If you run a WordPress site and you have wished out loud that you could ask Claude or ChatGPT to “back up my site,” “create a staging environment,” “check my DNS zone,” or “edit this content for me” and have it actually happen, there is now a practical answer. It involves an acronym you have probably seen this year (MCP, the Model Context Protocol), and it runs through the part of your stack you may not have thought about yet: your hosting provider.
This is a guide to the simplest path from a WordPress site to a working AI integration. It uses tools you are already paying for, takes about three clicks once you know what to click, and once you have it set up, the way you work with your site changes meaningfully.
Why MCP servers are the right primitive
A quick definition. MCP stands for Model Context Protocol, the open standard that AI clients like Claude and ChatGPT use to talk to external tools and data. An MCP server is a small endpoint that exposes a set of capabilities (read this, write that, run this action) in a format the AI can use directly. Once an AI client is connected to an MCP server, those capabilities show up in the conversation as something the AI can actually do.
For WordPress specifically, that matters because everything in a WordPress site (posts, pages, media, users, settings, plugins) is already structured. What MCP adds is the layer that lets an AI read and act on that structure without you having to copy and paste anything.
Several paths exist for getting MCP working with WordPress. You can write your own MCP server. You can install one of the growing number of WordPress plugins that ship an MCP endpoint. Or you can use the MCP server your hosting provider gives you. The third option is the simplest by a wide margin, and for most users it is the right starting point.
An MCP server set up by your host has one big structural advantage over a plugin-based one. It exposes both the WordPress site and the hosting infrastructure underneath. That means you can ask the AI to do things at the WordPress level (edit a post, manage media, change settings) and things at the hosting level (enable maintenance mode, trigger a backup, create a test site, modify DNS), all from a single connection.
The principle: your host already has the best MCP server
A growing number of WordPress hosting providers now expose an MCP server directly from their control panel. The pattern is consistent: log into your hosting account, find the MCP server settings for one of your sites, activate the server, and copy the URL. That URL is what you give to ChatGPT or Claude.
What gets exposed through that URL is more comprehensive than most users expect. Tools that interact with your WordPress installation (managing content, users, plugins, settings) sit alongside tools that interact with your hosting environment (backups, deployments, DNS, staging). With one connection, the AI on the other side gets both layers.
For the rest of this guide, the workflow assumes your host offers this feature. If yours does not yet, it is worth checking. The major WordPress-focused hosts are rolling this out steadily, and the feature is becoming a meaningful differentiator. If your host has not announced an MCP server, the plugin path is the next-best fallback.
Connecting to ChatGPT
ChatGPT supports MCP connections through its developer-mode applications system. The setup is a one-time configuration.
- Open Settings. In ChatGPT, go to Settings and open Applications.
- Enable developer mode. Inside Applications, open Advanced Settings and turn on developer mode. ChatGPT will show a warning when you enable it; you can read and accept it.
- Create a new application. With developer mode active, a Create Application button appears. Click it.
- Paste the MCP server URL. In the form, paste the URL from your hosting panel. Give the application a name (your site name works well; it makes the connector easy to address in conversation if you connect more than one site). Add an icon and a short description if you want.
- Set authentication to “none.” This sounds wrong but it is correct. The URL itself already contains a unique token that authenticates the connection. Selecting “no authentication” means you are not layering anything on top of the token already in the URL. The token does the work.
- Confirm and accept. Click Create, then accept the additional security pop-up that follows. The site is now connected.
From the active applications view, scroll through the full list of tools the MCP server has exposed. You will see a mix of site-level capabilities (managing content, users, plugins) and hosting-level ones (backups, staging, DNS). The combination is the point.
Connecting to Claude
Claude’s setup is more direct. In your Claude account, open Settings and go to Connectors. Click the plus button to add a custom connector. Give it a name (again, the site name is the natural choice), paste the same MCP server URL you used for ChatGPT, and click Add.
Claude immediately lists the tools the connector exposes. A small productivity tip worth mentioning: any read-only tool can be marked “always allowed” in the connector settings. Tools that only read data, never modify it, are safe to pre-authorize, and doing so saves you a confirmation step every time the AI checks something.
The first time you use the connector in a Claude conversation, you may need to nudge it toward the right site if you have multiple connectors active. Asking explicitly (“using the [site name] connector, list every tool available for the full MCP server, not just the WordPress site”) is the clearest way to confirm what Claude actually has access to.
When asking the AI to enumerate available tools, frame the question carefully. The natural phrasing “what WordPress tools are available” often returns only the in-site tools and hides the hosting-level capabilities. Asking about “the full MCP server” or “every tool this connector exposes” surfaces the complete list, which can run to dozens of tools.
The three-layer architecture: site, cloud, account
The simplest setup connects one WordPress site to one AI client. Some hosts have taken the model further, exposing MCP servers at three nested levels of granularity.
A site-level MCP server exposes a single WordPress installation, plus the hosting controls associated with that site (its backups, its staging environment, its DNS, and so on). This is the right starting point and the level most users will spend their time in.
A cloud-level MCP server (the “cloud” here being the host’s term for the container or plan that holds one or more of your sites) exposes everything in that container. One MCP connection now reveals every site inside the cloud, plus all the cloud-level controls (scaling resources, adjusting the plan, managing the shared environment). This is useful when you manage multiple related sites and want one AI conversation to be able to move between them.
An account-level MCP server exposes everything in your hosting account. Every cloud, every site within every cloud, every site-level tool, and every account-wide control. This is the broadest possible attack surface, and that phrasing is deliberate. It is also the most powerful configuration if you genuinely manage a portfolio of sites and want the AI to coordinate across them.
The tradeoff is the obvious one. Broader connections give the AI more reach. They also give the AI more reach if something goes wrong. For most users, the site-level connection is the right default, and the higher-level connections are worth adding only when there is a specific reason to.
Combining with WordPress AI plugins
The MCP server your host provides exposes whatever is in your WordPress installation, plus the hosting layer. That means the tools available to the AI grow automatically whenever you install something inside WordPress that exposes additional AI capabilities.
In practice this gives you a clean two-step workflow. First, install and configure any WordPress plugin that exposes its capabilities as AI tools. A growing number of plugins are designed specifically to add AI-callable actions to WordPress: content generation hooks, structured query tools, custom workflow triggers. Second, refresh your MCP connection on the AI side. The new tools appear automatically in the list.
This is the right pattern for getting the most out of MCP without committing to a more complex custom setup. The plugin adds capability inside WordPress. The host’s MCP server exposes that capability. The AI client picks it up the next time it queries the connector.
What you can actually do with it
The use cases that matter most are the ones that previously required jumping between tools.
Content work becomes conversational. You can ask the AI to draft a post, save it as a draft directly in your site, then open it in the WordPress editor to refine. No copy-paste step.
Maintenance work becomes voice-driven. “Enable maintenance mode.” “Trigger a backup before I push this update.” “Spin up a staging environment from the current site, I want to test a plugin update.” These are instructions that work, not demos.
Diagnostic work becomes conversational. When something is wrong, you can ask the AI to read the relevant logs, summarize the error, and propose a fix. The AI has the same access to the site and the hosting environment that you do.
Multi-site management becomes coordinated. With a cloud-level or account-level MCP server, “find every site in my account where plugin X is out of date and schedule an update” stops being a script you have to write and becomes a conversation.
The deeper shift here is what an AI assistant actually does for a WordPress workflow. The chatbot pattern, where you ask questions and copy answers back into your site, evaporates. The replacement is something closer to a colleague who can both think and act, with you still in the loop for approval.
A note on safety
Three principles worth internalizing.
First, the MCP connection is exactly as powerful as your hosting account. Anything you can do, the AI can do once it is connected. That cuts both ways. Useful, yes. Worth keeping in mind, definitely.
Second, “always allow” should be reserved for read-only tools. Write actions (publishing, deleting, modifying settings, billing operations) should retain the confirmation step. The small friction is the safety mechanism.
Third, account-level connections deserve more caution than site-level ones. The convenience of one connection seeing your whole portfolio is also the risk of one bad instruction affecting your whole portfolio. Most users should start with site-level and graduate upward only when a specific need justifies it.
Frequently Asked Questions
What is MCP and why does it matter for WordPress?
MCP, the Model Context Protocol, is the open standard that lets AI clients like Claude and ChatGPT call external tools and data sources. For WordPress, an MCP server is the endpoint that exposes the site’s capabilities (and often the hosting environment behind it) to the AI in a format the AI can use directly. Once connected, the AI can act on your WordPress site, not just discuss it.
Do I need WordPress 7.0 to use MCP servers?
No. The MCP server provided by your hosting provider operates independently of any specific WordPress version. WordPress 7.0 added a native AI connector for plugins to share AI access at the WordPress level; the host-side MCP server exposes capabilities on the AI client side. The two complement each other, but neither requires the other.
Is “no authentication” actually safe?
In this context, yes. The MCP server URL contains a unique token that authenticates the connection. Selecting “no authentication” in the ChatGPT setup means no additional authentication layer is being added on top of the token already in the URL. Treat the URL itself as a secret, the way you would an API key.
What is the difference between site-level, cloud-level, and account-level MCP servers?
A site-level connection exposes one WordPress site and its associated hosting tools. A cloud-level connection exposes every site within a hosting plan or container, plus the controls for that plan. An account-level connection exposes everything in the account: every plan, every site, every tool. Each broader level gives the AI more reach, both for legitimate work and for mistakes. Most users should default to site-level.
What happens to the MCP server if I install a new plugin?
If the plugin exposes its capabilities as AI tools, they automatically appear in the MCP server’s tool list. The next time the AI client refreshes the connection, the new tools are available. This is the cleanest way to extend what the AI can do without building anything custom.
Can I connect the same MCP server to both ChatGPT and Claude?
Yes. The URL is the same; only the client-side setup differs. Many users keep both connected, defaulting to whichever client they prefer for a given task. Read-only lookups and quick questions often go to ChatGPT; longer, more structured workflows that benefit from Claude’s writing and code review tend to go to Claude. The MCP server itself does not care which client is calling it.
