Why we connected Mocxam to Claude and ChatGPT, and what we refused to build

Here is a scene most coaching directors will recognise. A teacher exports a results sheet, pastes it into ChatGPT, and asks which students are slipping. It works, sort of. The AI sees every column, including ones it never needed. Nobody knows which version of the sheet it read. Next week someone does it again from scratch.

Teachers were not waiting for us to give them AI. They already had it. What they did not have was a safe, accurate way to point that AI at their own institute. That gap is why we built the Mocxam AI connector, and it shaped almost every decision that followed.

Decision one: meet teachers inside the assistant they already use

The obvious move for a software company in 2026 is to build its own chatbot and put it in a corner of the dashboard. We considered it and said no. A teacher who already pays for Claude or ChatGPT does not want a second, weaker assistant that only knows one app. They want the one they use every day to know about their batches.

So we built on MCP, the Model Context Protocol. It is an open standard that Claude, ChatGPT and a growing list of assistants use to connect to outside apps. We are not the first education platform to use it, and we do not claim to be. Our job was to make it fit the way an Indian coaching institute actually works: exams with blueprints, batches, faculty, and results that matter on a Monday morning.

Decision two: the server computes, the assistant narrates

Large language models are very good at explaining things and unreliable at arithmetic over a few thousand rows. If we handed raw attempts to the assistant and asked it to rank students, we would get confident, plausible and occasionally wrong answers. In a coaching institute, a wrong "weakest student" is not a small error.

So every number comes from Mocxam. When a teacher asks who needs attention in Quant, the assistant calls a tool, and our server works out accuracy against the centre average, falling scores, skipped questions and inactivity from submitted tests in a stated window. Rankings ignore students with fewer than three tests by default, so one lucky paper does not top the list. The assistant receives a short, structured result and turns it into a sentence. Its job is language, not maths.

Decision three: the institute comes from the token, never from the prompt

Each teacher connects with their own Mocxam login. The connection is pinned to their institute at sign-in. No tool takes an institute as an argument, so no prompt, however clever, can ask for another institute's data. The plan is checked at sign-in, at every token refresh and on every call. If an institute moves to a plan without the connector, access stops at the next request.

What we refused to build

The most useful part of this project was the list of things we decided not to do.

• No free-form query tool. It would have been the fastest way to make the connector look powerful. It would also have made the AI the gatekeeper of our database. Instead there are 15 named tools, each with a narrow job.
• No student connections. The connector is for admins and faculty. Students use their institute's portal.
• No identity changes by chat. An admin can change portal colours and text or update staff access, always with a preview first. Logos, domains, the institute name and admin accounts stay in the console, where you see the full result before saving.
• No deletes. The AI can add questions to your bank. It cannot change or delete existing questions, students or results. Staff are disabled, never deleted.
• No meter. There is no usage quota and no per-call fee from Mocxam. The connector is included on Academy Pro and above.

What it means for an institute

A faculty member can ask for twenty on-pattern SSC CGL Quant questions and find them in the question bank a minute later, filed under the right section and ready for review. A director can ask how the morning batch compares with the centre before a parents' meeting. An admin can add a new evaluator without opening a settings page. None of this replaces teaching. It removes the clerical work between a question and an answer.

There are limits, and we will be plain about them. AI-written questions go straight into your bank today; a draft-and-approve step is on our list. Free ChatGPT accounts cannot add custom connectors yet, so those teachers should use Claude, which allows one custom connector on its free plan, until we are listed in OpenAI's directory. And what Anthropic or OpenAI do with a conversation is governed by their terms and the teacher's settings, which every institute should review. Our security page sets out exactly what the AI can and cannot see.

Why this matters now

Software for coaching institutes has mostly competed on features you can see in a demo: the test screen, the portal, the report. The next difference will be less visible. It is whether your system can be trusted inside the tools your team already thinks with. We would rather earn that trust slowly, with a narrow, logged and honest connector, than chase it with a flashy one.

If you run an institute and want to see it working on your own data, start a pilot and connect it from Claude or ChatGPT. I would like to hear what you ask it first.