Claude Code is good at writing your app. Getting it onto a server usually means leaving the agent, opening a dashboard, and copying settings by hand. An MCP server removes that step: the agent gets tools to create apps, deploy them, and read logs, and you ask for the result in plain English.

This guide connects Claude Code to CloudPloy’s MCP server and deploys a repo. It works the same in Cursor, Codex, ChatGPT, OpenCode, and any other MCP client.

What you need

  • A CloudPloy account. The Free plan covers one small server and one app.
  • A Git repository on GitHub, GitLab, or Bitbucket.
  • A server to deploy to: one you already own, or an AWS account you connect with an access key.

Step 1: Add the MCP server

Run this once in your terminal:

claude mcp add --transport http cloudploy https://app.cloudploy.com/mcp

Your browser opens so you can sign in to CloudPloy and approve access. After that, Claude Code lists CloudPloy as a connected server.

Step 2: Connect your Git provider

Ask Claude Code:

Connect my GitHub account to CloudPloy.

The agent returns a sign-in link. Open it, approve access in the browser, and tell the agent you’re done. You never paste a token into the chat.

Step 3: Detect and create the app

Detect what kind of app is in my my-org/my-app repo and create it on CloudPloy.

The agent inspects the repo and picks a build. Node, Python, Go, Rust, Ruby, Java, PHP, and static sites build through Nixpacks. Laravel and WordPress have one-click presets.

Step 4: Check the cost, then deploy

Estimate the cost of a server for this app, then deploy it.

Compute is billed at your cloud provider’s rate with no markup from CloudPloy. The agent shows the estimate before it provisions anything. Once the server is ready, it deploys the app and reports the result.

Step 5: Redeploy on every push

Turn on auto-deploy for the main branch.

From then on, a push to main redeploys the app. If a release goes wrong, ask the agent to roll back to the previous deployment.

What the agent can and can’t do

The agent can create apps, trigger deployments, set environment variables, add domains, provision SSL, read logs, restart apps, and resize or stop servers.

Destructive actions are guarded. terminate_server and delete_app are permanent, so they run only when you ask for them and the call includes confirm=true.

Why use MCP instead of a dashboard

  • One place to work. You stay in the agent you already use to write code.
  • No pasted secrets. Git and cloud access are granted in the browser, not in chat.
  • Your infrastructure. Apps run on your own AWS account or VPS, so you keep the bill and the data.

Try it

Start on the Free plan, add the MCP server, and ask Claude Code to deploy a small app. The MCP setup page has copy-paste commands for every client, and the pricing page lists the paid plans.