Let your AI agent run your whole infrastructure

Everything you do by hand in ArduMaker, your agent does for you: with your permissions and with every action on the record.

MCP endpoint: https://mcp.ardumaker.com/mcp · OAuth 2.1 + PKCE with dynamic client registration · Full tool reference

How do you connect?

Three steps, and the agent handles two of them

You give your agent the MCP server address — it registers itself
The agent opens an authorization link in your browser
You pick the scope and approve. Done: it works on your behalf

Copy, paste, done

Copy this message and paste it to your AI agent — it connects on its own and will only walk you through the part you have to do (approving access in your browser).

Connect to the ArduMaker MCP server at https://mcp.ardumaker.com/mcp (Model Context Protocol, OAuth 2.1). Add it as a remote MCP server and walk me step by step only through what I need to do myself, such as approving access in the browser.

Automatic discovery via OAuth metadata
Dynamic Client Registration (DCR)
PKCE — no secrets to leak
The session renews itself: connect once

Manual configuration (Claude Desktop, Cursor, etc.)

{
  "mcpServers": {
    "ardumaker": {
      "url": "https://mcp.ardumaker.com/mcp"
    }
  }
}

Connect your agent

Two steps per client. No API keys: you approve access in the browser and you're done.

Claude: Claude → Settings → Connectors → «Add custom connector». Paste the server URL and approve access in the browser.
```
https://mcp.ardumaker.com/mcp
```
ChatGPT: ChatGPT → Settings → Apps & connectors → enable developer mode. Create a connector with the server URL and authorize it.
```
https://mcp.ardumaker.com/mcp
```
Gemini CLI: In Gemini CLI run the command below. Approve access when the browser opens.
```
gemini mcp add --transport http ardumaker https://mcp.ardumaker.com/mcp
```
Cursor: Save this block in your project's .cursor/mcp.json. Approve access when Cursor opens the browser.
```
// .cursor/mcp.json
{ "mcpServers": { "ardumaker": { "url": "https://mcp.ardumaker.com/mcp" } } }
```
VS Code / Copilot: In VS Code: «MCP: Add Server» → HTTP, or save this block in .vscode/mcp.json. Approve access when the browser opens.
```
// .vscode/mcp.json
{ "servers": { "ardumaker": { "type": "http", "url": "https://mcp.ardumaker.com/mcp" } } }
```
MCP: Any MCP client: point it at the server URL (HTTP, OAuth 2.1). Approve access when the browser opens.
```
https://mcp.ardumaker.com/mcp
```

What your agent can do

VPS and commands: Create and control servers, run commands, read logs and metrics, manage Linux users
Domains, DNS and nginx: Your own domains and their DNS records, nginx configuration, reverse proxies and port forwards
Email: Mail domains, mailboxes, aliases, filters and folders — plus reading and sending messages
Databases: PostgreSQL and MySQL: explore the schema, query, insert, edit rows and run DDL
Projects and team: Stages, tasks, checklists, calendar, meetings, attachments and audit trail
Services and scheduled tasks: systemd services and cron jobs: create, edit, start, stop and read their logs
GitHub deploys: Link repos, redeploy, and read the commits and logs of every deployment
WordPress: Sites, Elementor, WooCommerce, plugins, SEO, backups, staging and moves, on your own VPS
AI and search keys: Router IA keys and Search API tokens: create them, revoke them and check usage
Blog and content: Write, translate and publish blog posts, with their sections and images
SSH tunnels: Tunnel receivers and initiators, keys, and the state of every connection
Workspace and usage: Contracted services, usage for the period, and the workspace's verification status
Issue reporting: The agent can report a problem when something breaks, and look up its reports

What happens to my information?

It's the first question anyone asks before connecting an agent to their infrastructure. Here are the specifics.

You authorize, not the agent: An agent cannot grant itself access. Registering with the server gives it nothing: a person has to approve the connection from their account, in the browser.
You choose the scope when you approve: You can give it all your projects or a selection, and all your VPS or just some. They are two independent decisions, and you can narrow them later: scope is evaluated on every request, not frozen into the token.
Every action is on the record: The agent acts as one more member of the team, and its activity shows up in the project history under its own name, just like a person's. Nothing it does is invisible.
You cut access whenever you want: From your account you see every connected agent, when it was last used, and how long its session is good for. Revoke one or all at once and it loses access instantly — with no way to renew itself back in.

How long access lasts

The session renews itself while you use the agent, so you connect once and forget about it. It closes after 30 days unused, and expires for good after 180 days, when you authorize again. If someone tried to replay an old credential, the session cuts itself off.

What your agent will never be able to do

The agent inherits your permissions, but with explicit limits it cannot work around.

Delete an entire project
Add or remove members, or change anyone's role
Create, edit or delete a note or attachment marked as secret
Delete a VPS, or a Linux user inside one

See the full list before you grant access

The reference for every tool, with its parameters and what it returns, generated live from the server itself. No surprises.

Read the documentation

Frequently asked questions

What is the ArduMaker MCP server?

Public MCP server: connect Claude, ChatGPT or Cursor with OAuth and let it manage your VPS, domains, email and projects, with every action logged.

How do I connect my AI agent to ArduMaker?

1. You give your agent the MCP server address — it registers itself 2. The agent opens an authorization link in your browser 3. You pick the scope and approve. Done: it works on your behalf

What can a connected agent do?

Practically everything you do yourself, under the same rules and permissions: VPS and commands, Domains, DNS and nginx, Email, Databases, Projects and team, Services and scheduled tasks, GitHub deploys, WordPress, AI and search keys, Blog and content, SSH tunnels, Workspace and usage, Issue reporting.

What can an MCP agent never do?

Delete an entire project. Add or remove members, or change anyone's role. Create, edit or delete a note or attachment marked as secret. Delete a VPS, or a Linux user inside one.

How long does an agent's access last?

Connect your agent and start delegating

Create your account, authorise your agent and ask it for the first task. MCP is included at no extra cost.

How it works under the hood

The technical documentation for what this page covers: what each tool does, what changes, how to verify it and which MCP tools operate it.

MCP server: MCP server with OAuth and human approval so AI agents can operate ArduMaker, with profiles, safe retries and limits.
MCP agent permissions: What an MCP agent can do: its human's permissions, scope by project and VPS, read-only mode and operations that are always forbidden.
Agent audit: Log of every MCP agent call with its result, ongoing operations and command history, without storing values.

All the documentation