Why this is in the programme
Useful enterprise AI systems need access to capabilities outside the model itself, but giving a probabilistic system access to tools and data immediately raises questions about interfaces, permissions and control.
Model Context Protocol provides one emerging approach to that problem.
Course
Build AI Agents using MCP — IBM
The course covers MCP architecture, servers and clients, tools, resources and prompts, as well as interaction with external capabilities and the role of permissions and human approval.
Role in the experiment
The aim is to understand MCP as an engineering mechanism rather than simply learn its API.
Particular attention will be paid to how tool boundaries, least privilege, approval and provenance might apply when an AI system operates in a security-sensitive enterprise environment.
This page will develop alongside the practical work.