Skip to content

Connect your own ChatGPT or Claude to AI Teacher

The AI Teacher connector lets a member of staff add AI Teacher to their own ChatGPT or Claude. It acts as that person, in their own role, on the programmes they build or administer, and leaves drafts that a person reviews and publishes in AI Teacher.

Server address
https://api.ai-lms-teacher.com/mcp
Paste it where ChatGPT or Claude asks for the connector’s address. You then sign in to AI Teacher itself: there is no key or password to paste.
A remote MCP server over Streamable HTTP, signed in with OAuth 2.1 and PKCE. The assistant registers itself.

Who can use it

How to connect

You need the server address above and your AI Teacher sign-in. Both products change these screens often: where yours differs, follow its own screen. All it needs from AI Teacher is the address and your approval.

In Claude

  1. 01Open Customize, then Connectors.
  2. 02Press + Add, then Add custom connector.
  3. 03Give it a name you will recognise, paste the server address and press Continue.
  4. 04Keep the sign-in settings Claude suggests, press Continue, choose Sign in now and press Add.
  5. 05AI Teacher’s approval page opens. Sign in if it asks, read what the assistant will be able to do, and press Approve.
  6. 06In a chat, press +, then Connectors, and switch it on.

On a Team or Enterprise plan of Claude, an Owner adds it once under Organization settings, Connectors, and each person then presses Connect and approves it for their own account.

In ChatGPT

  1. 01Open the Plugins page, at chatgpt.com/plugins.
  2. 02Press the plus button, then Add custom MCP server.
  3. 03Give it a name and a description and, under Connection, paste the server address, with /mcp at its end.
  4. 04Set the authentication to OAuth, read the warning, choose I understand and want to continue, then Create as a plugin.
  5. 05AI Teacher’s approval page opens. Sign in if it asks and press Approve.

Whether a ChatGPT account may add one depends on its plan and, in a workspace, on that workspace’s own admin.

The steps are as each product’s own help described them on 9 October 2026.

The approval page

AI Teacher opens its own page and asks you, signed in as usual, whether to connect. It shows you:

  • who is asking, named by the address its key will be sent to (Claude or ChatGPT for the address each publishes for itself), and that address in full;
  • the account you are signed in as;
  • what it will be able to do with your role, and what it never can;
  • that what it reads goes to your own assistant’s provider, under your own account, and is not paid from your organisation’s AI budget;
  • a caution above Approve and Deny: approve only a request you started yourself, just now, and choose Deny for a link somebody sent you.

Setting the language a programme is taught in is its own line on that page, the one thing the assistant changes that is not a draft. An assistant connected before the page listed it cannot set a language until you disconnect it and connect it again; Connected assistants on your Profile says which can.

When you approve, AI Teacher emails you and leaves a notice in your portal: which assistant, when, and where its key was sent.

Disconnect any time

  • In AI Teacher: Profile, Connected assistants, Disconnect. Its key stops working at once.
  • In the assistant: in Claude, Customize, Connectors, Remove; in ChatGPT, in its own settings.
  • Setting a new AI Teacher password ends every assistant connected to your account. If you see one you do not recognise, disconnect it and change your password.

What it can do

Fifteen functions, each offered only to the roles that can do the same in their own portal, and only on a connection approved for it. Your assistant calls them when you ask in your own words.

Eight that read

Declared as read-only. They change nothing.

  • List my programmes

    list_my_programmes

    The programmes you work on, each with its status, language, children or adult mode, whether it is for learners or staff training, and how many levels it has. Archived programmes are left out.

    Who: Teacher: the programmes they build. Admin: their branch’s. Super Admin: the whole organisation’s.

  • Read a programme’s outline

    get_programme_outline

    Its levels, skills and units, each unit’s materials by title and kind (never a file or its text), the languages a video can be watched in, and how many published and draft questions each unit and skill has. Where a Start from your materials setup is open, the names of the files waiting in it.

    Who: Teacher, admin, Super Admin

  • Read how a question is written

    get_question_format

    The question types and the rules of the question import, so that what it drafts fits.

    Who: Teacher

  • Search the help

    search_help

    Up to five matching articles of your own portal’s help, each with a short excerpt.

    Who: Teacher, admin, Super Admin

  • Read a help article

    get_help_article

    One whole help article, as your portal shows it to your role.

    Who: Teacher, admin, Super Admin

  • Read the learner theme

    get_learner_theme

    The theme your learners see and its colours, and whether your own pictures were uploaded, never the pictures.

    Who: Super Admin

  • Check a colour set

    check_theme_colours

    Checks proposed theme colours by the same contrast rules as Settings and, where they fail, gives the nearest colours that pass. Saves nothing.

    Who: Super Admin

  • List my drafts

    list_my_drafts

    The drafts your own assistants left that still wait for you, newest first, at most 50, each with where to find it.

    Who: Teacher

Five that leave a draft

Declared as writing, which tells your assistant to ask you before each one. What they leave waits in AI Teacher, marked as coming from an AI assistant, until a person reviews it.

  • Draft questions

    draft_questions

    Multiple choice, choose several, true or false, fill in the blank or written answer, in one level and skill. Each lands in the question bank labelled for its answer to be checked; the teacher publishes each one by hand, one at a time, or turns it down.

    Who: Teacher, on a programme they build

  • Propose a programme’s outline

    propose_programme_outline

    Levels, skills and units for a programme with no unit yet, and which unit each file the teacher uploaded goes in. None of its levels, skills or units is created and no file moves until the teacher accepts it in the builder.

    Who: Teacher, on a programme assigned to them

  • Draft a lesson text

    draft_lesson_text

    One written lesson on a unit, in plain text. No learner can open it until the teacher publishes it.

    Who: Teacher, on a programme they build

  • Suggest theme colours

    suggest_theme_colours

    A colour set that waits on the learner theme card in Settings, to preview, then use or dismiss. The theme does not change.

    Who: Super Admin

  • Create an empty draft programme

    create_draft_programme

    A programme with a name and nothing in it, which no learner sees. An admin then assigns the teacher who builds it.

    Who: Super Admin, in any branch. Admin, in their own branch.

One that changes a setting

The one function that changes something at once and leaves no draft. Your assistant is told to call it only when you ask for it yourself.

  • Set a programme’s language

    set_programme_language

    Bahasa Melayu, English or Chinese: the language the programme is taught in, which new drafts for it are written in. It translates nothing, changes no question or lesson already written and never touches an archived programme, and your assistant tells you which language it was and which it is now.

    Who: Teacher: a programme they build. Admin: in their branch. Super Admin: any.

One that takes back a draft

Declared as destructive, because a question turned down cannot be turned back, which tells your assistant to confirm with you first. It deletes nothing.

  • Take back a draft

    withdraw_draft

    One draft your own assistant left that nobody has reviewed: a question is turned down, a lesson retired, which the teacher can restore, an outline taken out of the builder. Anything a person published, accepted or wrote is refused.

    Who: Teacher

Limits

  • At most 50 questions in one request. While 100 of an assistant’s drafts wait unreviewed in a programme, it drafts no more there until the teacher has read them.
  • A lesson text of at most 4,000 characters. A longer one is refused whole, never cut.
  • Up to 600 calls an hour for one person and 3,000 for an organisation, and within them 60 drafts an hour for one person and 300 for an organisation, each question counting as one.

Things to ask it

Ask in your own words. Four requests the functions above answer:

What it never does

Data and privacy

  • It acts as the person who signed in, with their role and nothing more. Their organisation’s own access rules decide what it reads, and every call is checked again: the connection, the person still active staff, the organisation’s switch still on.
  • It returns authored teaching content and help, and nothing else: programme names and outlines, materials’ titles and kinds (never what is in a file), question counts, AI Teacher’s own help and question rules, and for a Super Admin the theme’s colours. Nothing about a learner or a guardian is read or returned.
  • What it reads goes to the assistant you use and is handled by its provider under your own account with them, and what it drafts is written there. None of it is paid from your organisation’s AI budget.
  • Its keys are issued by AI Teacher, sent only to the address ChatGPT or Claude publishes for its own assistant, and work only at the connector, nowhere else in AI Teacher. AI Teacher keeps a one-way fingerprint of each key, never the key.
  • A connection ends when its person disconnects it, sets a new password, changes role or is made inactive, and when the organisation switches the connector off or moves to a plan without it.
  • The log of each call names who made it, the function and its outcome, never what was asked or answered. Every draft it leaves and every change it makes is in the organisation’s audit trail, under the person’s name.

How AI Teacher handles personal data is in the privacy notice.

Support

Questions about the connector, or a connection you do not recognise:

Whether the connector is on for your organisation is your Super Admin’s to decide, in Settings.