Skip to content

AI readyAgent ready

Manage your hosting from your AI agent.

Corefinity's MCP server gives your AI agents over 400 tools to be able to manage your infrastructure, cluster and sites. Your agent can read everything and change only what you allow.

Order online

Billed monthly. No setup fee. 60-day money-back guarantee.

Real panels, right in the chat.

Some answers come as a small piece of the control panel, inside the conversation.

Start a fresh WordPress site for saltbrae.co.uk on Brindlewyck UK.

The version picker: WordPress chosen, three version cards with their status, PHP and MariaDB, and a Tell the agent button.

Connect your client, then sign in once.

Your browser opens Corefinity's approval page; you choose what the app may do and go straight back.

Claude Desktop

  1. Open Customize › Connectors, then choose Add › Custom › Web.
  2. Name it Corefinity, paste the server URL, choose Continue and sign in.
https://manage.corefinity.com/api/mcp

Cursor

  1. Add to ~/.cursor/mcp.json:
    {"mcpServers":{"corefinity":{"url":"https://manage.corefinity.com/api/mcp"}}}
  2. In Cursor Settings › Tools & MCP, choose corefinity and sign in.

VS Code

  1. Open your user mcp.json (Command Palette: "MCP: Open User Configuration") and add:
    {"servers":{"corefinity":{"type":"http","url":"https://manage.corefinity.com/api/mcp"}}}
  2. Start the server from the editor; VS Code opens the sign-in.

ChatGPT and Claude.ai

Paste the server URL as a custom connector or app. OAuth does the rest; no token to copy.

https://manage.corefinity.com/api/mcp
Claude Code and Codex, in a terminal

Claude Code

  1. Run:
    claude mcp add --transport http --scope user corefinity https://manage.corefinity.com/api/mcp
  2. In Claude Code, run /mcp, choose corefinity, then Authenticate.

Codex (CLI or ChatGPT desktop app)

  1. Run:
    codex mcp add corefinity --url https://manage.corefinity.com/api/mcp
  2. Then sign in:
    codex mcp login corefinity

Connecting needs a Corefinity account with two-factor sign-in. New here? Order online, then connect. Prefer a token? Create one on the panel's MCP & API card with the scopes and sites you choose.

400+tools across every part of your hosting.

Read is always on. The rest are switches you choose, per token.

  • Runtime & configuration
  • Deployments & pipelines
  • Clusters & sites
  • Environments & copies
  • Logs & usage
  • Backups & restores
  • Access & firewall
  • Domains & integrations
  • Billing & team
  • MCP Apps

In your words, in your vocabulary.

Click a prompt to copy it.

The boundary is in the infrastructure, not in the prompt.

Four scopes: read, write, exec, destructive

Chosen per token or app. Read sees your clusters, sites, logs and usage and changes nothing.

Destructive is off by default

Never pre-ticked. Deleting, or overwriting production, needs the scope and a typed confirmation that matches exactly.

Two-factor to connect

A fresh code for every new token or app approval. A missing scope sends you a link to your own panel; nothing widens without you.

Safety snapshots and dry runs

Restores and clones snapshot first. Quotes, capacity checks, validations and pipeline rehearsals come before any change.

Where it works

Pin a token to clusters, sites or environments, allow-list addresses, set an expiry, revoke in one step.

Every action logged

Each call is written before it runs and shown in Activity by name, with arguments, result and reason. Kept forever.

Your AI agent never sees your passwords: secrets open in your browser by default.

Admin logins, database credentials and SFTP passwords open on a Corefinity page in your browser, never in the chat. Your agent never pays, verifies your email or passes two-factor for you.

Questions, answered.

Is it safe to let an agent touch my hosting?

Your agent gets exactly the permissions you tick. Destructive is never pre-ticked, connecting needs two-factor sign-in, destructive actions take a typed confirmation and a safety snapshot, and every call is logged by name before it runs. If a call needs more than the agent has, you get a link to your own panel; nothing widens without a code and a save.

Which clients work?

Claude Code, Codex, Claude Desktop, ChatGPT, Cursor and VS Code. Interactive views (charts, logs, smoke tests) render inside Claude Desktop, ChatGPT desktop, Cursor and VS Code; the terminal clients, Claude Code and the Codex CLI, show the same results as text.

Can my agent sign up and pay for me?

No, by design. It can price a cluster, check capacity, and open a ready-filled order page. You create the account, verify your email, set up two-factor and pay in your browser.

Does my agent get my passwords?

Not unless you say so. Admin logins, database credentials and passwords open on a Corefinity page in your browser. An agent receives a secret only when its token has exec and the call asks for it explicitly.

Can I see what an agent did?

Yes. Every call appears on the MCP activity page with its arguments (redacted), result and reason, and a site's Activity names the agent next to the people. There is an Agent filter.

Does it cost extra?

No. The MCP server and the REST API come with every CoreStack and CoreCluster. What your agent orders is billed exactly as if you had ordered it in the panel.

Let's Talk

Tell us about your platform and goals. Our engineers will help you find the right setup: no sales pressure, just expert advice.

Rather start on your own? Order online

By sending this message you agree to our Privacy Policy.

Pictured: in Claude Desktop, an AI agent is asked to add staging to saltbrae.co.uk, and Corefinity’s app in the chat shows the new staging environment ready next to production.

Manage your hosting from your AI agent.