Agent-based Generative AI architectures are becoming increasingly popular, allowing the model to overcome its limitations by leveraging external smart tools that can perform specialized tasks. In this talk, we will start from the Microsoft.Extensions.AI library using local functions to create agentic applications with both offline and online models. We will then extend the application by leveraging the Model Context Protocol (MCP) to orchestrate multiple agents such as the Playwright MCP server which let the model acquire the knowledge of the web. We will see how orchestrate multiple MCP Servers to create powerful agentic applications and how to get the most out of Completion and Reasoning models.