Creative Learning

What it does and what makes it unique

Creative Learning MCP is an MCP server for Creative Learning, a multi-tenant LMS used in Portugal by training entities, companies, trainers, and trainees. It exposes the existing REST API as tools that assistants like Cursor, Claude, or Grok can call directly.

Each user signs in with their own Creative Learning account through OAuth: Cursor or Claude opens the platform's login and consent page, and access can be revoked at any time under "Aplicações ligadas". A personal access token is available as a fallback. The API remains the permission boundary, so the assistant only sees what the user can already see. Personal data is filtered out before it reaches the assistant, and the platform has zero AI costs, because the model belongs to the user.

The available tools are:

• list_courses: the courses the user can see.
• get_course: the details and modules of one course.
• get_training_plan: a course's training plan and schedule.
• list_trainees: the trainees in a course, showing only name, status and progress (company, trainer and training entity).
• get_trainee_progress: one trainee's lesson and quiz progress.
• get_training_hours_summary: the hours each trainee has completed per course (company, trainer and training entity).
• list_at_risk_trainees: trainees at risk of dropping out, meaning 14 or more days inactive, more than 20 points behind, or 2 or more failed quizzes.
• create_session_plans_draft: drafts in-person session plans for a trainer, who reviews and confirms them in the platform before anything is saved.

The role of AI

AI becomes the platform's interface. Instead of clicking through dashboards, a trainer can ask "Which trainees are at risk in my course?" and a company can ask "How many training hours has each employee completed?" The assistant picks and combines the right tools, summarizes the results, and proposes next steps. AI also does work, not just reporting. It flags trainees at risk of dropping out (inactive, falling behind, or failing quizzes) so trainers can act early, and it drafts in-person session plans. The only write action creates a draft that the trainer reviews and confirms in the platform before anything is saved. Every request runs with the user's own permissions, so anything outside their scope is refused by the API itself.

The whole project was built with AI: all development was done with Cursor and Cursor Cloud Agents, from the MCP server to the OAuth login and the in-app "Usar com IA" guide.

Technologies

TypeScript, Node 24, MCP SDK, OAuth 2.1 with PKCE, Express, WordPress REST API, JWT, React, Cursor, and Cursor Cloud Agents.