AI & MCP
You can connect an AI assistant to your account and ask it, in plain language, to do the work you would otherwise do in the dashboard: "deploy my Essentials lab and give me the SSH command", "why is my site saying not secure?", "start the dev server in my project and check it answers".
By the end of this page you will know what that gives you, how it works, what your plan allows the assistant to do, and where to go next.
What MCP gives you
MCP (Model Context Protocol) is an open standard that lets AI assistants use outside tools. The platform offers a set of tools — 154 of them — that cover labs, the files and processes inside them, domains, HTTPS, databases, private networks, devices, SSH keys, storage, organizations and your own lab templates.
Once you connect a client, the assistant can:
- Answer questions about your account — which labs are running, what a lab's address and password are, whether a certificate is about to expire, how much storage you are using.
- Do things for you — deploy, redeploy, stop and pause labs, put a site on a domain, create a database, register a laptop on the VPN.
- Work inside a lab — read and edit your code, run commands and tests, start a dev server and check that it answers.
- Walk you through something new — describe what you want to build and it takes you through the decisions in the right order, showing you the real options on your platform at each step.
Who it is for
- Developers who already use Claude Code, Codex, Cursor or VS Code and want their assistant to reach their labs directly, instead of copying commands between windows.
- Students and newcomers who know what they want to build but not which lab, domain or database they need. The assistant can lay out the steps.
- Organization owners who want to manage members, departments and shared resources by asking rather than clicking.
You do not need to know the tool names. You talk to the assistant; it picks the tools. The names appear in these pages so you can recognise them in your client's output.
How it works
- You add the platform to your AI client once, and sign in through your browser. You approve the client by name. See Connect an AI assistant.
- When you ask for something, the client calls the platform's tools as you. It has your permissions and nothing more — it can reach what you can reach and is refused what you are refused.
- Every call is recorded. You can read the record on the MCP page's Activity tab, and deploys and stops also appear on the lab's own Activity tab marked via MCP.
The assistant never sees your password, and there is no token for you to copy. You can disconnect a client at any time from Account Settings → Security.
What your plan allows
Which tools the assistant can use depends on your account.
| Free | Pro | Administrator | |
|---|---|---|---|
| See your labs, their status, addresses, stats and deploy logs | Yes | Yes | Yes |
| See domains, certificates and diagnose broken HTTPS | Yes | Yes | Yes |
| See services, databases, networks, devices, SSH keys, storage, organizations | Yes | Yes | Yes |
| Add, download and delete VPN devices | Yes | Yes | Yes |
| Add, delete, enable and disable SSH keys | Yes | Yes | Yes |
| Deploy, redeploy, stop, pause, resume, rename, terminate labs | — | Yes | Yes |
| Change lab settings: routing, passwords, startup script, networks | — | Yes | Yes |
| Work inside a lab: files, commands, background processes | — | Yes | Yes |
| Add domains, request certificate re-issues | — | Yes | Yes |
| Create database users and databases, private networks | — | Yes | Yes |
| Manage organization members and departments | — | Yes (where your role allows) | Yes |
| Author your own lab templates | — | Yes, where template authoring is enabled for you | Yes |
| Read your action history, plan a new build | — | Yes | Yes |
| Act across other people's accounts | — | — | Yes — see MCP admin tools |
On a Free account the assistant can look at everything and manage your devices and SSH keys, but it cannot change the state of a lab — not even redeploy one you already have. A tool your plan does not include simply does not appear in your client; asking for it gets a polite "not available" rather than an error to work around.
Plan limits still apply on top of this. A Pro account's assistant can deploy, but only as many labs as your plan allows at once — see Plans and limits.
If you cannot find MCP in the user menu, or the page says MCP is not enabled on this deployment, your administrator has not turned it on. Ask them.
What it will not do
- Billing. There is no tool for payments or plans.
- Anything outside labs and their resources. Requests outside that scope are refused as a matter of policy, for everyone including administrators. Rephrasing does not change the answer; use the web dashboard instead.
- Labs it cannot see. The assistant works with the labs its own lab list shows. Anything else is managed in the dashboard.
More on this in History and safety.
In this section
| Page | Read it when |
|---|---|
| Connect an AI assistant | You are setting up a client for the first time |
| What you can ask | You want to know what is possible, area by area |
| Worked examples | You want to see a whole task from request to result |
| Workspace and files | The assistant is going to write code in a lab |
| History and safety | You want to know what it did, or keep it on a short leash |
Connect an AI assistant
Connect Claude, Codex, Cursor, VS Code and other AI clients to your account, check the connection works, and disconnect a client when you are done.
What you can ask
Every area a connected AI assistant can work in — labs, files, domains, HTTPS, databases, VPN, SSH keys, storage, organizations and templates — with example requests, the tools behind them and the one thing to know about each.
Worked examples
Ten complete conversations with a connected AI assistant — from an idea to a running app, publishing a site with HTTPS, adding a database, fixing a failed deploy, VPN, templates and organizations.
Workspace and files
How a connected AI assistant works inside a lab — opening a project, reading and editing files, running commands, keeping a dev server running, and what survives a redeploy.
History and safety
What a connected AI assistant can and cannot do, where every action it takes is recorded, which tools cannot be undone, and how to steer it or cut it off.