Screenshot to Agent

  • Desktop app
  • All plans
View as MarkdownFeature page

What it does

Captures a region of any screen and attaches it straight to the composer of the focused agent, as a numbered thumbnail ready to send with your prompt. It uses the operating system's own region picker, so there is no third-party tool, no PNG to save and find, no drag into the window. Two entry points run the same capture: a Screenshot button in the composer and a global keyboard shortcut that works even when AgentsRoom is in the background.

Where to find it

  • Composer toolbar of any agent: the Screenshot button opens a small popover (Attach a screenshot) with the permission state, the current shortcut and Capture now.
  • Shortcut: Cmd+Shift+2 on macOS, Ctrl+Shift+2 on Windows and Linux by default, rebindable in Settings > Composer > Screen region capture.

How to use it

  1. Press the shortcut from anywhere, or click Screenshot > Capture now.
  2. The native picker opens (crosshair on macOS, the Snip overlay on Windows, grim/slurp, spectacle, gnome-screenshot, maim, scrot or flameshot on Linux). Drag a box; Escape cancels silently.
  3. The capture is attached to the focused agent's composer as File #N, with a confirmation toast. Write your prompt and send: the image goes with the message.
  4. To circle or annotate the capture before sending, click the pencil on its thumbnail: it opens in Sketch.
  5. To say something about one attached image in particular (captured, pasted or dropped), click its thumbnail to open the preview (Left and Right arrows move between images) and type in the field under it ("Add a note about this image: it will be sent next to it"), then Done. The note leaves right after that image in the message, whatever the provider. A grey dot on the thumbnail marks an annotated image; hover it to read the note ("Note on this image"). The field offers the prompt library and dictation; Enter adds a line, it never sends.

With the shortcut, the capture goes to the agent whose terminal is focused in the current project; if no agent is open, a toast says "Open an agent first to attach a screenshot". When a project is open in its own window (see Project Window), the capture goes to that window's composer if it is in front, or was the last one you used before switching to the app you capture (since 2026-09-23).

Pasting an image works the same way. An image copied from a browser, a chat app or a screenshot thumbnail, pasted with Cmd+V / Ctrl+V into the composer or into the agent's terminal, lands in the composer as a File #N chip you can see and remove (Remove from message); it leaves with your next message, on purpose, and never on its own. Pasting text into the terminal still types it into the CLI.

Settings

  • regionScreenshotEnabled (global, Settings > Composer > Screen region capture > Enable the global capture shortcut): registers the system-wide shortcut. Default on.
  • regionScreenshotShortcut (global, same section, Capture shortcut): the key combination, recorded by pressing it; default CommandOrControl+Shift+2, chosen to avoid the macOS Cmd+Shift+3/4/5 captures.
  • composerHiddenActions (global, Settings > Composer > Composer toolbar buttons): hide the Screenshot button.

Agent tools (MCP)

None. The capture reaches the agent as a plain image attachment.

Providers

All providers whose model accepts image input (Claude Code, Codex and the other multimodal CLIs). The file is a standard PNG passed by path; nothing is provider-specific.

One difference in how the message travels, invisible unless it goes wrong: Claude Code and Grok both read a paste event in their prompt as "the user pasted an image" and go and read the system clipboard themselves. AgentsRoom therefore never sends an empty paste to any CLI (since 2026-09-17), and types a message to a Grok agent key by key instead of pasting it (since 2026-09-18), so a screenshot sitting in your clipboard is no longer attached to a message you never pasted it in. On Windows, where the terminal has no paste protocol, the question never arose.

Mobile

Not available on the phone (no global shortcut, no OS capture tool). The mobile companion has its own way of sharing a phone screenshot or photo with the focused desktop agent, see Mobile Companion. The same two protections apply to a message sent from the phone's terminal: no empty paste, and a message to a Grok agent typed rather than pasted. Notes under each image are desktop only.

Limits

  • Captures are stored locally in your home folder, outside the project, and deleted after seven days.
  • macOS: needs the Screen Recording permission, otherwise the capture is blank. The first capture triggers the system prompt; the popover shows the state and a Settings button.
  • Windows: the Snip tool has no callback, so the capture is read from the clipboard; cancelling ends after a 90-second timeout, and a second capture cannot start meanwhile.
  • Linux: requires one of the supported region tools on the PATH; without one the popover says "No screenshot tool found" and lists options.
  • A shortcut already taken by another app fails silently; the button always works as a fallback.

Common questions

  • Does it work when AgentsRoom is in the background? Yes, the shortcut is global; the capture is routed to the focused agent of the current project.
  • Can I change or disable the shortcut? Yes, Settings > Composer > Screen region capture: record a new combination or switch the toggle off.
  • Can I annotate before sending? Yes, via the pencil on the thumbnail (Sketch). Screenshot to Agent attaches the raw capture.
  • Are my screenshots uploaded somewhere? Only to the agent, with your message. They are otherwise local and transient.
  • Why is the capture black on macOS? Screen Recording permission missing: grant it in System Settings and relaunch the app.
  • A screenshot I copied earlier was sent with my message, as [Image #1], though I never pasted it. That was the CLI reading the clipboard on its own when AgentsRoom handed it a paste (Claude Code on an empty paste, Grok on any paste). Fixed on 2026-09-17 for Claude Code and on 2026-09-18 for Grok, desktop and phone: update the apps. An image now reaches the agent only as a visible chip in the composer.
  • Can I explain each image separately instead of writing "on the second screenshot..."? Yes, since 2026-09-25 on desktop: open the image preview from its thumbnail and write a note under it. Each note is sent next to its own image. Not on the phone yet.
  • I pasted an image in the terminal and nothing was typed. Expected: the image went to the composer as a File #N chip, where you can check it or remove it before sending.

Related

  • Sketch: annotate the capture.
  • Smart Hints: suggests this feature when you paste screenshots by hand or use the button often.
  • Chrome Extension: send the HTML of a page rather than its pixels.