dtu-learn: DTU Learn MCP server and API client for AI assistants

Use your DTU Learn courses with Claude and other AI assistants. dtu-learn is an MCP server and a command-line DTU Learn API client: it downloads your own DTU Learn (Brightspace / D2L) courses to your computer (slides, files, announcements, assignments, deadlines, grades and Panopto lecture transcripts), so your AI can answer "what is due this week?" or "which slide explains attention?".

Unofficial. A student project, not made or supported by DTU or D2L. Use it only with your own account and follow DTU's rules. Everything stays on your computer: there is no server and no shared login.

Install

You need a DTU account and Google Chrome (or Edge).

Claude Code (recommended)

/plugin marketplace add CarlSvejstrup/dtu-learn-mcp
/plugin install dtu-learn@dtu-learn

Restart Claude Code and ask: "Set up DTU Learn for me." It installs the tool and opens the DTU login. You type your password and MFA in the browser yourself.

Claude Desktop, Cursor, Codex and other MCP clients

curl -LsSf https://astral.sh/uv/install.sh | sh          # uv, if you do not have it (Windows: see uv docs)
uv tool install git+https://github.com/CarlSvejstrup/dtu-learn-mcp
dtu-learn setup                                          # login, first download, connects your AI apps

Restart your AI app. To add the skills too: npx skills add CarlSvejstrup/dtu-learn-mcp.

Related MCP server: Moodle MCP Server

Use it

Ask your AI:

• "What is due this week, and have I handed it in?"

• "What is new on DTU Learn since yesterday?"

• "Which lecture in 02132 covers pipelining? Give me the slide page."

• "Search the deep learning lecture transcripts for 'batch norm'."

• "Give me my DTU week." / "Quiz me on the NLP slides."

Fresh data is one question away: say "refresh DTU Learn" and the MCP refresh tool downloads what changed. If the DTU login has expired, the AI calls login and a browser window opens for you. Without an AI: dtu-learn scrape --current.

Optional extras

All off by default. Use the ones you want.

dtu-learn schedule install      # launchd (macOS), crontab (Linux) or Task Scheduler (Windows)
dtu-learn schedule status
dtu-learn schedule remove

It checks at 07:30 (or when the computer wakes) and refreshes at most once per 12 hours. After every check it writes ~/.dtu-learn/STATUS.md: result, whether you must log in, next refresh and what was new. Ask your AI "what is the dtu-learn status?" to read it. When the login has expired you get a notification; run dtu-learn login.

Copy sync.example.json to ~/.dtu-learn/sync.json and map each course id (dtu-learn courses --current) to a folder. dtu-learn sync copies new or changed files, never deletes, and skips a file whose content you already keep elsewhere in that folder's parent. The MCP refresh tool and the schedule sync automatically once sync.json exists.

If a destination is <course>/material/learn, sync also writes <course>/announcements.md (all announcements, newest first) and adds upcoming deadlines under ## Next in <course>/backlog.md. --no-vault turns that off.

Make an executable file ~/.dtu-learn/hooks/after-refresh. When it exists, every refresh groups new lecture slides per lecture in out/new_lectures.json, and the scheduled run calls the hook (at most 2 lectures per run). Guards: more than 4 new lecture folders at once counts as a re-download, not new teaching; old lectures never re-trigger.

Reference

course can be the org unit id, the course number (02132) or part of the name (nlp).

Update: uv tool upgrade dtu-learn. Remove: dtu-learn schedule remove, uv tool uninstall dtu-learn, then delete ~/.dtu-learn/.

Everything lives in ~/.dtu-learn/ (or DTU_LEARN_HOME). Per course in out/<code> <name>/: content/ (all files plus links.md), announcements/, assignments/ (instructions, attachments, your submissions), grades.json, linked/ (PDFs linked from course pages), videos.md and recordings/ (Panopto caption text with timestamps, never the video). Each refresh writes out/CHANGES.md. Not covered: quizzes, discussions, calendar, Locker.

• You log in once in a real Chrome window (DTU SSO + MFA). The session cookies stay in ~/.dtu-learn/state.json.

• Downloads use Brightspace's own REST API (/d2l/api/lp, /d2l/api/le), 3 requests at a time, and only fetch files whose modified date changed. One refresh runs at a time.

• Panopto transcripts use the same endpoints the Panopto player calls, one request at a time.

DTU rules

From DTU's own pages (checked 2026-10-03). Not legal advice.

• Your own login and copy only. Never share state.json, the browser profile or downloaded files. Each person downloads with their own account.

• Teachers' material into AI tools needs permission. DTU counts uploading material to AI chatbots as sharing, and allows only your own work, public domain and CC-0. Ask the teacher first.

• No other people's personal data into AI tools (names in announcements, lists or recordings).

• Keep the load low. Do not raise the parallel requests or refresh more than needed.

Sources: DTU IT security policy · Generative AI and copyright · Copyright for students

Troubleshooting

Development

git clone https://github.com/CarlSvejstrup/dtu-learn-mcp && cd dtu-learn-mcp
uv tool install --editable .
uv run --group dev pytest -q        # offline tests: no network, browser or scheduler
claude plugin validate . --strict

MIT licence. Not affiliated with DTU, D2L or Panopto.