Publish a site from Cursor's agent, no terminal command needed
17 September 2026 · 3 min read
Cursor already writes real files to your machine when its agent builds a site, so publishing doesn't have to mean copying a folder into a hosting dashboard or running a deploy command. Connect MakeMySiteLive's hosted MCP server once, and the same agent that wrote the site can publish it too, straight from the chat you're already in. Here's how that path works, and when a script is the better fit.
Two ways to publish from Cursor
Cursor gives you two paths to a live URL. You can zip the folder the agent wrote and upload it in the MakeMySiteLive dashboard. Or you can connect the MCP server and skip that step. You ask Cursor's agent to publish, in plain words, and it runs the deploy for you, in the same chat where it wrote the code.
Connect the MCP server once
Setup lives under Customize → MCPs in Cursor (it moved out of Settings), and it's a one-time step. Add an entry for makemysitelive under mcpServers pointing at MakeMySiteLive's connector URL, save the file, then authenticate when Cursor flags the new server as needing attention. That's a normal OAuth sign-in to your own MakeMySiteLive account, and the agent only ever acts as you, within your plan.
The full click-by-click walkthrough, with screenshots of the actual Customize → MCPs flow, is in the setup guide.
Publish by asking, not typing a command
Once it's connected, there's no deploy command to remember. Ask Cursor to “publish this folder as a live site” or “create a site called my-demo and deploy this build,” and the agent checks the subdomain, runs the deploy, and hands back the live yourname.makemysitelive.com URL in the same reply. The chat where the site got built is also where it goes live.
What the agent can do beyond the first deploy
The same connection covers more than the initial publish. Ask it to list your sites, roll back to a previous version after a bad deploy, check whether a subdomain is free before you commit to a name, or turn on password protection for a Pro site, all without opening the dashboard or a terminal.
When a script is the better fit
MCP is built for a chat session with a person in it. A CI pipeline, a scheduled job, or any deploy that runs with nobody there to approve each step is a job for the REST API instead. It uses an API key, kept as a secret environment variable, in place of a sign-in. Use MCP inside the editor, and the REST API anywhere a script runs on its own.
Key takeaways
- Cursor's MCP connector lets the agent publish a site it just built, without a terminal command.
- Setup is one-time: add the connector under Customize → MCPs, authenticate, done.
- The agent can also list sites, roll back versions, check subdomain availability, and toggle password protection.
- For CI and other unattended deploys, use the REST API. MCP is built for chat sessions with a person in them.
Frequently asked questions
Do I need to install anything to use the MCP connector?
No. The MCP connector is one URL added to Cursor's mcp.json. There is nothing to install.
Can the agent overwrite my site by mistake?
Each deploy still goes through MakeMySiteLive's own versioning, so a full-site publish through MCP creates a new immutable version and can be rolled back the same way any other deploy can. You can also set MCP tools to require approval before running.
Does this work the same way in Claude and ChatGPT?
Yes. All three connect to the same MCP server and connector URL. Cursor puts the setting under Customize → MCPs, Claude under Customize → Connectors, and ChatGPT under Settings → Plugins; the underlying tools are identical.
What if I want to publish without giving the agent write access?
Set the write tools (create, deploy, rollback, delete) to require approval when you connect the server, so nothing publishes without you confirming it first, while read actions like listing sites still work automatically.