I spent 6 hours trying to connect Claude to my WordPress site. So I built the plugin that should have existed.

I had a simple goal. I wanted to run a content refresh across 400 published articles on a WordPress site built with Elementor. I wanted to do it from Claude Desktop, the AI tool I already use every day, with the audit data from my own crawler feeding into the same conversation. Read a post, see what’s broken, fix it, move on.

This should be a solved problem in 2026. It is not.

After six hours of trying every plausible option, I gave up on the existing landscape and built IATO MCP, a free, AI-agnostic WordPress plugin that connects your site to whatever AI tool you already use. WordPress and Elementor native, with more page builders coming. This post is the story of why it had to exist, and what each of the alternatives got almost-right.

The workflow I was actually trying to do

The job was content refresh, the kind of work agencies do constantly:

• Pull a post
• Look at what the audit data says about it (thin content, broken links, missing alt text, weak title, orphan page, whatever)
• Update the content, the SEO meta, the alt tags
• Move to the next one

400 times.

If the site had been plain Gutenberg WordPress, almost any AI plugin could have handled it. But this was Elementor. And Elementor doesn’t store content as HTML in post_content like a normal WordPress post. It stores it as a serialized JSON blob in a meta field called _elementor_data. Touch that JSON wrong and the page breaks. Update it without clearing Elementor’s CSS cache and the frontend doesn’t reflect your changes. Skip the content regeneration step and your search results, RSS feed, and any non-Elementor reader of the post still show the old version.

So I needed an MCP server inside WordPress that understood Elementor natively. I tried three.

Plugin 1: msrbuilds/elementor-mcp

This one looked perfect on paper. Built specifically for Elementor. Open source. Active development.

The problem was distribution. It lives on GitHub, not the WordPress.org plugin directory, and it depends on a second plugin called MCP Adapter that also lives on GitHub. Both plugins are distributed as source; they expect you to run composer install after cloning to pull down their PHP dependencies into a vendor/ folder before the plugin will function.

I installed both plugins through the WP Admin uploader. They activated cleanly. No errors.

And then nothing. No settings page appeared. No REST endpoints registered. Silent failure.

The reason: missing Composer autoloader. Without the vendor/ folder, the plugin’s classes can’t load, and there’s no graceful failure mode that tells you that’s why. The fix would have been to:

1. Install Composer on a local machine
2. Clone the repo
3. Run composer install
4. Zip the plugin with the vendor/ folder included
5. Upload the zip through WP Admin
6. Repeat that entire dance every time the plugin updates

If you have SSH access to your server, this is annoying but doable. I had WP Admin only, which is the situation most agencies are in for most of their client sites. There was no clean path to production.

This is a real pattern, by the way. A lot of “MCP for WordPress” projects on GitHub assume you’re a developer with full server access. That assumption excludes the entire WordPress audience these tools claim to serve.

Plugin 2: AI Engine by Meow Apps

AI Engine is a mature, heavily-installed plugin on WordPress.org with an MCP server feature in its free tier. After the msrbuilds saga, the install experience felt like sunshine. One click, no Composer, no dependencies. The MCP endpoint worked on the first try.

But the tools it exposed were generic WordPress post CRUD: get post, update post, create post. Which is fine, for a normal WordPress site.

For Elementor, it wasn’t enough. To update an Elementor page through generic CRUD tools, I would have had to:

• Read the post meta manually to get the _elementor_data JSON
• Parse it, modify it, re-serialize it correctly
• Write it back, hoping the JSON escaping survived the round trip
• Manually invalidate _elementor_css so the frontend reflected changes
• Regenerate the rendered post_content so search and RSS stayed in sync

For one post, an afternoon’s worth of careful prompting. For 400 posts, an unacceptable amount of surface area for things to break silently. Elementor breakage usually only shows up when a real visitor lands on the page.

AI Engine is a good plugin. It just wasn’t built for this.

Plugin 3: IATO MCP (the one I ended up building)

Six hours in, I had two options. Spend another day duct-taping AI Engine into something that could safely handle Elementor JSON, or build the plugin I actually wanted.

I built it.

IATO MCP installs directly from the WordPress.org plugin directory. No Composer. No dependencies. No vendored PHP libraries to build. On activation the plugin generates an Application Password, surfaces a setup wizard that produces ready-to-paste config for Claude Desktop and other MCP clients, and exposes its endpoint at a predictable URL.

The toolset is purpose-built for content audit-and-fix workflows. 52 tools split into two categories:

The first 40 are WordPress native tools that work immediately with no account required: get and update posts and pages, read and write Elementor JSON safely with dry-run preview and automatic CSS cache clearing, update SEO meta through Yoast, RankMath, or SEOPress (auto-detected), manage alt text, menus, taxonomies, redirects, structured data, and canonical URLs. Every write returns a change receipt so any AI-made change can be rolled back with a single call.

The second group is 12 IATO bridge tools that connect your WordPress site to your IATO crawl data: SEO audit results, broken link reports, orphan page detection, content gap analysis, AI-prioritized fix suggestions, and crawl control. These require an IATO API key entered in the plugin settings.

That last part was the unlock for me. One MCP server, one Claude conversation, both the audit step and the fix step. Instead of running two separate tools and copy-pasting between them, the AI can look at the audit findings for a post and update the post in the same turn.

The comparison, honestly