← Home

Say what you want,
your platform does it

Geduk exposes an MCP server: some fifty operations your AI assistant can carry out for you, in your platform, with your permissions. Create a session, enrol a cohort, turn on a reminder, publish to the catalogue. One sentence instead of half an hour of clicking.

Try for free See the setup

The real cost of an LMS

It is not the platform
that takes up the time

It is the September intake with eleven sessions to open, the enrolment list that arrives as a spreadsheet, the reminder you forget to turn on, the session you forget to publish. None of these tasks is hard. All of them are long, repetitive, and land at the worst time of the year.

A well-built LMS reduces the number of clicks. It does not remove them: someone still has to say what they want. The MCP server moves that boundary. You describe the result you expect in plain language, and the assistant chains the operations in the platform by itself, with your permissions and under your control.

MCP, for Model Context Protocol, is an open standard. It describes how an application publishes its operations so that an AI assistant can call them. Geduk publishes the list, your assistant reads it and picks the tools that fit your request. Nothing is tied to one AI vendor, and nothing locks you in: the same server answers any compatible client, today and tomorrow.

What it looks like

One sentence,
four operations

The example below is an ordinary start-of-term request. It takes about twenty minutes in any administration interface, yours included. A filmed run of the same kind of request on app.geduk.fr, captioned in French, is on the French page.

You Open a “Cybersecurity” session from 9 to 13 November, remote, with the learning path of the same name. Enrol the twelve people on the list I just gave you, remind them two days before it starts, and publish it to the catalogue at €250.
Your assistant Done. I found the “Cybersecurity” learning path and created the session on your dates, with its modules. I scheduled the reminder before enrolling anyone, so that everyone gets it. Two people on the list already had an account, so I reused them rather than creating duplicates.
  • Session created, 9 to 13 November, remote, reminder at D−2 lms_create_session
  • 10 accounts created for the newcomers lms_create_user
  • 12 learners enrolled lms_enroll_user_to_session
  • Published to the catalogue at €250 lms_publish_session_to_catalogue

The assistant names the tools it uses and reports back on what it did. Each operation is an ordinary write in the platform: it appears in the logs, under the name of the person the token belongs to.

The point is not to replace a click with a sentence. It is to chain operations that, by hand, take you through four different screens, and to accept data as it arrives: a list pasted from an email, a messy spreadsheet, a request made out loud in a meeting.

What is exposed

Some fifty tools

Each tool matches an LMS operation, with the same business rules and the same checks as the interface. It is not a back door into the database: it is the same door, opened another way.

Users and roles 11 tools

Create, edit and search for an account, import a whole list from a CSV or Excel file with its groups, assign a role, view existing roles and privileges, create new ones.

Modules and learning paths 8 tools

Create and update modules and learning paths, list them, read their structure. File upload lets you attach images, videos and documents to them.

Sessions and catalogue 6 tools

Open a session, edit it, publish it to the catalogue with its price and enrolment mode, list existing publications.

Enrolments 3 tools

Enrol a person in a session, list the enrolled learners, view an enrolment and its progress.

Notifications 6 tools

Read the templates that ship with Geduk, create or edit an email notification, adjust its trigger, delete it.

Pages and home page 6 tools

Create and edit portal pages, read and adjust the learner home page configuration.

Domains and settings 6 tools

Create a domain, edit its advanced settings, switch the current domain for the following operations — useful to organizations that run several entities or brands.

Translations and files 4 tools

Read and edit the platform's translated labels, upload a file and get its address to use it in content.

The exact list is published by the server itself: your assistant discovers it when connecting, so you have nothing to keep up to date. It grows with each LMS release.

The question that comes right after

An assistant never has
more permissions than you

That is the only thing separating a useful feature from an unacceptable risk. Here is exactly what is in place.

The person's permissions, not the server's Operations run in the context of the account the token comes from, within its domain. A trainer remains a trainer, including through the assistant.
The tool list is filtered The assistant only sees the tools the account is allowed to use. Without permission to edit users, the creation tool does not even appear — and a direct call would be refused.
Privileges are re-read on every request No snapshot taken at sign-in: a removed role stops applying immediately, even if the conversation is already open.
Access you can cut off whenever you want Through OAuth 2, the assistant receives a fifteen-minute token, renewed for as long as access is authorized: deleting the client in Geduk stops the renewals. A personal token, for its part, is named and dated, and its revocation, checked in the database on every call, takes effect right away.
A session cannot be borrowed The technical conversation ID is not a password: it is bound to the fingerprint of the token holder, and is rejected when presented by another caller.
Logs without secrets Authentication headers are masked before writing. A log file must never become a key ring.

What we will not let you believe: what the assistant reads in your platform goes through the AI provider you chose, exactly like everything you write to it. Geduk adds no intermediary and sends nothing on its own, but cannot promise anything about the processing done downstream. For use with sensitive data, connect the assistant with an account with deliberately limited permissions, and check your provider's commitments.

Setup

Three steps, five minutes

This is the path for Claude, which connects through OAuth 2: you grant access on a Geduk page, and the assistant never sees your password. For Mistral's Le Chat or a client configured through a file, see the box below.

  1. 1

    Register Claude in the platform, once and for all

    An administrator opens Supervision → Integration → OAuth2 clients and registers a client: a name, Claude's redirect address, and the scopes offered, keeping roles ticked, since that is what carries your permissions. Copy the client ID and secret: the secret is never shown again. One client is enough for the whole team; everyone then signs in with their own account.

    Redirect URI
https://claude.ai/api/mcp/auth_callback
  2. 2

    Add the connector in Claude

    In Settings → Connectors, add a custom connector. Its address is your platform's address followed by /mcp. Under Advanced settings, paste the client ID and secret. Click Connect: Geduk shows its consent page, you sign in with your account and authorize. The assistant receives a fifteen-minute token, renewed on its own, tied to your account.

    MCP server URL
https://app.geduk.fr/mcp

    On a dedicated instance, replace the address with your platform's. The assistant works in the domain you were signed in to when you authorized it.

  3. 3

    Check, then ask for something real

    First ask the assistant to test the connection: it has a check tool for exactly that and will answer within a second. Then start with a read — “how many learners are enrolled in the October session?” — before trusting it with a write. Trust is built in that order.

Mistral's Le Chat, or a client configured through a file? These clients connect with a personal token rather than through OAuth. Create it under Supervision → Integration → Personal tokens, with a name and a validity period: it starts with pt_ and is shown only once. In Le Chat, choose the HTTP Bearer Token authentication and paste it. In a configuration file, it goes in the authorization header:

"geduk": {
  "type": "streamable-http",
  "url": "https://app.geduk.fr/mcp",
  "headers": {
    "Authorization": "Bearer pt_your_token"
  }
}

A personal token carries your permissions for its whole lifetime: keep it short, and create one token per connected tool, so you can revoke one without cutting off the others.

Nothing to install, nothing to host. The server is part of the platform: it is already there, on the SaaS plan as on a dedicated instance, with no paid add-on. On a dedicated instance, the endpoint is on your own domain, and the traffic never leaves your hosting, except towards the AI provider you use.

To decide with your eyes open

What it does,
and what it does not

What the assistant does well

  • Opening sessions in bulk from a calendar
  • Enrolling a cohort from a pasted list, with no prior formatting
  • Finding a technical ID from a simple name
  • Answering a tracking question without opening the platform
  • Creating the skeleton of a module or a learning path to fill in later
  • Setting up notifications and labels, tedious screen by screen
  • Chaining operations that span several screens

What it does not replace

  • Instructional design: it executes, it does not decide
  • Fine work in the authoring tool, which remains visual
  • Human review before publishing to the catalogue
  • Approving a proof file before an audit
  • Repeated mass operations, which belong to an import or the API
  • The confidentiality you owe your learners: the connected account must stay proportionate to its use

There is another, quieter reason to value this feature: it forces the platform to expose its own operations cleanly. An LMS that an assistant can drive is an LMS whose API holds up. That is true for the assistant, true for your management software, and true for you.

FAQ

The MCP server, plainly

MCP, for Model Context Protocol, is an open standard that describes how an application exposes its operations to an AI assistant. The server, here Geduk, publishes a list of tools and what each one expects. The assistant reads that list and calls the ones that serve your request. Nothing is tied to one AI vendor: the same server answers any compatible client, which saves rewriting the integration each time you change tools.

No. It acts with the permissions of the person the token belongs to, never more. The tool list itself is filtered: a trainer without permission to edit users does not see the account creation tool, and a direct call would be refused. Privileges are re-read on every request, so a removed role stops applying immediately, without waiting for the conversation to end.

What the assistant reads in the platform goes through the provider you chose, like everything else you write to it. Geduk sends nothing on its own and adds no intermediary: the connection goes from your client to your platform. The choice of provider, and therefore of where the data goes, is yours. For use with sensitive data, connect the assistant with an account with limited permissions.

By a token carried in the authorization header. Four forms are accepted: an OAuth 2 token issued by the built-in authorization server, which is how Claude connects, a personal token created from the platform and revocable at any time, for Le Chat and clients configured through a file, an ordinary session token, and a service key for technical uses. The personal token is checked in the database on every call, so its revocation takes effect immediately instead of waiting for it to expire.

No. The server is built into the platform and its endpoint is your platform's address followed by /mcp. There is no software to deploy alongside it and no gateway to maintain. On the assistant's side, you declare an address and authorize the connection, which takes a few minutes.

Yes, on the SaaS plan as on a dedicated instance, with no paid add-on. On a dedicated instance, the endpoint is on your own domain: the traffic stays within your hosting, except for exchanges with the AI provider you use.

It exists and is not going anywhere. Geduk exposes a documented REST API, API keys and webhooks, as well as an OAuth 2 and OpenID Connect server to authenticate your own applications. The MCP server is aimed at conversational use and replaces nothing: for repeated, scheduled, high-volume processing, the API remains the right tool.

Try it on
your own platform.

Create a free space, connect your assistant, and ask it to open your first session.
No installation, no credit card, no commitment.

Try for free Request a demo

Included in every plan · Hosted in France · Real human support