Most WordPress AI tools live in a chat box, where you describe seeing a white screen on your site, the bot suggests five things that might be wrong, and you go back to your error logs to do the work yourself.
CortexWP is trying something different. It’s a WordPress engineering agent that connects to a real WordPress install, reads the actual files and logs, and executes approved actions directly on the site.
What CortexWP Actually Does
CortexWP bills itself as a WordPress engineering agent rather than just some support chatbot.
You connect a site by SFTP or through their connector plugin. From there, the agent inspects what a senior engineer would check: files, logs, database context, DNS records, cache layers, PHP configuration, and connected hosting details. It works from the actual WordPress install itself rather than guessing from a symptom you typed into a box.
Once it’s connected, it can take action. CortexWP can activate, deactivate, delete, update, and audit plugins, themes, pages, posts, custom post types, and the WordPress config. It can write PHP snippets, patch CSS, toggle maintenance mode, set up Redis, and make edits to the config file.
Where it really gets useful is the debugging. It targets fatal errors, white screens, broken updates, PHP notices, cron failures, and plugin conflicts, using evidence from the site rather than generic steps. If you’ve ever spent an afternoon isolating which one of your site’s 30 plugins broke a checkout page, like I have a couple of times, you already know how much time that saves.
How CortexWP Handles Approval Before Acting
The obvious worry with an AI that has write access to production is that it just does something that isn’t right, and you don’t have the time to intervene and stop it.
CortexWP’s answer to that problem is an approval gate. Every real action asks you for approval first, so before it changes anything, the agent shows what it will run, why, the likely impact, and any rollback notes. You approve the run, it executes it, and then you get evidence of what changed, what was tested, and what still needs human judgment.
It’s the right approach to have that human in the loop throughout the process, but whether the process holds up in the real world or not is something to test out. That’s why it’s always recommended to try any new approach like this on a staging site before trusting it with a live site.
How CortexWP Pricing and Tokens Work
CortexWP charges with token credits rather than a flat per-seat fee.
In their model, tokens are usage credits for the whole task. They cover the model reasoning, tool calls, WordPress inspection, log reading, commands, code changes, and response text. A deep debug session burns more than a quick question, obviously, because the agent is doing more work. That means your cost scales with how hard your problems are, not how many you have.
The plans on offer are:
- Starter: $5/month for 1M tokens, for single-site owners wanting help with small fixes.
- Pro: $15/month for 5M tokens, for agencies and production sites needing recurring debugging.
- Ultra: $50/month for 20M tokens, for heavier incidents and multi-site work.
New accounts also get 100,000 free tokens with no credit card required, which enough to run the agent through a real problem on a staging site before you commit to the paid plans.
The token-per-task pricing is the thing to watch out for. Although it’s transparent, it makes budgeting your tokens harder than a fixed subscription. A single brutal incident could eat a chunk of your allocation, and agencies billing clients hourly will want to map token spend against their own rates to get the most out of their plan.
What CortexWP Cannot Do Yet
CortexWP is upfront that it’s backend-focused for now. The current build handles fatal error triage, plugin conflict isolation, maintenance mode, WP-CLI style operations, config edits, Redis enablement, PHP snippets, CSS fixes, and backend feature work.
What it cannot yet do is see the front end, but visual site understanding and page design support are listed as coming soon, as is Cloudflare integration for DNS, cache, and edge troubleshooting. So if your problem is a layout that looks wrong rather than a site that is broken at its core, you’ll need to wait a bit longer before CortexWP can come to your rescue.
Is CortexWP Worth Trying?
WordPress debugging is tedious, evidence-heavy work, and most of it follows patterns an agent could plausibly handle faster than a human clicking through logs. We’ve done this internally with some custom solutions so I can confirm that the idea works. An AI that reads the actual install and proposes changes with a rollback note is definitely going to be more useful than a chatbot guessing from symptoms.
As you try it out, keep a few questions in mind to keep asking yourself. How well does the approval gate describe risk before you click yes? How predictable is the token spend across a month of real incidents? And how does it behave on a site with the kind of plugin sprawl that causes these problems in the first place?
The real answers to these questions will only come from connecting a staging site and running it through a problem you already understand, which is what the 100,000 free tokens are for. If you try it, I’m curious to read in the comments below where the approval model held up for you and where it didn’t.
