In this article we present a 3-layer architecture (MAPIS3) that facilitates the integration of adaptive characteristics in an IMS-LD based computer-supported collaborative learning scenario (CSCL script), by making use of external and independent software components. The proposed architecture introduces a mediator component (MC) as a key element which mediates the communication of IMS-LD based representations (Learning Designs) with external services (e.g. a forum in a VLE, an agent or any software component that provides a specific functionality). Overall, the MAPIS3 architecture facilitates managing the complex task of controlling learning and data flow among these external services, through the MC and towards the IMS-LD module, which in turn acts as the orchestrator of the whole CSCL script. This work presents an example scenario, as a proof of concept, analyzing the architecture and demonstrating the flexibility of architecture implementation. In this scenario IMS-LD based modeling and Coppercore engine are used to implement adaptations by setting IMS-LD properties depending on input from three external tools: a) a Moodle questionnaire b) a group formation tool and c) a Moodle forum tool.