment card. Pando does not meter agent calls and will not: the tokens are paid for by your own Claude subscription, so there is nothing for us to resell.
Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this