Most AI “content generators” fail for one reason: they have no idea what is already on your site. They guess from a generic training set. The Model Context Protocol (MCP) for WordPress.com changes that. Bridge your site’s real data to Claude and the WordPress Claude content ideas you get back are responses to what your audience is actually doing.
Fourteen years in, I have watched a lot of people throw spaghetti at the wall with content strategy. It usually involves a messy spreadsheet and a lot of guessing. Treat the site as a live data source over MCP instead, and the model can read your posts, your pages and your comments, which is where you find out how you are failing your users.
Why MCP is the bridge that was missing
Getting Claude to look at your site used to mean pasting URLs or uploading a CSV of your post list, which tested my patience more than it tested the model. MCP is an open standard that lets AI assistants discover and call external tools, so your WordPress site becomes a set of functions Claude can invoke. I went into the technical side of that in an earlier post on why we’re moving toward MCP for plugin development.
Step 1: turning on the connection
Start with the handshake. In your WordPress.com account, go to AI and MCP and switch on “Enable MCP Tool Access.” That creates the endpoint Claude connects to. It runs on OAuth 2.1, so Claude can read your data without leaving a wide-open door into your database.
Then open your Claude settings, find the connectors list and add the official WordPress.com tool. Once you authorize it, Claude gets read-only access to your site’s REST API through the MCP server, plus the occasional ability to write a draft.
Finding the content gaps that matter
Now do not ask it for “blog post ideas.” Ask for a gap analysis. A prompt worth running tells Claude to scan your /posts endpoint, read the /comments for technical questions nobody ever answered, and hold both against your competitors’ sitemaps.
Something like: “Review my existing blog posts and identify recurring themes in user comments that I haven’t addressed in a dedicated article.” That is where the WordPress Claude content ideas with real buying intent come from.
From idea list to content calendar
Some ideas sit on a real technical bottleneck and some are top-of-funnel noise, so they should not carry equal weight. I tell clients to score them. Have Claude weight each new idea against two criteria:
- Prospect value: Does it answer a “how-to” or “pricing” question that ends in a sale?
- Client retention: Does it help an existing user troubleshoot a legacy code problem they keep hitting?
Claude can push that scored list straight into a Google Sheet, which beats the manual entry we were doing back in the 4.x days. My guide on applying Claude Code to non-technical tasks covers more workflows in this shape.
Drafting and publishing
With a priority topic picked, Claude can outline it using the context of your existing internal links, which keeps you out of the duplicate content trap. When the draft is close enough, one prompt saves it to your site as post_status => 'draft'.
// Behind the scenes, the MCP server is essentially hitting endpoints like:
// POST /wp/v2/posts
// With a payload containing your markdown converted to Gutenberg blocks.If setting all of this up is eating your dev hours, I can take it off your hands. I’ve been wrestling with WordPress since the 4.x days.
What changes in practice
The Model Context Protocol changes how site management gets done rather than adding another toy to the pile. With Claude reading your own data, “I think we should write about this” becomes “the data says there is a gap here.” The official MCP documentation covers the protocol itself in detail.
