Simular Box

The cloud computer
for your agents.

Your agents get their own desktop, browser and terminal, so they can use apps, browse and run code without touching your computer. One line sets it up.

Paste this into your AI chat or agent
Set up box from https://box.simular.ai/install

Your agent does the setup. You approve it once in your browser.

Or set it up in your browser →

Open in Claude Open in Codex Open in Cursor
agent session Claude Code
box_7f2q · ubuntu desktop live view
Activities Tue 10:42
agent
running  VNC · watch or take over chrome · fernrun.shop
box_7f2qyou
running
box_k9m1codex
hibernated
box_b3x8cursor
waking · ~1s
box_q8c4my-agent
hibernated
01  HOW IT WORKS

Three steps to a computer.

No VM images, no VNC setup, no config files to edit. Your agent sets it up and connects.

[1]

Give your agent one line

Paste it into Claude Code, Codex, Cursor or any agent that speaks MCP.

Set up box from https://box.simular.ai/install

[2]

Approve it in your browser

Your agent opens a sign-in page. Sign in with Google or GitHub and allow it. It does everything else.

signed in · box added · box reachable

[3]

Your agent has a computer

It runs commands, browses, and works any app on the desktop. Watch live, or take over.

> give me today’s top 5 Hacker News stories in a table

02  SET IT UP YOURSELF

One command, or edit the config.

box is one remote MCP server, box. One command adds it to the AI apps on your computer, or add the URL yourself; sign in when your AI asks, no key to paste. In Claude or ChatGPT, add it as a custom connector.

terminal
npx @simular-ai/box connect
# finds your AI apps; sign in when asked
03  TOOLS

What your agent can do.

Every desktop box exposes these MCP tools.

box

Shell & files

A persistent bash session, and reading, writing and editing files on the box.

shellread_filewrite_fileedit_filelist_dir

Browser

Drive Chrome through page snapshots and element refs, fill a form in one call, read a page in a hidden tab, or work in a tab you already have open.

browser_navigatebrowser_snapshotbrowser_clickbrowser_typebrowser_fillbrowser_readbrowser_fetchbrowser_tabs

Desktop

See the screen, its windows and accessibility elements, and click, type or press keys in any app.

screenshotcomputer_statecomputer_act

Automation

Run a multi-step GUI plan in one call, script the desktop in TypeScript, or query the accessibility tree directly.

gui_planscript_rundesk

Ubuntu desktop, Chrome, shell

A full desktop, ready on first connect.

Hibernates when idle

Wakes in about a second; files and state persist.

Live screen

Watch over VNC and take control when you need to.