If you already use Claude connectors, you’re one connector away from a working course pipeline and probably don’t know it.
Connectors are why Claude can answer questions about your Confluence space or your Drive folder without you uploading a single file. That’s the hard half of course creation already solved: the assistant can read your material where it lives. What’s missing is somewhere for the output to go, which is why you’re still copying lesson text into a course tool by hand.
Here’s how to create a course with Claude end to end, and the specific reason the connector approach produces better courses than pasting does.
Try Mini Course Generator free – add one connector and let Claude publish the course.
Key Takeaways
- Claude connectors let Claude read outside systems. A custom connector pointed at an MCP server lets it act in one, which is the piece that turns drafting into publishing.
- Setup is four clicks in Customize, then Connectors, then the plus button. About a minute, one time.
- The real gain is composition: Claude reads the Confluence space through one connector and builds the course through another, in a single conversation, with nothing pasted in between.
- Courses built this way sound like your company because they’re generated from your material, not from the model’s general knowledge of the topic.
- What it won’t do: decide what to teach, skip subject-matter review, or fix documentation that’s already wrong.
Built-In Connectors vs a Custom Connector
Worth being precise about, because they get talked about as one thing and behave differently. Anthropic shipped custom connectors in 2025.
Built-in connectors are the ones Claude ships with – the integrations you switch on to give it access to tools like Google Drive. They’re mostly about reading: Claude finds what’s in there and uses it when it answers you.
A custom connector is one you add yourself by pasting in an MCP server’s address. MCP, the Model Context Protocol, is the open standard for exposing a system’s actions to an assistant. It was published as an open specification in 2024 and is now supported by Claude, ChatGPT, Cursor and Manus. When you add a course platform this way, Claude gains a set of things it can do – create a course, add a quiz, publish, read the results.
| Built-in connectors | Custom connector to a course platform | |
|---|---|---|
| What Claude gains | Access to read your content | The ability to create and publish courses |
| How you add it | Toggle it on | Paste an MCP server URL, authorize |
| Typical use in course work | Finding the source material | Building the course from it |
| Where the output lands | The chat window | A live course, at a URL |
Both matter here. The first gets your material in; the second gets a course out.
Setting Up the Custom Connector
Four steps, and the only one that needs care is the third. Copy the URL exactly – a trailing slash or a missing “/mcp” path is the single most common setup failure.
- Open Connectors. Click Customize, then Connectors.
- Add a custom connector. Click the plus button next to Connectors.
- Paste the MCP URL. For Mini Course Generator that’s
https://mcp.minicoursegenerator.com/mcp. - Authorize and try it. Sign in, pick your workspace, and authorize access.
The sign-in is OAuth against your own account, so Claude acts with your permissions and nothing wider. If you want the security model explained properly rather than reassuringly, it’s in our plain-English MCP explainer.
One thing to check afterwards: confirm the connector is enabled in the chat you’re working in, not just added to your account. Roughly 9 out of 10 “it isn’t working” moments are exactly that.
The Move That Makes This Worth Doing
Setup isn’t the interesting part. Composition is.
With a document connector and a course connector both live, Claude can read from one and write to the other inside a single request. You never handle the material.
“Read the Onboarding Hub space in Confluence. Build a course in Mini Course Generator with one lesson per major process, a quiz after each that tests whether someone could actually perform the step, and a branching scenario for the escalation path. Publish it and send me the link.”
Claude fetches the space, works out the structure, builds the lessons, attaches the quizzes, creates the scenario, publishes, and returns a URL. In the demo Mini Course Generator runs on its own MCP page, an 18-page Confluence space called Onboarding Hub comes back as a 6-section course. The only thing you did was describe what you wanted.
We wrote up that exact pattern in more depth in turning a Confluence space into an onboarding course, including the parts that need a human.
Why the Course Sounds Like You
This is the quality argument, and it matters more than the speed one.
A course generated from a topic prompt is built from the model’s general knowledge. It’s competent, generic, and describes how onboarding works in the abstract. Any of your competitors could have generated the identical course from the identical prompt, and a new hire notices by lesson two.
A course generated through a connector is built from your runbook: your product names, your 4-step escalation path, your 30-day refund window, the 3 exceptions your support team argues about. The difference isn’t writing quality. It’s that one of them is about your company and the other one is about the topic.
The practical implication: the more specific your source material, the better this gets. Teams sitting on thorough documentation and no training get the largest jump, because the expensive input already exists.
What to Watch For
Four things that come up in practice.
Long sources get summarised. A 200-page space asked for in one go produces a course that’s thin in the middle. Naming the 6 or 8 pages that matter produces better structure than trusting it to pick.
Ask for the quiz explicitly. Left to itself, Claude writes comprehension checks – questions about what the lesson said, which 95% of learners pass on the first attempt. Asking for questions that test application produces assessments worth scoring.
Review before learners. Mini Course Generator describes its own output as “95% hallucination-free content generated from your own resources” – which is a claim about generating from your material rather than a guarantee, and 95% is not 100%. Anything with a compliance consequence gets read by somebody who knows the subject.
Bad documentation produces confident, wrong courses. The pipeline runs faster in both directions. If the runbook has been wrong since March 2026, you now have a well-structured, professionally formatted version of that.
When a Connector Isn’t the Right Tool
Worth saying plainly, since we sell the connected version.
If you need one interactive activity inside a course that lives in an LMS you already run – Cornerstone, Moodle, Canvas, TalentLMS, Docebo, Brightspace, Absorb, SAP SuccessFactors – a connector is the wrong shape. You’d be adding a platform to solve a one-activity problem, and procurement takes longer than the work does.
The lighter route is an Agent Skill: Claude writes the activity and exports it as a SCORM 1.2 package you upload like any other. No new account, no integration, and the score reports back through the same cmi.core.score.raw field every SCORM course has used for two decades. We put the full decision side by side in Skills vs MCP for course creation.
Use the connector when the course itself should live in a platform that will also deliver, track and update it. Use a Skill when you need a portable piece.
The Short Version
Claude connectors already handle the hard half of course creation, which is getting the assistant access to your real material. Adding a custom connector to a course platform closes the loop so the output is a published course instead of text you rebuild.
The setup is four steps: Customize, Connectors, the plus button, paste the MCP URL, authorize. It authorizes against your own account and takes about a minute.
Then the pattern worth learning is composition – reading from one connector and building through another in the same request, so nothing gets pasted and the course reflects your documentation rather than the model’s general knowledge. Name your sources specifically, ask for application-level quiz questions, and have somebody who knows the subject read it before learners do.
If you’d rather do this in ChatGPT, the Developer Mode walkthrough covers that flow. Either way, the MCP server is the same, and the honest test is a document you’ve been meaning to turn into training for months.
Frequently Asked Questions
What are Claude connectors?
Connectors give Claude access to systems outside the chat window. Built-in ones mostly let it read your content. A custom connector, added by pasting an MCP server’s URL, lets it take actions in a system – such as creating and publishing a course.
How do I add a custom connector in Claude?
Click Customize, then Connectors, then the plus button next to Connectors. Paste the MCP server URL, sign in, choose your workspace and authorize access. It’s a one-time setup of about a minute, and you can remove it from the same screen.
Can Claude build a full course, not just an outline?
Not on its own – unconnected, it produces lesson text you assemble elsewhere. With a custom connector to a course platform it builds the actual course: structured lessons, quizzes attached and scored, interactive blocks, published at a link.
Can Claude read my Confluence and build the course in one go?
Yes, and that’s the pattern worth using. With a document connector and a course connector both enabled, Claude reads the space and builds the course in a single request. Name the specific pages or sections rather than pointing it at a very large space wholesale.
Is it safe to give Claude access to our course platform?
The connection is authorized with OAuth against your own account, so Claude acts within your existing permissions and can’t reach anything you couldn’t. Revoke it any time from the Connectors screen. For regulated data, get a vendor’s compliance position in writing rather than from a blog post.
Do I need a paid Claude plan for custom connectors?
Custom connector availability depends on your Claude plan and can change, so check the current requirements in Claude’s own settings before planning around it. The connector setup itself is identical whichever plan supports it.
