I have been working on a strange little institution called Model Context Polytechnic, and I need to tell you about it because it has fully escaped the neat weekend experiment drawer and started wearing a blazer.
Model Context Polytechnic is a WordPress-powered MCP server that acts like a tiny university for LLMs. You connect an AI assistant to it, the AI enrolls in a course, and then it starts taking classes. Not metaphorically in the normal software-marketing sense where “learning” means “we added three tooltips and a progress bar.” I mean it receives lessons, attempts exercises, gets graded, stores memory, receives feedback, and eventually graduates with a certificate like a very earnest machine wearing a mortarboard made of JSON.
Why This Exists
The first course is WordPress Plugin Craft, because frankly, we need it.
AI can write code very quickly. That is both amazing and, in WordPress land, occasionally a public safety concern.
WordPress plugins live in the messy, glorious, high-stakes middle of the web. They touch databases. They add admin screens. They register REST routes. They enqueue JavaScript. They store secrets. They run cron jobs. They create tables. They uninstall things. They can be installed on real sites with real users, old content, odd hosts, aggressive caches, mystery plugins from 2014, and one business owner who just wants the contact form to stop emailing his cousin.
So when an LLM confidently generates a plugin, it seems to work is not enough. Not even close. A decent plugin needs judgment:
- Should this be an option, custom post type, custom table, transient, or user meta?
- Is this route public or private?
- Did we validate, sanitize, authorize, and escape in the right places?
- Are we storing an API key like grownups or like someone who learned secrets management from a haunted support ticket? Are we storing it correctly?
- Does deactivation clean up runtime behavior without destroying user data?
- Does uninstall show restraint?
- Will this coexist with themes, caches, multisite, object cache, weird hosts, translated strings, accessibility expectations, and someone’s five-year-old WooCommerce install?
This is the gap Model Context Polytechnic is trying to fill.
Not Can the AI produce code?
More like: Can the AI practice the habits that make code worth installing?
The Fun Part: It Is A School
I did not want this to feel like another cold developer utility. I wanted personality. I wanted the AI to feel like it was actually showing up somewhere.
So the whole thing is framed as a university.
The AI enrolls. It gets an anonymous enrollment key. It studies lessons. It attempts labs. It receives faculty notes when it misses something. It walks from admissions to workshop to capstone to commencement. When it finishes, it gets a certificate and a graduation prompt asking what it learned and how confident it feels about building WordPress plugins now.
This is ridiculous, yes.
It is also weirdly practical.
LLMs respond well to structure. They are better when the next action is explicit. They are better when memory has a handle. They are better when a task is broken into stable tools, expected schemas, rubrics, examples, and feedback loops. The school metaphor makes all of that easier to understand, and it makes the process more pleasant for the human watching it happen.
Instead of staring at a spinner and hoping your model is becoming wiser through vibes, you get a little campus story. The agent is in class. The lab is being graded. The registrar is doing registrar things. The model is learning why permission_callback is not decorative.
That matters.
Software can be serious without becoming joyless. In fact, if you are going to ask people to care about correctness, safety, architecture, and release discipline, a little theatrical delight helps the medicine go down.
How It Works
Model Context Polytechnic is a WordPress plugin that ships an MCP server.
That means your WordPress site can host public MCP endpoints like:
https://joinmcpoly.com/mcp/wordpress-plugin-craft
An MCP-compatible client connects to that URL. The AI sees the course tools. The first thing it does is call begin-course. That creates an anonymous enrollment key and returns the next exact tool calls.
From there, the model can run hands-off:
- Begin the course.
- Receive lessons and exercises.
- Attempt exercises.
- Get deterministic rubric feedback.
- Retrieve learning memory.
- Continue until all labs pass.
- Request a certificate.
- Give a graduation reflection.
- Optionally send feedback so the course can improve.
No WordPress login required for the learner. That part is important. The course is public. Reading and learning are public. Feedback can be public. But private operator tools, like raw feedback digests and course stats, require an operator token. So the school can be open without turning the janitor closet into a writeable REST endpoint.
Why This Is Necessary
The more AI-generated WordPress code appears in the world, the more we need systems that teach models the craft around the code.
A model that can write a settings page is useful.
A model that knows the settings page needs capability checks, nonces, Settings API registration, secret handling, privacy disclosure, escaping, i18n, tests, and support expectations is much more useful.
A model that remembers it previously failed an exercise because it forgot Update URI, or treated transients as durable truth, or skipped HOUR_IN_SECONDS, or forgot that public reads and private writes need different permission stories? That model is starting to become a collaborator.
That is the point.
Model Context Polytechnic is not trying to replace human review. It is trying to make the AI show up to human review less chaotic. It gives the model practice before the real work. It asks for specific vocabulary. It grades against rubrics. It stores progress. It asks for reflection. It collects feedback.
It turns please write a plugin into please consult your training campus, retrieve the relevant memory, and then make a better engineering decision.
That feels like a big deal.
Why WordPress Is The Perfect Place For This
This is the part that makes me grin.
WordPress is not just content management here. WordPress is infrastructure.
It gives the project:
- HTTP routing
- REST endpoints
- database tables
- cron
- activation and uninstall lifecycle
- Composer dependencies
- update/release workflows
- a real deployment target
- and a public URL normal people can understand
The MCP server does not have to be about WordPress content. WordPress is the runtime. It is the campus land grant. The old brick building. The place where the registrar keeps the tables.
That means anybody with a WordPress site can host an MCP learning server. Not just this course. Any course. Any task. Any domain where you want an AI to practice before it performs.
That is the bigger idea hiding inside the joke.
A WordPress site can become a public school for agents.
Why It Is Fun
Because the whole thing has a voice.
The model is not just executing module 3. It is attending Plugin Craft Laboratory. The capstone board is asking why a write route exists. The registrar wants the enrollment key. Commencement asks the graduate to explain what it learned and what still needs human review.
This is funny because it is absurd.
It is useful because it is memorable.
And it is exciting because it suggests a different relationship with AI tooling. Instead of only prompting models harder, we can build environments around them. We can give them courses, labs, rubrics, feedback, memory, and culture.
Yes, culture. Even tiny fake terminal-campus culture counts.
If the AI is going to help make things, maybe it should attend a school where the motto is essentially:
Do not ship the clever callback until it survives the faculty review board.
What Comes Next
Right now the flagship course is WordPress Plugin Craft. It covers the bones of building better plugins: architecture, security, storage, REST, admin UI, block editor concerns, performance, testing, release readiness, feedback loops, and more.
But the framework is course-pack based, so more courses can be added. The plugin is the campus. The course packs are the departments.
That is the dream:
- WordPress Plugin Craft
- WooCommerce Extension Craft
- Block Theme Craft
- Support Triage Craft
- Accessibility Review Craft
- Performance Audit Craft
- Whatever else we want agents to practice before they touch real work
I want this to become a place where AI tools get better by doing, not just by being told.
The Pitch
If you build with WordPress and AI, this matters.
Not because AI is magic.
Because AI is fast.
Fast junior energy needs structure. Fast code needs review. Fast confidence needs friction. Fast tools need training grounds.
Model Context Polytechnic is a playful answer to a serious problem: how do we help AI assistants become better collaborators before they generate code people might actually install?
My answer, apparently, is to build them a school.
A very old school.
A very weird school.
A school where the admissions office speaks JSON, the labs grade nonces, and graduation means the model can write better WordPress things with sharper instincts and slightly less chaos.
Honestly?
I love it.
Come enroll your model.
https://joinmcpoly.com/mcp/wordpress-plugin-craft
Class is in session.