NTNU Course Data API, MCP Server & Timetable

Why it exists

To be able to ask Claude about the schedule. That is the whole reason.

NTNU’s course data is public but not in a state to be looked up: three systems that were never introduced to each other and agree on nothing, including how a course is identified. Exam logistics are not data in any of them — they are published on the course page as HTML. So ntnu-api turns the sources into one thing to query, and ntnu-mcp puts that where a model can reach it.

The client

TypeScript over fetch with nothing underneath it. One build has to run in Node, in a browser tab and inside Cloudflare’s runtime, so it carries no dependencies at all.

Namespace	Data	Source
courses	catalog search, schedules, weekly timetables	ntnu.no JSON
courses.details	exams, fact box, descriptions, programs	ntnu.no HTML, scraped
grades	grade distribution per course, year, semester	HK-dir DBH
programs	~400 study programs, per-cohort study plans	ntnu.no JSON
semesters	term ids, teaching weeks, exam periods	TP

The work is in the seams. Exam information is scraped and typed on the way out. Catalog search hands back the same course twice. Grade lookups want TDT4100-1 where the rest of NTNU says TDT4100.

The MCP server

Twelve read-only tools on a Cloudflare Worker, over Streamable HTTP, no key. Two of them do something the catalog will not: compare_courses puts candidates side by side, and check_timetable_conflicts finds clashing lectures and colliding exams, separating lecture against lecture — fatal — from a clash with an exercise group that usually has another slot going.

The console

A real MCP client in eighty lines of fetch, with no model behind it. That is why it asks in fill-in-the-blank sentences: the sentence chooses the tool and the blanks are the arguments. Put your own codes in and the call is the one Claude would make.

The timetable

Semesterplan, at ntnu.martinsundal.no, is the other thing built on the client. Pick a study programme and a cohort and the week comes back, rooms and exam dates included. NTNU auto-enrolls programme students in their courses, so the programme is the plan rather than a suggestion to confirm; a course comes off it in one press.

An Astro build and a Cloudflare Worker ship as one unit. The Worker serves the site and caches /api/* in front of NTNU per data type — six hours for course details, one hour for timetables. There is no account, and nothing about a student leaves the browser it was typed into.