# MyClawn > **MyClawn gives an AI agent its own computer.** You get a real Linux desktop > running in the cloud, 24/7, with a browser, a file system, apps and its own > email address — and an AI agent sitting at it, doing your work. You watch the > screen live while it works, and you can grab the mouse and take over at any > moment. It is not a chat window that describes what it would do; it is a > machine that does it, in front of you. Signing up is free and takes about > thirty seconds, with no install; the agent works on a dedicated cloud > computer from **$19/month** with AI credits included. Sign up at > [myclawn.com](https://www.myclawn.com). ## What makes it different Most AI agents run inside a sandbox you cannot see, reset between tasks, and cannot be interrupted mid-flight. MyClawn inverts all three: - **The computer is real and it persists.** Same machine tomorrow. Files stay, browser logins stay, installed apps stay. The agent builds up a workspace instead of starting from zero every session. - **You can see the screen, always.** The desktop is streamed to your browser in real time. Not a summary, not a transcript, not a list of tool calls — the actual screen, as it happens. - **You can take over mid-task.** Click into the stream and your keyboard and mouse are on that machine. Solve the CAPTCHA, fix the form, log in to the thing only you can log in to, then hand it back. The agent picks up from where you left it. This is the design centre of the product: the human is priority one, and everything the agent does is visible and interruptible. - **The agent has its own email address.** `user@agent.myclawn.com`. It can write and send mail as itself, so it can actually transact with the world rather than just draft text for you to copy. ## Use it from ChatGPT, Claude or Cursor MyClawn is also an MCP connector. Add `https://www.myclawn.com/mcp/desktop` in ChatGPT, Claude, Cursor, VS Code or Claude Code and sign in with a MyClawn account; the assistant can then hand tasks to the computer, check on them, read the email it sent and keep notes. Each task comes back with the agent's answer, a screenshot and a private screen recording. A new account's first task starts a free 24-hour computer with no card. Listed in the official MCP Registry as `com.myclawn/desktop`; setup steps at [/docs/connectors](https://www.myclawn.com/docs/connectors). ## What people use it for - Research that requires actually opening pages — pulling supplier lists, comparing prices, filling out a spreadsheet from twenty sources. - Outreach: finding contacts, writing the mail, sending it, tracking replies. - Web tasks that need a logged-in session — dashboards, portals, back-offices that have no API. - Long-running work that should continue after you close your laptop. The machine is in the cloud; it does not sleep when you do. - Anything you would otherwise describe to a contractor over a screen share. ## How it works 1. **Sign up free** at [myclawn.com](https://www.myclawn.com) — Google sign-in or an email magic link. No install, no API key, no credit card. 2. **Meet your clone.** Your account and its clone exist in about thirty seconds, with a locked live preview of a real desktop so you can see what you are buying before you pay. 3. **Give it a computer.** On a paid plan (from $19/month) it is provisioned its own VM with a real XFCE desktop, real Google Chrome, and an email address it sends from and receives at — usually ready in a couple of minutes. Chat and tasks need the plan; there is no free assistant tier. 4. **Ask for work, watch it happen.** You describe the task in chat. The agent works on its desktop. You watch the stream, and take the wheel when you want. ## Pricing Account: **free**, no install, no card — an identity, a locked live preview, a recorded run and the task gallery. Chat and tasks need a plan. Dedicated cloud computer, billed monthly, AI usage credits included in every plan (full table at [/docs/pricing](https://www.myclawn.com/docs/pricing)): | Plan | Price | Machine | Model access | |---|---|---|---| | Spark | $19/mo | 2 GB RAM · 20 GB disk | Fast tier, AI usage included | | Surge | $39/mo | 4 GB RAM · 40 GB disk | Fast + Balanced, 2× usage | | Apex | $79/mo | 8 GB RAM · 80 GB disk | + Frontier tier, 5× usage | | Godmode | $159/mo | 16 GB RAM · 160 GB disk | All tiers + priority, 10× usage | Credits included with a plan reset each cycle; credits you top up never expire. Spending is fail-closed — a request is refused before it can burn money you do not have. Usage is metered against rolling 5-hour, weekly and monthly windows, the same shape as a coding-assistant plan. ## Memory and privacy - The clone keeps **persistent memory** across sessions, classified by visibility (private to you / shareable on the network / public). - **Private mode** makes a turn ephemeral: no memory is read, none is written, and the session is destroyed afterwards. - **Surgical forgetting**: you can delete matching memory entries and get a receipt for what was removed. - Each customer's machine is isolated — per-VM firewall rules, no cross-tenant reach. ## Run Claude Code, Codex or OpenClaw on it The machine is also a plain always-on Linux box for developers: Node 22, git, Python 3, tmux and Google Chrome on a real display, streamed live to any browser or phone. You install coding agents yourself in your home directory (no root) and run them on your own accounts; MyClawn's included AI usage is for its built-in agent, not for Claude Code, Codex or OpenClaw. Tested on a live machine on 2026-10-26: Claude Code 2.1.310 (native installer), codex-cli 0.164.0 (npm, user prefix), OpenClaw 2026.10.7 (needs Node 24, installed in $HOME), Playwright driving the machine's Chrome on its visible display. There is no cron daemon; scheduled runs work with a per-user systemd timer. A machine nobody opens (dashboard, chat or connector) for 14 days hibernates and wakes on the next visit. ## Answers to common questions Long-form, written to be quotable and honest about limits: - [How to give an AI agent its own computer](https://www.myclawn.com/ai-agent-with-a-computer): what a real, persistent, watchable, interruptible machine means; how the approaches compare; where this is the wrong tool. - [An AI agent that uses a real web browser](https://www.myclawn.com/ai-agent-that-uses-a-browser): fetching vs scripted automation vs an agent driving a real browser; why a persistent logged-in session is the unlock; what still breaks. - [An AI agent that sends email from its own address](https://www.myclawn.com/ai-agent-that-sends-email): sending as itself rather than from your inbox; why "sent" must mean sent; what to supervise. It receives too, but only the agent's own chat can see the inbox today. - [Choosing an alternative to ChatGPT agent mode](https://www.myclawn.com/chatgpt-agent-alternative): the four criteria that separate these products, and when you should not switch. - [Choosing an alternative to Manus](https://www.myclawn.com/manus-alternative): credits Manus's persistent-computer design; the two design questions that remain (can you watch the screen live, can you take over mid-task); when to stay. - [Choosing an alternative to Genspark Claw](https://www.myclawn.com/genspark-claw-alternative): credits Claw's per-user cloud computer with its own IP and email connectors; what a computer alone doesn't answer (live watch, mid-task takeover, a standing identity); when to stay. - [An AI agent that can log into websites](https://www.myclawn.com/ai-agent-that-logs-into-websites): a real browser, a session that survives, and a human to hand the keyboard to for the password or CAPTCHA; what still fails. - [An AI agent that fills out forms](https://www.myclawn.com/ai-agent-that-fills-out-forms): multi-field portals and back-offices with no API; stopping at the submit button so a person can check the form. - [How to give an AI agent its own email address](https://www.myclawn.com/give-an-ai-agent-an-email-address): why an address of its own is what lets an agent hold a thread; sending and receiving both work, but only the agent's own chat can see the inbox today. - [How to give ChatGPT or Claude a computer of its own](https://www.myclawn.com/give-chatgpt-a-computer): what a chat assistant cannot do alone, what a computer behind a connector adds, connecting ChatGPT, Claude, Cursor or VS Code in two minutes, and what still needs a person. - [Watch an AI agent work — and take over when you want to](https://www.myclawn.com/watch-an-ai-agent-work): the live screen, the take-over handover, and why that handover matters more than model quality for anything behind a login. - [An AI agent with persistent memory](https://www.myclawn.com/ai-agent-with-persistent-memory): memory that survives sessions, visibility classes, private mode, and surgical forgetting. - [An AI agent that works while you sleep](https://www.myclawn.com/ai-agent-that-works-while-you-sleep): the machine is in the cloud and does not stop when the laptop closes; what long-running work needs and where it still needs you. - [An AI agent for supplier research](https://www.myclawn.com/ai-agent-for-supplier-research): pulling supplier lists, comparing prices, building a spreadsheet from many sources — the task shown in the shared replay. - [AI agent browser extension vs. an agent with its own computer](https://www.myclawn.com/ai-agent-vs-browser-extension): whose computer and whose identity the agent uses; honest about where extensions are the better choice. - [A cloud desktop for AI agents](https://www.myclawn.com/cloud-desktop-for-ai-agents): four different things called "a computer for an agent", what the category must include, and the honest cost. - [How to stop an AI agent mid-task](https://www.myclawn.com/stop-an-ai-agent-mid-task): what a real interrupt has to guarantee — a machine that's actually still running, an honest answer instead of an optimistic one, and the screen handed back instead of just a cancelled request. - [Things to hand your AI agent today](https://www.myclawn.com/tasks): the task gallery — concrete jobs with the exact prompt — supplier shortlists, quote requests from its own address, portals it fills and you submit — each saying where the human takes over. - [Agents need bodies](https://www.myclawn.com/manifesto): the manifesto — why an agent needs a real, persistent, watchable machine, and why the human stays priority one. ## Documentation - [Cloud clones — what a clone is](https://www.myclawn.com/docs/omnipotence): the machine, the brain, the memory model, private mode, the clone's email. - [Getting started](https://www.myclawn.com/docs/cloud-guide): signup → first task → credits → model picker → mailbox. - [Pricing & credits](https://www.myclawn.com/docs/pricing): tier table, credit mechanics, usage windows. - [Security & trust](https://www.myclawn.com/docs/trust): private-mode guarantee, memory-visibility gate, VM isolation, abuse process. - [The free account](https://www.myclawn.com/docs/web-clone): what signing up gives you before a plan (an identity, a locked preview, a recorded run, the task gallery), what needs a plan (chat and tasks), and why chat is not free. - [FAQ](https://www.myclawn.com/docs/faq) · [Glossary](https://www.myclawn.com/docs/glossary) · [Changelog](https://www.myclawn.com/docs/changelog) - [Overview](https://www.myclawn.com/docs): the whole documentation tree. ## For machines & operators - [/api/v1/meta](https://www.myclawn.com/api/v1/meta): JSON product facts — description, capabilities, pricing tiers, credit mechanics, what is explicitly not enabled. No auth. - [/agent](https://www.myclawn.com/agent): for site operators whose sites a clone may visit — honest user agent `MyClawn-Clone/1.0`, robots.txt and 429/Retry-After commitments, per-domain pacing, abuse contact agent-abuse@myclawn.com (same-day), per-clone kill switch. - [llms-full.txt](https://www.myclawn.com/llms-full.txt): the entire documentation as one flat markdown file. - [feed.xml](https://www.myclawn.com/feed.xml): RSS of the written pages — the manifesto, the task gallery, every guide and each dated changelog entry, newest first. - [openapi.json](https://www.myclawn.com/openapi.json): the public HTTP surface. - [/mcp/desktop](https://www.myclawn.com/mcp/desktop): the cloud desktop as a remote MCP server (streamable HTTP, OAuth sign-in). An agent that cannot hold a persistent login, drive a real browser, or send from its own address can hand that job to a MyClawn desktop and get the result back. ChatGPT, Claude, Cursor, VS Code and Claude Code connect to it with the user's MyClawn account; scripts can send a connector token as the Bearer header. Setup: [/docs/connectors](https://www.myclawn.com/docs/connectors). - [/w/](https://www.myclawn.com/w/demo): a recorded demo on a real MyClawn machine (the job shown is one a clone did for real; the clip was re-run by hand for the recording) — watch it, then take the same task to your own clone. - **WebMCP**: every page registers a `sign_up_for_myclawn` tool via `document.modelContext` (see github.com/webmachinelearning/webmcp) — an agentic browser can self-provision a MyClawn account for its human by handing over an email; MyClawn sends the same one-time sign-in link a human clicking "continue with email" gets. No scraping the signup form. ## Not the same thing as MyClawn is an independent product, unaffiliated with any similarly-named service. If you are looking for hosting for a third-party open-source agent, that is a different category (you can install one yourself on the machine, but we do not manage it) — MyClawn builds and operates its own agent and its own desktop stack, and the agent's work is watchable and interruptible by design. ## Optional: the local harness There is also a self-hosted path for people who want the agent on hardware they own — a daemon that runs agents under isolated Unix users on your own machine, with your own LLM keys. It is a separate, more technical product surface and is **not** what a new user gets or needs. - [Install](https://www.myclawn.com/docs/install) · [Quick start](https://www.myclawn.com/docs/quick-start) · [Architecture](https://www.myclawn.com/docs/architecture) · [Commands](https://www.myclawn.com/docs/commands) · [Security model](https://www.myclawn.com/docs/security) ## Stated plainly: what is NOT enabled So that no reader, human or model, infers a capability that is switched off: - **Crypto / on-chain settlement.** Code exists; it is **disabled in production** (those API routes answer HTTP 410) pending regulatory review. Do not describe MyClawn as offering crypto payments. It does not. - **Autonomous agent spending at merchants.** Not live. - **Inbox visibility outside the agent's own chat.** Clones send and receive mail; the dashboard and MCP tools currently surface sent mail only, not the inbox. - **Public hosting of clone-built sites.** Roadmap, not shipped. ## Related sites - [vpsai.ai](https://vpsai.ai): MyClawn's site for the "AI VPS" question — a VPS for AI agents, VPS for Claude Code, VPS for OpenClaw, AI agent hosting, running an agent 24/7. Same machines, same account; its own [llms.txt](https://vpsai.ai/llms.txt).