The Learn:Up MCP Server lets you connect your AI assistant to Learn:Up and ask questions about your learning data in plain English. This article explains what it can and cannot access, how access is controlled, and what to share with your security team.
What the MCP Server can see
Once connected, your assistant can read:
People — learners, user groups, and manager relationships
Content — courses, learning paths, SCORM packages, certifications, and the steps inside a course
Progress — assignments, completions, overdue training, and expiring certifications
Live training — event schedules, rosters, and attendance
Setup — webhook configuration and API key metadata
It works through a fixed set of 18 defined tools. There is no open-ended query access and no direct access to the underlying database.
What it cannot do
The MCP Server is read-only. It cannot:
Create, edit or delete any content
Assign or unassign training
Register, invite or remove anyone from an event
Grade or review a submission
Send reminders or change a due date
Change any user, group or permission setting
It answers questions. You still make the changes yourself in the product.
What's coming. Create, update and edit are planned for a future release. Delete will come after that. We're sequencing it this way on purpose — the guardrails that stop an AI assistant removing large amounts of your data need to be in place first.
How access is controlled
You only ever see what you could already see.
Every request runs as you. It goes through exactly the same permission checks as if you'd clicked through the Learn:Up interface yourself. A manager sees their own team. A learner sees their own record. An administrator sees what their admin role allows.
Connecting an AI assistant does not widen anyone's access. It only changes how they ask.
Who can connect at launch. Employee Learning administrators, Manager and learners. Academy (Customer Learning) administrator access, will follow in later releases.
How the connection works
Connection uses OAuth 2.1 — the same standard behind "Sign in with…" buttons across the web — and happens in two stages.
Stage one, once per organisation. An administrator in your AI assistant's workspace adds Learn:Up as a custom connector, using your organisation's Learn:Up MCP server URL. This makes it available to your people. It doesn't connect anyone.
Stage two, once per person. Each individual adds the connector for themselves. That sends them to Learn:Up to sign in with their normal credentials or through your SSO, shows them a consent screen describing the access being granted, and binds the connection to their own account.
Stage two is the important one for security: there is no shared or service connection. Every connection belongs to a named individual and carries their permissions. What a person can even see differs by role — an administrator is offered the full set of tools, a learner a much smaller set scoped to their own record.
📌 Note: You'll need a paid Team or Enterprise plan on your AI assistant. Free and personal tiers can't add custom connectors.
Your password is never shared with the assistant, and the assistant never stores your Learn:Up credentials.
To disconnect, remove the connector in your assistant. We are working on adding controls within Learn:up so admins can revoke the connection.
Which AI assistants work with this
MCP is an open standard, so any assistant that supports it can connect.
We actively test and support Claude and ChatGPT — those are the two where we guarantee everything works correctly. Other MCP-compatible assistants will generally work, but we don't test them.
A common question: this isn't tied to a particular AI model. MCP connects applications to assistants; it doesn't integrate with GPT, Claude, Gemini or any other model individually. Your choice of model is yours, and it doesn't change what the Learn:Up MCP Server does.
Where your data goes
The MCP Server runs inside Learn:Up itself. It isn't a separate service, and it doesn't introduce a new data store or a new third party into your setup.
When you ask a question, your assistant calls Learn:Up, Learn:Up returns the answer, and the answer appears in your assistant. From that point, the data sits in your AI assistant's conversation — governed by your own agreement with that AI provider.
Learn:Up does not use anything retrieved through the MCP Server to train AI models.
What to give your security team
Ask your Customer Success Manager for the Learn:Up MCP Server Security FAQ. It covers authentication, token handling, permission scoping, logging, data residency and compliance in the detail a security review needs.