Let Claude, Cursor or Codex read what you’ve copied.

What a connected tool can do

• Search everything you’ve copied, including the words inside screenshots.
• List your most recent copies.
• Open one card and read it.
• Look at the picture itself, for an image or a screenshot.
• Pull a board, the last few copies, or a handful of cards as one ready-to-read block, the same block Copy for AI makes.
• List your boards, each with the note you wrote about what it is for.
• Count things: how many links this week, how many screenshots.

AI search may not reach the oldest cards in very large histories.

Eight tools that read. Every one, with its arguments.

What is never shared

Anything from an app you’ve told Just Copy to skip, and anything shaped like a key or a token, never reaches a tool, by any tool, in any form. A hidden card looks exactly like a missing one. The filter sits in front of every request, so there is no way around it, and a self-test proves it before every release.

Worth knowing: whatever a tool reads, it sends on to its own cloud model. That is how these assistants work. Keep the switch off for the days you would rather it didn’t.

Tools can leave cards, if you allow it

A second switch, “Let tools add cards”, is off until you turn it on. With it on, a tool can add a card to your history and file it into a board it names. The card lands as the newest, badged with that tool’s own icon, so you always know where it came from. It never pastes, and it never touches what you’ve copied unless the tool asks to, for that one card, and that is logged too.

Text that looks like a key or a secret is stored concealed, exactly as if you had copied it yourself. Thirty cards a minute, a hundred thousand characters each. With the switch off, the server is read-only, and it says so to every tool that connects.

The same block, three ways

Copy for AI in the app, the context tool for a connected assistant, and justcopy context from the command line all produce the same thing: a header with the board’s name and note, the count and the time range, then every card oldest first with its kind, source app, time and content. Text as text, links as title and address, images as the words read from them, colors as hex, contacts as fields, files as name and path.

How to connect

Open Settings → AI and turn on “Let AI assistants read my clipboard”. Claude Code, Claude Desktop, Cursor and Codex each get a Connect button; it writes the entry into that app’s own config, with the server’s address and a token. Claude Desktop goes through a small Node bridge, so it needs Node.js installed, or you can copy the config and paste it in yourself. If the address or token ever changes, the same row offers Update.

Recent activity, on the same screen, lists what was read or added: queries and counts only, never clip contents.