Deploy VPS with artificial intelligence: the GINERNET MCP server

Imagine opening your AI assistant's chat and typing: "Deploy this website on a server". A few minutes later you have a freshly created VPS, your application running, the firewall properly configured and the exact hourly cost confirmed before anything was created. That is exactly what the GINERNET MCP server enables: a secure channel for AI agents such as Claude, Cursor, VS Code Copilot, Windsurf or Zed to deploy and manage pay-per-use VPS servers on your behalf.

In this article we explain what an MCP server is (even if you have never written a line of code), how to enable it on your account step by step, what a connected agent can and cannot do, and we include real prompt examples you can copy and paste - from deploying your first website to managing a small server fleet.

What is MCP (Model Context Protocol)?

MCP stands for Model Context Protocol, an open protocol created by Anthropic in 2024 and since adopted by practically the whole AI ecosystem (OpenAI, Google, Microsoft, Cursor...). Its goal is easy to grasp: give AI models a standard, secure way to use external tools.

The most common analogy is USB-C: before it existed, every device needed its own cable. Before MCP, every integration between an AI and an external service was a custom development. With MCP, any compatible assistant can "plug into" any service that exposes an MCP server and use its tools right away, with no extra programming.

An MCP server is therefore the piece that publishes those tools. GINERNET's exposes the operations of our cloud VPS platform (browsing the catalog and prices, creating servers, opening firewall ports, resizing...) so that your agent can use them just like you would use the panel - but conversationally.

And an "AI agent"? It is an assistant that doesn't just answer: it acts. Claude Code, Cursor or Copilot can run commands, write code and, thanks to MCP, operate external services. You describe the goal ("I want this website online") and the agent chains the technical steps for you.

What can an agent connected to GINERNET do?

The GINERNET MCP server exposes 10 tools covering the full lifecycle of a VPS:

With those pieces, an agent can complete end-to-end flows like this one:

1. It checks your balance and the price catalog.
2. It proposes the smallest server that can run your application and confirms the real total cost (public IPv4 included).
3. It creates the VPS with a new SSH key generated for that server.
4. It waits for provisioning to finish (1-3 minutes), connects over SSH and installs your application.
5. It opens only the firewall ports the application needs.
6. It hands you the working URL plus the hourly and monthly cost.

What an agent CANNOT do (designed for your peace of mind)

Giving an AI access to your infrastructure sounds risky, so the channel is designed around the principle of least privilege:

• It cannot delete servers. VPS deletion only exists in the web panel, for humans. This is deliberate: the worst possible agent mistake can never be destructive.
• It cannot spend more than you decide. You configure a maximum number of API-created servers and a maximum hourly spend for the channel. If the agent hits them, the operation fails with a clear message. The prepaid balance is an additional hard cap: no balance, no spend.
• It cannot touch the limits. Channel limits can only be changed from the panel, with your session and your passkey. An agent can never raise its own limits.
• Keys with separate permissions. Each API key is scoped to read-only (inspect) or provision (create and operate). You can give a monitoring agent a key that cannot create anything.
• Firewall closed by default. Every new VPS starts with inbound traffic blocked except SSH and ping. The agent must explicitly open each port, and the channel rules forbid exposing databases to the internet.
• Full audit trail. Every key records its last use and every action is traced in the VPS activity history.
• The key is shown only once and stored encrypted (SHA-256 hash). If you lose it, revoke it and create another one in seconds.

How to enable it, step by step

You need a GINERNET account with balance. Then:

1. Log into the Manager and open AI Agents in the sidebar (or go straight to /agent-access).
2. Enable the channel with the toggle. It will ask you to confirm with your passkey: it is a sensitive action and we want to be sure it's you.
3. Review the spending limits: maximum API-created VPS and maximum euros per hour. Start conservative; you can always raise them.
4. Create an API key, give it a descriptive name (for example "Claude laptop") and pick the scope (read or provision). Copy the gcp_... key right away: it will not be shown again.
5. On the same page you will find the endpoint (https://manager.ginernet.com/mcp) and tabs with the exact configuration snippet for Claude Code, Claude Desktop, Cursor, VS Code, Windsurf and Zed.

For example, Claude Code takes a single command:

claude mcp add --transport http ginernet https://manager.ginernet.com/mcp \
  --header "Authorization: Bearer gcp_YOUR_KEY"

And Cursor takes a block in ~/.cursor/mcp.json:

{
  "mcpServers": {
    "ginernet": {
      "url": "https://manager.ginernet.com/mcp",
      "headers": { "Authorization": "Bearer gcp_YOUR_KEY" }
    }
  }
}

Clients that don't support native HTTP headers yet (Claude Desktop, Windsurf, Zed) connect just as easily through the mcp-remote bridge; the exact snippet is in their tab in the panel.

Real prompt examples

The following prompts work as-is in any connected agent. Notice the pattern: you describe the goal, not the technical steps.

1. Your first website online (beginner)

Deploy a website for my photography business on a Ginernet server.
I want something simple: a landing page with my services and a
contact form. Use the cheapest possible server and tell me how much
it will cost per month before creating it.

The agent will browse the catalog, propose the default VPS (1 vCPU / 1 GB / 10 GB), confirm the total price with the IP included and, after your approval, create the server, generate the website, serve it with nginx and open ports 80 and 443. Total: a few minutes and not a single line of code written by you.

2. A production-ready WordPress (intermediate)

Set up WordPress on a new Ginernet VPS for my domain mycompany.com.
Install PHP, MariaDB and nginx, configure HTTPS with Let's Encrypt
and leave it ready for me to run the install wizard. The database
must NOT be reachable from the internet. When you're done, tell me
which DNS records I need to create.

Here the agent chains more pieces: it creates the VPS, installs the full LEMP stack, keeps MariaDB local-only (never opening port 3306, exactly as the channel rules require), issues the TLS certificate and returns the exact DNS records you must point at the server's IP.

3. Scaling when traffic arrives (intermediate-advanced)

The website on the "online-store" VPS has been slow since yesterday.
SSH in, check whether the bottleneck is CPU, RAM or disk, and grow
it to the next size if needed. Confirm the new hourly cost with me
before applying the change.

The agent diagnoses live (top, memory, I/O), decides whether the problem is fixable with configuration or hardware and, only if needed, uses resize_vps to grow the server - always confirming the new cost with you. Disk can only grow, never shrink, so the best practice the channel enforces is to start small and scale when it's time.

4. A multi-server environment with a strict firewall (advanced)

I need a staging environment with two Ginernet VPS: one running the
API (Node.js, port 443 open to the public) and another with
PostgreSQL reachable only from the first one's IP. Generate new SSH
keys for each, document IPs, credentials and the environment's total
hourly cost, and shut both servers down when you finish configuring
them: I'll start them myself when the test begins.

This prompt combines almost the whole channel: two creations, SSH deployment, selective firewall rules (the database is never exposed to the internet), power management with vps_power and a final cost report. Keep one nuance in mind: a stopped VPS keeps billing while it exists; to stop paying you must delete it, and only you can do that from the panel.

For developers: how it works under the hood

If you want to integrate the channel into your own tooling, these are the technical details:

• Protocol: MCP over JSON-RPC 2.0 with Streamable HTTP transport at POST https://manager.ginernet.com/mcp. Stateless: every request is independent, no sessions or cookies.
• Authentication: Authorization: Bearer gcp_... header. No OAuth: one revocable key per integration.
• Limits: 120 requests per minute per key.
• Discovery: tools/list returns the full JSON schema of every tool; initialize includes the channel's operating instructions (cost, firewall and SSH best practices that agents follow automatically).

You can try it with curl in a minute:

curl -s https://manager.ginernet.com/mcp \
  -H "Authorization: Bearer gcp_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Long operations (creation, resize) are asynchronous: they return a taskId you poll with get_task until the status is succeeded. Typical VPS provisioning takes between 45 seconds and 3 minutes.

Frequently asked questions

Do I need to know how to code to use this? No. Installing an agent like Claude Desktop is like installing any application, and the panel gives you the connection snippet ready to copy. From there, everything is a conversation in your language.

Which agents and editors are compatible? Any MCP client: Claude Code, Claude Desktop, Cursor, VS Code (Copilot), Windsurf and Zed each have their own tab with instructions in the panel, and the list keeps growing because MCP is an open standard.

Is it safe to give an AI access to my hosting? The channel is limited by design: no deletion, spending caps only you control, per-key permissions, firewall closed by default and an audit trail for every action. The agent can create and operate; irreversible decisions remain yours.

How much does it cost? There is no fee for the MCP channel itself: you only pay for the servers, hourly and from prepaid balance. The smallest VPS starts at €5/month (~€0.007/hour) with the IPv4 included, and the agent always confirms the exact cost before creating or growing anything.

Can the AI delete my servers or drain my balance? It cannot delete anything: deletion is exclusive to the web panel. And spending is triple-capped: an API server limit, an hourly spend limit and the prepaid balance itself.

What if the agent makes a mistake? Possible mistakes are low-impact by design: an extra server gets shut down and deleted from the panel in one click, and a wrongly opened port is closed with the same tool. Everything is logged in the activity history so you can review what the agent did and when.

Ready to try it? Enable the channel in Manager → AI Agents, connect your agent and ask for your first server. If you have any questions, our support team is one ticket away.