Abstract:

Current developments in generative artificial intelligence show that large language models (LLMs) are highly powerful tools capable of solving a wide range of natural language tasks. The educational process, including its preparation, undoubtedly falls into this category, and the possibilities for using AI in this area have already been discussed many times. However, significantly less attention has been paid to external tools that fundamentally extend LLMs' capabilities. Their common feature is typically a text-based input generated by the main LLM, which at the same time coordinates the entire solution to the assigned task, including selecting suitable tools and communicating with them.

At present, there is a wide range of such tools focused on specific IT activities – from working with program code, through information retrieval, to the creation of specialized outputs. Ongoing standardization in this area (e.g., Model Context Protocol – MCP) also enables users to create their own tools, which can be effectively utilized, for example, when designing solutions that support the educational process. When selecting appropriate tools as well as the main LLM, it is also necessary to consider additional factors, particularly the economics of operating the whole system (externally provided services paid periodically or according to usage versus local solutions operated on the user’s hardware) and data security (only local solutions guarantee full control over the software, the organization’s know-how, and above all, user data).

The paper also presents a practical, fully local, and comprehensive system based on LLM, with specifically designed MCP tools for supporting education, using exclusively proprietary components. The proposed solution covers the entire educational process – from the initial diagnosis of the student and assessment of his/her knowledge level in the given topic, through lesson planning and its delivery in various forms according to the student’s preferences (including the creation of study materials), to practical exercises, final knowledge assessment, and reflection. The system was designed to be universal; the educational topic is an input parameter when communicating with a specific student. Student information is also stored locally, and students can repeatedly return to their profiles. Thanks to the MCP standard, the system is open and allows further extension with both proprietary and externally provided tools. Practical experience with the deployment of this solution will also be presented.

Keywords:

LLM tools, AI in education, MCP, complex education system.