Inclusive education requires learning environments that accommodate diverse cognitive profiles, including students with slow learning who often experience working memory overload when processing abstract mathematical symbols and procedures. Empirical evidence indicates that slow-learning students face persistent difficulties in identifying patterns, reasoning logically, drawing conclusions, and transferring mathematical concepts to new contexts. These challenges highlight the urgent need for adaptive learning innovations that can provide individualized support and real-time assessment to support the development of meaningful mathematical reasoning. This study aims to design and validate the MAGENTA AI Robot Learning Ecosystem as an adaptive support model for inclusive mathematics learning. This research uses a development approach with a 4D model (Define, Design, Develop, Deploy), which integrates needs assessment, prototype development, expert validation, revision, and limited field trials. Data were collected through classroom observations, teacher and student questionnaires, initial reasoning tests, expert assessment validation sheets, and a practicality response instrument. The MAGENTA ecosystem integrates a physical robot diorama (a line-tracing robot) with a tablet-based AI platform that provides multimodal, game-based assessment and three levels of adaptive scaffolding: problem interpretation (yellow), step-by-step solution guidance (red), and formula/tool ​​support (blue). The system also incorporates a cognitive pause mechanism that automatically pauses the robot when the scaffolding is accessed, supporting cognitive load management. Findings indicate strong stakeholder demand for contextual visualization media (93%), adaptive scaffolding (89%), and an automated feedback system (85%). Expert validation results confirmed that the MAGENTA model achieved a high level of validity (mean score of 3.64/4), while the modular teaching materials were also categorized as valid (mean score of 3.55/4). Practicality evaluations demonstrated very high usability based on teacher feedback (mean score of 133.67, categorized as "very practical") and positive student feedback (mean score of 62.96, categorized as "practical"). These results demonstrate the viability of MAGENTA as an inclusive learning ecosystem that supports mathematical reasoning through adaptive scaffolding and real-time assessment. This study suggests that integrating AI-based scaffolding with embodied robotic interactions can enhance deep learning opportunities in inclusive classrooms by reducing cognitive load, strengthening reasoning processes, and supporting differentiated instruction. MAGENTA offers a scalable framework for future AI-robotic learning environments in mathematics education.