Urlbox CLI: screenshots from your terminal — or your agent

The Urlbox CLI lets you call the Urlbox screenshot API straight from your terminal. You can also hand it to an AI agent and have it taking screenshots in under a minute.

>

Everything the CLI does was already possible through the Urlbox API. The difference is that you can now tell an agent "use the Urlbox CLI to screenshot these URLs" and it will install the tool, work out how to use it, and save its first screenshot in about a minute.

To take a full-page screenshot from your shell and save it to a file:

urlbox screenshot https://stripe.com --full-page --output stripe.png

PDFs and videos work the same way:

urlbox pdf   https://stripe.com/pricing --output pricing.pdf
urlbox video https://theverge.com       --output verge.mp4

Signing in

urlbox login opens your browser, you approve the device, and the CLI is signed in to your account. From there you can manage your account without opening the dashboard:

urlbox login            # sign in through your browser
urlbox me               # the signed-in account, active org and project
urlbox orgs select      # switch organisation
urlbox projects select  # switch project
urlbox usage            # render usage for the current period
urlbox storage list     # storage credentials (proxies and llm too)
urlbox commands         # everything else the CLI can do

On CI, or any machine without a browser, skip login and set URLBOX_API_SECRET to your project's secret instead. Rendering works with just that.

Urlbox is agent-accessible

Three pieces work together:

• The API is the rendering engine, same as always.
• The CLI puts it in your terminal.
• The agent skill teaches your agent how to use it.

Anything you can do with the Urlbox API, you can now get an agent to do in plain English.

The experiment we ran

We walked the exact cold path an agent takes (install, discover, render) and timed every step. From a cold machine to a saved screenshot took about a minute. Here's where the time goes:

• Installing from npm takes about 30 seconds.
• urlbox commands returns every command and flag as structured JSON, so the agent can read the tool instead of guessing at it.
• urlbox screenshot https://stripe.com --full-page --dry-run comes back with "payload validated, no API call made". The agent has confirmed the correct command without holding any credentials.
• The one wall it hits is auth: code: auth, "no API secret configured", with a link straight to where you get one.
• Grabbing the key takes about 30 seconds, and then the first screenshot lands in under 6 seconds.

Nearly all of that minute is npm installing and you fetching your key. The render itself takes seconds. The one thing the agent can't do for you is paste in your own credential, and that's deliberate: you decide when your credentials get written to disk, not the agent.

How we built it for agents

We took our cue from 37signals, who recently made Basecamp agent-accessible by pairing their API with a CLI and a set of agent skills. We applied the same idea to screenshots, with one goal throughout: make the tool as machine-friendly as possible.

If you're building something an agent should be able to pick up cold, a few decisions made the biggest difference:

• Make every command self-describing. urlbox commands and --help emit structured JSON, so an agent can read the tool instead of guessing at it.
• Give it a dry run. --dry-run validates a request without spending anything, so an agent can confirm it built the right call before committing to it.
• Ship it as an agent skill. Running urlbox skill install --target claude-code drops a skill file the agent auto-discovers, so it knows the tool from the first prompt (Cursor, Codex and opencode too).
• Don't hide the rest of the API. The CLI passes any option straight through, so anything you can render through Urlbox, an agent can render for you.

That's why "take every URL in this CSV and screenshot it into a folder" becomes a sentence, not a script.

Frequently asked questions

What is the Urlbox CLI?

The Urlbox CLI is a command-line tool that calls the Urlbox screenshot API from your terminal. You can capture screenshots, PDFs, and videos of any URL with a single command, and it ships with an agent skill so AI coding agents can drive it directly.

How do I take a screenshot from the command line?

Install the CLI with npm install -g @urlbox/cli, run urlbox login to sign in through your browser, then run urlbox screenshot https://example.com --output shot.png. Once you're signed in, the first screenshot lands in under 6 seconds.

How fast can an AI agent start using the Urlbox CLI?

In our timed cold-start test, going from never having seen the tool to a saved PNG took about a minute: roughly 30 seconds to install from npm, 30 to fetch an API key, and a few seconds for the render itself.

Can the Urlbox CLI capture PDFs and videos too?

Yes. Use urlbox pdf <url> --output file.pdf for PDFs and urlbox video <url> --output file.mp4 for videos. The CLI passes any Urlbox API option straight through, so anything the API can render, the CLI can too.

Which AI agents does the Urlbox CLI work with?

Any agent that can run a terminal command can use it, because it's just a CLI. To make it plug-and-play, we ship ready-made agent skills for Claude Code, Cursor, Codex, and opencode: run urlbox skill install --target <your-agent> and it knows the tool from the first prompt. Other agents can still drive it straight from urlbox --help.

Try it from your terminal or your agent

The CLI is under active development, and we'd love to hear about anything you run into. You'll need an Urlbox account to render: start a 7-day trial, then install from npm or GitHub:

npm install -g @urlbox/cli
urlbox login
urlbox screenshot https://urlbox.com --output hello.png

Or just ask "use the Urlbox CLI to screenshot this URL" and let your agent take it from there.