A Hands-on Learning Path

A structured progression can move a learner from using AI models to building custom MCP servers.

Begin with the AI Toolkit for Visual Studio Code. Explore the model catalogue, use the playground to test prompts, and change parameters such as temperature, maximum tokens, and top-p. Use the agent builder to define an agent's role, behavior, parameters, and tools.

Use an existing MCP server from the available catalogue. A browser-automation project can use the Playwright MCP server. The agent can open pages, click controls, extract content, capture images, and execute testing flows. The server and its tool capabilities are selected in the agent builder, and prompts are written to guide the automation task.

Use a Python template to create a weather MCP server that responds to natural-language questions such as asking for the weather in a city. Use the MCP SDK, configure debugging with MCP Inspector, and run the server beside the agent in Visual Studio Code.

This stage includes learning the project structure, updating dependencies, creating launch configurations and background tasks, and testing the server with both the agent builder and the Inspector.

Create a GitHub-clone MCP server that automates common development steps: cloning a repository, creating directories, and opening the project in Visual Studio Code. Add validation and error handling, include operating-system-aware logic for launching Visual Studio Code or Visual Studio Code Insiders, and connect the workflow to GitHub Copilot agent mode.

Create a spreadsheet server that lists, reads, writes, and appends data. Add a read-only sheet resource. Add prompt templates for analysis, report creation, and form-to-sheet workflows. Connect the server to more than one compatible host.

Choose a workflow and identify the tools, resources, and prompts it needs. Use an official SDK, test the server with MCP Inspector, deploy it locally and in the cloud, place an API-management layer in front of it, and apply identity, rate limits, monitoring, and security controls.

Additional project directions include a multi-provider server that routes requests between model providers, an enterprise prompt-management system that versions and approves prompts, and a content-generation platform with tracking and review workflows.