Your memory, in Claude and the other assistants you use
Until this change the web app was the only way into your vault. If you were writing to somebody in Claude and wanted to know what you had promised them, you switched tabs.
From 7 October 2026 your vault has an MCP address. You allow it once on a consent screen, and the assistant you already use can ask Tracebrain the questions you would have opened the app for: who this person is, what was said with them last week, what you owe, what a given day decided, where a project stands. The full list of what it can read and write is on the MCP page.
What it can do grew over the week:
- 7 October: read, and write what the app writes. Ten tools to read and a short list to write: add a task, tick one off, start a project, put somebody in it, correct a line a nightly pass got wrong. Every write goes through the same code the app uses, so it is dated, it refuses a duplicate, and it refuses anything that reads like a password or a card number.
- 8 October: things you told the assistant. A preference or a decision you settled in a chat with Claude never reached your vault, because the vault is built from your messages with people. Now the assistant can add it to My memory, dated the day it was written, and you can correct it there.
- 12 October: read-only is your switch. Under Manage account you choose whether an assistant may write at all. It is a setting on your side, not a permission the assistant asks for, so changing your mind does not mean connecting again. The same day a fact can be added to a person's page, and a line to a day whose note is already written.
- 12 October: a plugin for Claude Code. The connection plus six skills that know how to use it: a brief before a call, what you owe, a catch-up over several days, where a project stands, noting a promise, and remembering something about you.
- 15 October: Connect AI in the app. One page lists about twenty assistants (Claude, ChatGPT and Codex, Cursor, VS Code, Windsurf, Zed, Gemini CLI and more) with the steps, command or config each one takes. The address is always shown beside the button, so a link that stops working never leaves you stuck.
What an assistant cannot do matters as much. It never writes into the history of your conversations, which only grows. It deletes nothing. It does not create a person, and it does not change a line you typed yourself. And it only ever sees your vault: the address is the same for everybody, the vault it opens is the one you signed in to.