Okay so I'm a WordPress dev with 15 years of experience, and of course I've fully embraced the AI coding tools. In the beginning I worked with Cursor for a few months and then moved over to Claude Code. I've been coding with Claude Code for the last few months now and it's getting better every week.
At the start I just used it for coding, then I started using WP CLI commands to test and debug stuff. But over time I realized it can basically do anything for me. It can run bash scripts, PHP eval for quick small scripts, and it can create SQL queries 100x faster than I can (yes, I have 15 years of coding experience and my SQL still sucks, haha). I also realized it can set up plugins for me, activate them, be my strategic partner, analyze SEO content... I was doing all of this in my local (DDEV) environment.
But then it hit me. In all those 15 years, one of the most annoying things about WordPress development is that debugging production sites is slow and painful. Sure, you can copy your production files to local, copy the database, but then you need to think about connected systems that shouldn't receive API calls and mess up your data. Meanwhile the production site keeps changing, new plugins get installed, I have to sync/migrate everything etc. It would be so much easier to just work directly on production — run your debug scripts there, run your SQL queries on the real database with up-to-date data.
So I thought: wouldn't it be awesome to have Claude Code on my production website? That way I can let it do anything I want on real-time data.
So I set up a Hetzner server, installed Claude Code, Apache, and Node.js. I created a folder to host my WordPress files and a server user for Claude with permissions limited to only the WordPress files directory. I set up a Node.js server on the same box in another folder and created a React chat interface. The requests from the interface go into the Node server, into Claude Code, and responses flow back through the same channel (see it in action here).
I installed an mu-plugin that enables the Node.js server to authenticate via REST, so I can use my admin login credentials to access the chat interface.
The result: a subdomain chat.example.com where I can log in and talk to Claude Code to diagnose errors, install plugins, write new blog posts, build landing pages, and make adjustments.
I use the CLAUDE.md file to prompt the agent to behave in a certain way — for example, to never change third-party plugin code without asking for my approval first.
I also gave it its own "agent" folder where it can store logs, snapshots (backups) and other stuff. I prompted it to create a JSON formatted log file in agent/logs every time it makes a change to code or the database, so I have a history of all changes and can command it to reverse things if needed.
I also use the .claude folder to create sub-agents and slash commands, which I enabled in the chat interface. I'll keep extending the library of expert agents — so it can be my theme developer following my WordPress coding standards, or do SEO research on my latest posts, add internal links, whatever I need.
Long story short: the sky is the limit and this 100x's my output when working on my WordPress sites or my clients' sites. Bugs are found quickly now. I just log in, activate my @ diagnose agent, provide a report of what's going wrong. It comes back with a plan, makes the changes with my approval, created a git commit, and logs everything so I can look it up later.
I've created a demo video. Check it out! I might create a business out of it and offer AI hosting with hundreds of sub agents, slash commands and an easy-to-use chat interface.
rvermeulen1993•2mo ago
At the start I just used it for coding, then I started using WP CLI commands to test and debug stuff. But over time I realized it can basically do anything for me. It can run bash scripts, PHP eval for quick small scripts, and it can create SQL queries 100x faster than I can (yes, I have 15 years of coding experience and my SQL still sucks, haha). I also realized it can set up plugins for me, activate them, be my strategic partner, analyze SEO content... I was doing all of this in my local (DDEV) environment.
But then it hit me. In all those 15 years, one of the most annoying things about WordPress development is that debugging production sites is slow and painful. Sure, you can copy your production files to local, copy the database, but then you need to think about connected systems that shouldn't receive API calls and mess up your data. Meanwhile the production site keeps changing, new plugins get installed, I have to sync/migrate everything etc. It would be so much easier to just work directly on production — run your debug scripts there, run your SQL queries on the real database with up-to-date data.
So I thought: wouldn't it be awesome to have Claude Code on my production website? That way I can let it do anything I want on real-time data.
So I set up a Hetzner server, installed Claude Code, Apache, and Node.js. I created a folder to host my WordPress files and a server user for Claude with permissions limited to only the WordPress files directory. I set up a Node.js server on the same box in another folder and created a React chat interface. The requests from the interface go into the Node server, into Claude Code, and responses flow back through the same channel (see it in action here).
I installed an mu-plugin that enables the Node.js server to authenticate via REST, so I can use my admin login credentials to access the chat interface.
The result: a subdomain chat.example.com where I can log in and talk to Claude Code to diagnose errors, install plugins, write new blog posts, build landing pages, and make adjustments.
I use the CLAUDE.md file to prompt the agent to behave in a certain way — for example, to never change third-party plugin code without asking for my approval first.
I also gave it its own "agent" folder where it can store logs, snapshots (backups) and other stuff. I prompted it to create a JSON formatted log file in agent/logs every time it makes a change to code or the database, so I have a history of all changes and can command it to reverse things if needed.
I also use the .claude folder to create sub-agents and slash commands, which I enabled in the chat interface. I'll keep extending the library of expert agents — so it can be my theme developer following my WordPress coding standards, or do SEO research on my latest posts, add internal links, whatever I need.
Long story short: the sky is the limit and this 100x's my output when working on my WordPress sites or my clients' sites. Bugs are found quickly now. I just log in, activate my @ diagnose agent, provide a report of what's going wrong. It comes back with a plan, makes the changes with my approval, created a git commit, and logs everything so I can look it up later.
I've created a demo video. Check it out! I might create a business out of it and offer AI hosting with hundreds of sub agents, slash commands and an easy-to-use chat interface.
I call it: WP on Steroids! Here's a demo video: https://www.youtube.com/watch?v=QcZBIKIdDjU