
ChatGPT can edit your WordPress site — not by pasting text back to you, but by calling real functions on your server: creating posts, updating products, fixing SEO meta, reading orders. The mechanism is MCP, and connecting the two takes about five minutes once you know where the settings live.
This guide covers exactly that: what you need, where the connector settings are, how to connect ChatGPT to a WordPress site running an MCP server, the first prompts to try, and the limitations that are specific to ChatGPT rather than to WordPress.
A good first prompt, because it proves the whole chain works without changing anything:
Using the WordPress connector, list the five most recently modified pages
on the site with their titles, slugs and last-modified dates.
If that returns real pages, you are connected. If it returns a plausible-looking table of pages that do not exist, the connector is not being used — the model is guessing. Ask it explicitly to call the tool, and check the connector toggle.
Start with reads, then move to a small write with the approval gate on:
# read — safe, immediately useful
Which published posts have no meta description? Show a table with title, slug and word count.
# read — bulk insight
List every product with no featured image, grouped by category.
# write — small and reviewable
Draft a post from this brief: [brief]. House style, H2 structure, internal links to
our three most relevant existing guides, SEO meta set. Save as draft, don't publish.
More prompt patterns, with the full process each one triggers, live in the solutions library.
Being straight about these saves an afternoon of confusion:
If the connector refuses to connect at all, the error is almost always one of six things — they are all decoded in WordPress MCP not working.
The same rules apply as with any AI client, and they are worth stating because ChatGPT is often the first one a non-developer connects:
The full safety argument, including what an agent still cannot do: is it safe to give an AI agent write access to your WordPress site.
ChatGPT is the right choice if it is already your daily driver and your work is conversational — audits, content, one-off jobs, questions about the site. If you live in an editor, Cursor or VS Code keeps the site next to the code. If you script repeatable runs, a terminal client wins. They are not exclusive: the same site can serve all of them at once, because the server side is identical.
Side-by-side comparison: which AI clients work with WordPress MCP.
Yes, when your site runs an MCP server and you connect it as a custom connector. ChatGPT then calls typed tools on your site — creating posts, updating products, fixing meta — with the writes gated behind your approval.
Yes for custom connectors. They are available on paid tiers (Plus, Pro, Business, Enterprise); free accounts cannot add a custom MCP connector. In Business and Enterprise workspaces an admin may also need to allow it.
With the standard controls, yes: issue a scoped credential (start read-only), keep the approval gate on for writes, snapshot before bulk jobs, and check the audit log. The credential is per-connection and revocable.
Usually the connector is not enabled in that specific conversation, or the credential lacks scope. Toggle the connector on for the chat, then ask it explicitly to call the WordPress tool.
Not directly — remote connectors need a public HTTPS URL. Use a tunnel to expose your local site, or work against a staging site instead.
Yes, if your MCP server ships integrations for them. NibWP exposes 45+ plugin integrations, so WooCommerce products, Yoast meta, ACF fields and similar are callable as typed tools.