Content Generation with Copilot Studio and MCP Servers

An international education company offering language courses in 50+ languages, intercultural training, leadership and soft-skills programs, with 20,000+ teachers all over the world.

New teachers joining the organisation faced significant challenges in navigating through the large datasets of educational materials while structuring their lessons. The onboarding process was also time-consuming and inefficient, hindering the ability to rapidly deploy qualified instructors.

The client requested a solution to help new teachers rapidly adapt to the educational system while providing easy access to the existing content base. The key desired features to implement included:

• supporting document retrieval and processing of materials in different sizes and formats (PDF and HTML),
• adaptability, allowing connections to both cloud services and on-premises servers,
• enabling content creation based on available materials,
• facilitating easy document retrieval and efficient processing of materials stored in both cloud and on-premises environments.

Implementing the solution required addressing several complex technical constraints:

• The system must handle limited and commonly unstructured inputs generated by users (typically onboarded teachers with limited knowledge of content structure).
• The system must process content of varying sizes, from 100 KB to up to 100 MB.
• The system must identify and utilise appropriate data sources from both cloud and on-premises infrastructure.
• The system must provide fine-grained control for searching specific materials across multiple organisational dimensions: courses, units, document types.
• The system must log certain endpoints and allow administrators to browse the history of prompts and interactions.

These requirements demanded a sophisticated architecture, capable of intelligently routing requests, managing large file operations, and maintaining comprehensive audit trails while maintaining responsiveness and user experience.

To address these challenges, the team developed a comprehensive solution built on Microsoft Copilot Studio and a custom MCP* server.

Custom MCP Server Development

A custom MCP server was created using Golang, providing a comprehensive set of tools and resources to bridge the Copilot Agent with the organisation's data sources.

Custom MCP Server available for Materials Agent

Copilot Agent Content Delivery

The system feeds content to the Copilot Agent in multiple formats, allowing to intelligently handle different file sizes. For smaller documents, content is directly passed to the agent. For larger files, instead of transmitting the full content, the system generates presigned URLs and provides them to the agent, reducing processing overhead and improving performance.

Copilot Agent Configuration

The team assisted with the creation and configuration of the Copilot Agent, tailoring it to interpret educational context and provide accurate support to teachers.

Activity Tracking and Logging

Amazon DynamoDB was integrated to track and log prompt activity, enabling comprehensive audit trails and allowing administrators to review interaction history.

Third-Party Power Platform Connector Tools

The solution utilised third-party Power Platform connectors, specifically the "Get S3 object content" tool, to integrate seamlessly with the organisation's existing AWS S3 infrastructure.

Power Platform connector “Get S3 object content” available as a tool for Materials Agent

Fine-Grained Search Capabilities

The advanced search functionality allows teachers to retrieve materials from both cloud and on-premises sources, filtering by course, unit, and document type.

Custom MCP Server tools available for Materials Agent to search and retrieve documents

Beyond basic search functionality, the system exposes material data as raw JSON via an MCP resource, enabling administrators to configure and deploy more sophisticated data retrieval mechanisms.

Custom MCP Server resource available for Materials Agent to handle advanced document search

The solution was successfully integrated within three months. It enables new teachers to drastically reduce the time and effort spent searching for class materials across large, distributed databases.

Teachers can now quickly turn existing materials into new instructional ideas, speeding up their workflow and increasing productivity.

The dialogue with AI

The AI-powered approach has transformed the teacher onboarding experience, enabling rapid familiarity with available content while supporting creative adaptation of materials to specific classroom needs.

By combining the intelligence of AI agents with robust back-end infrastructure for content management and retrieval, the solution provides an intuitive yet powerful platform for educational content discovery and generation.