Host it yourself (advanced)

Most people do not need this page. monash mcp on your own computer is enough for Claude Desktop, Claude Code, Codex, Cursor and the others.

This page is for when you want ChatGPT, or Claude on the web or phone, to use Study Kit. Those apps connect to MCP servers from their own cloud, not from your computer, so Study Kit needs an HTTPS address they can reach.

It is for one person and their own data. Never host it for friends with your account; each person runs their own.

Needs Study Kit 1.4.0 or later. Run monash update first.

Two ways to get an HTTPS address

	Your own server and domain	Cloudflare quick tunnel
What you need	A server or always-on computer, a domain with HTTPS, nginx or Caddy	Just your own computer. No account and no domain
Address	Fixed, e.g. https://study.example.org	Random, e.g. https://some-words.trycloudflare.com
Works when	The server is on	Your computer is on and both terminals are open
Good for	Daily use	Trying it out

The author uses ChatGPT Plus with their own domain and server. The quick tunnel route is there so you can try it without buying anything; if something goes wrong with it, please open an issue.

Either way, first install Study Kit and finish monash setup on the computer that will run it. Log in to Moodle and paste the Ed token in that computer’s own terminal. Never paste the Ed token or a Moodle cookie into a chat.

Option A: Cloudflare quick tunnel

1. Install cloudflared:

   brew install cloudflared                       # macOS
   winget install --id Cloudflare.cloudflared     # Windows

2. In terminal 1, start the tunnel and copy the https://….trycloudflare.com address it prints:

   cloudflared tunnel --url http://127.0.0.1:8765

3. In terminal 2, start Study Kit with that address:

   monash remote serve --public-url https://some-words.trycloudflare.com

4. Connect your AI app with https://some-words.trycloudflare.com/mcp.

Every time you restart cloudflared the address changes. Then restart monash remote serve with the new address, and remove and re-add the connector in your AI app. When you are done, press Ctrl-C in both terminals and run monash remote revoke --all.

While the tunnel is open, anyone with the address can reach the sign-in page, but nothing works without a pairing code from your terminal.

Option B: your own server and domain

1. Start Study Kit on the server:

   monash remote serve --public-url https://study.example.org --port 8765

   You can set MONASH_KIT_REMOTE_URL instead of --public-url. It only listens on 127.0.0.1 and does not touch DNS, certificates, nginx or system services, so keep it running with whatever you normally use (tmux, systemd and so on). If you use a separate data folder with MONASH_KIT_HOME, use the same value for every monash remote command.

2. Forward only the MCP and sign-in paths to that port. An nginx example to adapt inside an existing HTTPS site:

   location ~ ^/(mcp|oauth/(register|authorize|token)|\.well-known/oauth-(protected-resource|authorization-server)(/mcp)?)$ {
       client_max_body_size 1m;
       proxy_pass http://127.0.0.1:8765;
       proxy_set_header Host $host;
       proxy_set_header Authorization $http_authorization;
       proxy_set_header X-Forwarded-Proto https;
   }

   Add your usual rate limits and timeouts. Do not forward monash library or the data folder.

3. Connect your AI app with https://study.example.org/mcp.

While monash remote serve runs it also syncs every hour and keeps the Moodle login alive, like Claude Desktop does. Add --no-background if something else already does that.

Connect your AI app

ChatGPT: in Settings, turn on developer mode, then create a new connector (app) with your address ending in /mcp and OAuth as the sign-in method. The menu names change from time to time; follow what your ChatGPT shows. The author’s ChatGPT Plus account works. Which plans get developer mode is up to OpenAI, so check whether your account has it.

Claude: in Settings, Connectors, choose Add custom connector and enter the same address. Once added it works on the web, in the desktop app and on the phone. The free plan allows one custom connector.

The app then opens a page asking for a pairing code. On the computer running Study Kit, open another terminal and run:

monash remote pair

Type the code into that page. A code works once, expires after 10 minutes and locks after five wrong tries. Each app you connect needs its own code.

Then try “What units am I taking?” and “What do I need to do this week?”.

Claude and ChatGPT callbacks are allowed by default. For another app, add its exact callback address with --redirect-uri (repeat it for more than one). Use the address that app tells you, not a guess.

See and remove connected apps

monash remote clients
monash remote revoke --client CLIENT_ID
monash remote revoke --all

After revoke the app has to pair again.

What remote mode can and cannot do

• It offers 16 read-only query tools and the same panels and prompts as the local server.
• Logging in to Moodle and starting a sync are left out. Do those on the computer itself.
• There is no way to set credentials, submit assignments, attempt quizzes, post on Ed or download files over the internet.
• Sign-in uses OAuth with PKCE. Access tokens last one hour. Refresh tokens last 30 days and change every time they are used. Only hashes are stored, in secrets/remote-mcp.db.
• Changing the address means every app has to pair again.
• On Windows, also restrict who can read the data folder.

How it was checked

The test suite runs the whole flow with made-up data on one machine: discovery, a 401 without a token, registration, pairing, PKCE, token expiry and rotation, revocation and refusal of the login and sync tools. On 30 October 2026 the 1.4.0 package from PyPI also passed the same flow end to end in a clean environment, using the Claude and ChatGPT callback addresses.

If you deploy it on your own server, check it once: the discovery URLs load over HTTPS, /mcp without a token returns 401, the app connects, a simple question works, and a second question still works after the app refreshes its token.

The full technical notes are in the project repository.