Why I built a quiz tool for my coding class — and then an MCP server so I'd never leave Claude

Tom Kim — Sat, 16 May 2026 15:19:06 +0000

The Problem

I'm a coding instructor in Korea, and after every class I give students short review quizzes.

Generating questions is easy now. ChatGPT or Claude can do that in seconds.

The annoying part starts after that:

The AI part took 10 seconds.

The delivery + grading workflow took 30 minutes.

That felt backwards.

I built a tool called Leafeep to remove the boring parts of the workflow.

The flow is simple:

No student accounts. No app install.

Just a fast way to go from “questions” → “responses”.

I already use Claude Code every day.

So why was I still opening a browser to create quizzes?

That led me to build an MCP server for the project.

Now the workflow looks like this:

me: "Make 5 Python list questions and send them to students"

Claude:
✓ Quiz created
✓ Student link generated
✓ Ready to share

me: "Grade the exam and summarize results"

Claude:
✓ 12 submissions
✓ 73% average score
✓ Q3 had the highest error rate

me: "Analyze weak points for student Kim"

Claude:
✓ Struggles with list slicing
✓ Improving on loops

At that point, the browser almost disappeared from the workflow.

Creating → distributing → grading → analyzing all happened inside Claude.

Before building this, I thought MCP was mainly useful for connecting LLMs to big external services like GitHub, Slack, or databases.

Now I think the most interesting MCP servers are much smaller and more personal.

They're wrappers around repetitive workflows.

The things you do every day with:

For me, that workflow was:

questions in → student link out → answers in → grades out

That ended up being a surprisingly good fit for MCP.

I originally built this for myself, but if you teach or tutor, I'd genuinely love feedback.