Why I built this
I'm Vaibhav Gupta. I founded Edumaths360, where I teach all three CAT sections, quantitative aptitude (QA), logical reasoning (DILR) and verbal ability (VARC), to students preparing for CAT, other MBA entrance tests and similar competitive exams. I have spent about six years teaching and presenting. I am now moving into developer relations, technical content and solutions engineering, and this project is where those two worlds meet.
The teaching problem
An aspirant who is stuck on a time-and-work or profit-and-loss question rarely needs the answer. They need a question that points them at the next step: "What is the cost price here?" or "How much of the tank fills in one hour?" Good teachers know this. A general AI chatbot, left to itself, often does not. It solves the problem neatly, and the aspirant learns very little.
That worries me. More and more aspirants practise with an AI chatbot open beside them, and it is easy to mistake reading a clean solution for being able to produce one. In the exam hall there is no chatbot and the clock is running. What counts is the student's own method, built through many attempts, mistakes and corrections. An assistant that removes the struggle removes the learning.
When I learned about the Model Context Protocol, I saw a way to put the teaching rules outside the model, in code that I control. The server decides when a hint is allowed, what a hint may say, and when the worked solution can be shown. The model can still explain things in a friendly way, but it works inside rules that come from the classroom. The hint pedagogy page describes those rules in detail.
Why the documentation matters as much as the code
Developer relations work is mostly about one thing: helping another developer succeed quickly. So I treated the docs as the main product, not an afterthought:
- Diátaxis structure. Tutorials, how-to guides, reference and explanation are kept separate, because a beginner learning MCP and an engineer looking up an error code need different pages.
- Docs as code. The docs live in the same repository as the server, are reviewed in pull requests, and are built and checked by CI. A broken link fails the build.
- Code that cannot drift. Code examples are imported from real files that the test suite runs, and the outputs shown are captured from the real server. If the code changes, the docs change with it, or the tests fail.
- Real errors only. The Troubleshooting page lists errors I actually hit, with the messages copied from my terminal. Some came from following older tutorials, which is exactly the experience a new developer has.
- Plain English. Many developers read documentation in their second language. I try to explain each term the first time it appears.
What I learned
- The ecosystem moves fast. Version 2 of the MCP Python SDK renamed
FastMCPtoMCPServer, and much of the content online still uses the old name. Pinning versions and testing the docs is not optional. - Descriptions are an interface. The model chooses tools based on their descriptions and schemas, so writing them is closer to writing documentation than writing comments.
- Safety is a design choice. Keeping the answer on the server and validating every argument was easier to build in from the start than it would have been to add later.
What this project is, and is not
This is a portfolio project. It is a working, tested MCP server and a full documentation site, built to show how I approach developer education. It is not yet used by Edumaths360 students in production, and I have not measured learning outcomes or exam scores with it. If I do, I will report the method and results honestly here.
Watch the walkthrough
A short walkthrough video (under 3 minutes) will appear here. It will show the server running in Claude Desktop: generating a problem, giving a hint, and marking an answer.
Placeholder: the video has not been recorded yet, so there is no link.
Get in touch
I'd love feedback on the docs, especially from people who are new to MCP. You can open an issue or email me at Edumaths360@gmail.com.