MCP (Model Context Protocol) for AI Applications in Rwanda

Problem Statement

An organisation wants to implement an AI-powered assistant to help its IT security and governance teams access security policies, compliance documents, incident response procedures, and access control guidelines. These documents are distributed across multiple systems, making information retrieval time-consuming and inefficient.

Connecting AI applications directly to enterprise systems also introduces risks, including unauthorised data access, unsafe tool execution, prompt injection, and insufficient audit logging. The organisation needs a solution that enables AI applications to interact with approved enterprise resources while maintaining security, governance, and accountability.

How Did We Overcome the Challenges?

Learners designed an MCP-based solution that connects an AI assistant with approved IT security resources through a controlled integration layer.

• Secure Integration: Developed an MCP server exposing approved tools for searching security policies and retrieving governance documents.
• Access Control: Applied authentication, authorisation, and least-privilege principles to restrict access to permitted resources.
• Input Validation: Defined structured tool inputs and validated requests to reduce invalid or potentially harmful operations.
• Prompt Injection Protection: Treated retrieved documents and external tool outputs as untrusted content rather than privileged instructions.
• Audit Logging: Designed logs to record relevant tool invocations, timestamps, authorisation outcomes, and operation results while minimising sensitive information.
• Governance Controls: Restricted the prototype to read-only operations and identified actions requiring additional approval and human oversight.

Overall Implementation

Step 1: Requirement Analysis

Identified the need for a secure IT governance assistant that searches approved security policies, compliance guidelines, and incident response documentation without directly modifying production systems.

Step 2: MCP Architecture Design

Designed an architecture containing an AI application, MCP client, MCP server, and approved enterprise data sources. The MCP server acts as the integration layer, while underlying systems enforce their own access permissions.

Step 3: MCP Server Development

Created an MCP server using a supported SDK, with tools for searching security documents, retrieving authorised procedures, and accessing approved compliance information.

Step 4: Client Integration and Security

Connected the MCP client to the server and implemented appropriate authentication, permission checks, input validation, secure credential handling, and error management.

Step 5: Testing and Evaluation

Tested authorised requests, restricted document access, invalid inputs, prompt injection attempts, and unsupported tool calls. Reviewed application responses and audit records to identify security gaps.

Step 6: Documentation and Demonstration

Prepared architecture documentation, tool definitions, security test cases, access-control rules, and a working demonstration of the AI assistant retrieving approved security information.

Learner Outcomes and Evidence

Learners demonstrate their skills through:

• MCP Development: A working MCP server with defined tools, resources, and validated inputs.
• AI Integration: A demonstration of an AI application retrieving relevant IT security information through an MCP client.
• Security Implementation: Documented access controls, permission checks, and safe handling of tool requests.
• Testing Evidence: Test results covering unauthorised access, invalid inputs, prompt injection, and error handling.
• Governance Evidence: Sample audit logs, risk assessments, and security checklists with sensitive information removed.
• Project Demonstration: A secure, read-only IT security and governance assistant that retrieves approved documentation and presents relevant information.

Expected Learning Outcomes

By completing this project, learners develop practical knowledge of MCP architecture, client-server communication, tool integration, access control, secure AI workflows, and governance considerations. They learn to test MCP-based applications, document security requirements, and identify risks associated with connecting AI assistants to enterprise systems.

Final Project Outcome

The final deliverable is a Secure IT Security and Governance Assistant using MCP that helps authorised employees find approved security policies and compliance information efficiently. The project demonstrates how MCP can support structured AI-to-tool integration while maintaining appropriate access controls and auditability. The prototype is intended for educational purposes; production deployment requires formal security assessment, monitoring, identity integration, and organisational compliance approval.