VS Code Learn: Extending Agents

Liam Conroy-Hampton walks through how to extend AI agents in Visual Studio Code, focusing on how agents use tools safely, how MCP connects agents to external data, and how Agent Plugins package tools for easy installation.

Overview

AI agents, tools, and tool sets

The video introduces key building blocks for agent extensibility:

• Tools: discrete capabilities an agent can invoke.
• Tool sets: collections of tools grouped for a particular workflow.
• Permissions: controls that govern what tools can do.
• Sandboxing: isolation boundaries intended to reduce risk when tools run.

Demo: extending an agent with tools

A hands-on demo shows how tools and tool sets are used in practice, including how permissions and sandboxing affect what the agent is allowed to access and execute.

MCP (Model Context Protocol)

The video introduces MCP (Model Context Protocol) and demonstrates how it can be used to connect agents to external data sources, enabling the agent to retrieve context beyond what is available locally.

Agent Plugins

The video covers Agent Plugins, describing them as a way to:

• Package tools so they can be installed with a one-click install experience.
• Share and reuse tool capabilities across environments.

Third-party agents in VS Code

The video closes by showing how VS Code covers the broader agent landscape, including support for third-party agents such as:

• Claude
• Codex

Links

• VS Code Learn playlist: https://aka.ms/vscode/learn
• Liam Conroy-Hampton (LinkedIn): https://www.linkedin.com/in/liam-conroy-hampton/
• VS Code on GitHub: https://github.com/microsoft/vscode
• VS Code on X: https://x.com/code
• VS Code on Bluesky: https://bsky.app/profile/vscode.dev
• VS Code on Instagram: https://www.instagram.com/vscode.ig/