How to paste an image or screenshot into Claude Code
Paste from the clipboard or drop a file onto the terminal. TermLink uploads it to the machine the agent is running on and types the saved path in for you, so your next prompt can just point Claude at the file — even when the agent is on another computer and you are holding a phone.
Some things are faster to show than to describe. A layout that is broken in a way you do not have the vocabulary for. A stack trace forty lines long. A design mock you want the agent to match.
Locally you would drop the file somewhere and tell Claude to look at it. The awkward part is doing that when the agent is running on a different machine and you are holding a phone.
Can Claude Code read images?
Yes — but it reads them the way any program does: off a filesystem, by path. Claude Code is a process in a shell on some computer, so the images it can open are the ones sitting on that computer's disk.
That is the whole reason getting a screenshot to it is fiddly. The picture is on your phone or in your clipboard; the agent can only see its own disk. Something has to move the bytes across and tell the agent where they landed.
Paste it, or drop it
Paste an image from your clipboard, or drag a file straight onto the terminal in the web client.
TermLink uploads it to the host — the machine Claude Code is actually running on — and then types the saved path into the prompt for you. Your next message just points at it:
> the mobile header overlaps the logo, see <the path TermLink just typed in>
Claude reads the file off its own disk, because that is where it now lives. There is no clipboard to bridge, no scp, no "put it in a gist and paste me the link."
Files up to 25 MB.
Why upload to the host and not to us
Because Claude Code is a program running in a shell on your machine, and the only file paths it can open are that machine's. Handing the agent a URL to somewhere else would just make it someone else's problem — and yours, when the link expires.
Putting the bytes on the host is the version that works with what the agent already is: a process with a filesystem.
It also means the image is subject to whatever you already do with that machine. It is not sitting in a third place you have to think about separately.
The phone case is the real one
This is a feature you appreciate away from your desk. You are looking at a bug on your phone — the app is misbehaving right there, in your hand. Screenshot it, open the TermLink tab, paste it onto the terminal, and tell the agent on your workstation to go look.
The alternative is remembering to do it when you are back at a computer, which means describing it to yourself in a note, badly, and losing the actual evidence.
More on the mobile flow in using Claude Code from your phone.
It works with other agents too
Nothing about the upload is Claude-specific. The file lands on the host and the path goes into whatever is reading input at the time — a prompt, an editor, another CLI agent.
TermLink relays a real terminal, so the image just becomes a path in it. That holds for Codex CLI and Cursor Agent the same way it holds for Claude Code, and for anything else you happen to be running in that shell.
Frequently asked questions
Can I paste an image into Claude Code?
Not into Claude Code directly — it reads files by path, not from your clipboard. With TermLink you paste onto the terminal instead: the image is uploaded to the host machine and the saved path is typed into the prompt, which is the form Claude can actually use.
How do I send a screenshot to Claude Code running on another machine?
Paste or drop it onto that machine's session in the browser. The upload goes to the host, not to your local disk, so the agent can open it immediately. You do not need SSH, scp, or a shared folder.
Does this work from a phone?
Yes. Take the screenshot, open the session in a mobile browser, and paste it onto the terminal. This is the case the feature exists for — the bug is usually on the phone, and the agent is usually not.
How large a file can I send?
Up to 25 MB.
Does it only work with images?
The upload puts a file on the host and types its path in, so the mechanism is not image-specific. Whether the agent can do anything useful with it depends on the agent.
Does it work with Codex CLI or Cursor Agent?
Yes. The file lands on the host and the path goes into whatever is reading input, so it does not depend on which agent is running. See the session manager for AI coding agents.
See getting started with TermLink to register the machine your agent runs on, or running Claude Code remotely for the setup this slots into.