ProStack
8 June 2026
Part 1 in our MCP server series.
We’re always looking for ways to make our web hosting platform Cito a genuinely great piece of kit for developers. Our latest update? Building an MCP server that lets you manage your Cito sites through the AI assistants you’re already using – think Claude, Cursor, and ChatGPT.
Here’s what an MCP server actually does, why we built one, and what you can do with yours.
What’s an MCP and why does it matter?
The Model Context Protocol (MCP) is a relatively new open standard that gives AI assistants a structured way to talk to external systems.
If you’ve ever installed a WordPress plugin, you already know roughly how it works. MCP lets a developer extend what an AI can do, without the AI itself needing to ship with that capability built in.
An MCP server declares a set of tools, the AI assistant connects and calls upon them when the conversation calls for it.
Why did we build an MCP server?
Lots of our customers are digital agencies, with development teams who are already using AI tools to make life simpler. For instance, when it comes to writing and debugging the code that ends up on our Cito servers or figuring out why something broke at 11pm on a Friday (we’ve all been there).
That’s what got us thinking. Why make those developers tab out to a dashboard or remember the syntax of a CLI command, when their AI assistant could just do the thing?
That’s when we decided to build a remote MCP server for Cito. It lets an AI assistant create sites, manage databases, read logs, deploy projects, and a handful of other things, all with natural-language prompts like:
- “Show me the error logs for my staging website”
- “Create a new WordPress site”
- “Clone example.com to staging.example.com”
- “Show me the last 25 lines of the error log for example.com”
What does it mean for you?
We’ve built the MCP server to help you out with two key workflows: deployments and debugging.
Here’s how they work in a nutshell, but you can find more information (including what the MCP won’t do) in our Cito documentation.
Deploying new projects to Cito servers
With the MCP server, you can say “deploy this into Cito” and have a working URL just a few seconds later. The MCP will:
- Detect the type of site you’re deploying (Laravel, WordPress, etc)
- Pick the correct deployment profile (or create a site if one doesn’t exist)
- Rsync files to your site of choice with an ephemeral SSH keypair
- Handle any post-install commands automatically
Debugging sites
Who doesn’t want to make debugging easier? Type in something like “why is my site broken?” and you’ll prompt the MCP to:
- Fetch the relevant error logs
- Interpret them and pinpoint what’s gone wrong
- Propose a fix and write the code
- Push the new code up
There are certain things the MCP server won’t do, and we’ve put those guardrails in for a reason. AI’s a useful tool, but it pays to be sensible: the last thing we want is for the MCP to cause you more issues than it solves.
We talk about the specific guardrails we’ve put in place in part 2 of this series: How we build the Cito MCP server – a deep dive. If you’re interested in the architecture the server sits on, the specific deployment workflows we’ve built, and the ways we’re keeping it secure, you’ll want to read this blog next.
Give it a go yourself
If you’re a Cito customer, the MCP server is yours to play with. You can connect it to your AI tool of choice (Claude Code, Claude Desktop, Cursor, or ChatGPT) in about 30 seconds right here.
The MCP server is currently in beta, so we’d also really appreciate your feedback. You can share your thoughts here – we’re all ears.
