# Visual Studio Code

> Microsoft's editor, with chat, agents, inline suggestions and MCP built into the product

- Category: AI Coding / IDE Copilots
- Language: TypeScript
- Licence: MIT (Code - OSS)
- Repository: https://github.com/microsoft/vscode
- Website: https://code.visualstudio.com
- Canonical page: https://ai-tldr.dev/tools/vscode/
- Updated: 2026-05-03

## In short

Visual Studio Code is Microsoft's code editor, developed in the open as "Code - OSS" under the MIT licence. The product you download from code.visualstudio.com is a distribution of that repository with Microsoft-specific customisations released under a Microsoft product licence. It sits in the IDE Copilots category of the AI Coding tools and is open-source, written in TypeScript and licensed MIT (Code - OSS).

## Overview

Visual Studio Code is Microsoft's code editor, developed in the open as "Code - OSS" under the MIT licence. The product you download from code.visualstudio.com is a distribution of that repository with Microsoft-specific customisations released under a Microsoft product licence. It combines editing, navigation, lightweight debugging and an extension model, and ships a monthly release.

The AI features are part of the editor rather than a separate product. VS Code's documentation groups them as agents (the most autonomous option, for multi-step tasks), chat and inline chat for conversational help, inline suggestions for completions, and smart actions for refactoring. Agents are gated behind the `chat.agent.enabled` setting.

Two things make it a platform rather than a single assistant. Custom instructions let you shape how the agent behaves in a given repository, and MCP server support lets it reach tools and data outside the editor. Model choice is open: you sign in with a GitHub account to use a GitHub Copilot subscription, or bring your own models — including ones running locally.

## What it does

- Agents for multi-step tasks, plus chat and inline chat, built into the editor
- Inline suggestions and smart actions for completion and refactoring
- Custom instructions to shape agent behaviour per project
- MCP server support, so the agent can call tools and data sources outside the editor
- Model choice: a GitHub Copilot subscription, other providers, or bring-your-own including local models
- Core editor developed in the open at microsoft/vscode under MIT, with a large extension marketplace

## Getting started

Download the product build rather than compiling the repository — the AI features are enabled from inside the editor.

1. Install VS Code

   Builds for Windows, macOS and Linux are on the download page; the Insiders build ships daily.

```text
https://code.visualstudio.com/Download
```

2. Pick where the models come from

   Sign in with your GitHub account to use a GitHub Copilot subscription, or configure your own model provider (BYOK), including a locally-run model.

3. Turn agents on

   Agent mode is behind a setting. Open Settings and enable it, then open the Chat view and pick the agent.

```text
chat.agent.enabled
```

4. Give the agent project context

   Add custom instructions so the agent follows your repository's conventions, and register MCP servers to give it tools beyond the workspace.

5. Build from source (optional)

   The Code - OSS repository builds on Linux, macOS and Windows; a dev container is included and wants at least 4 cores and 6 GB of RAM.

```bash
git clone https://github.com/microsoft/vscode.git
```

## When to use it

- Use one editor for both hand-written and agent-written code, without leaving for a separate tool
- Standardise how an agent behaves across a team by checking custom instructions into the repository
- Connect an in-editor agent to internal systems through MCP servers
- Keep code and prompts on your own model provider, or a local model, instead of a hosted subscription

## Version history

- 2026-05-03: Microsoft merged PR 313931, changing the default for the `git.addAICoAuthor` setting back to off after a 1,220-point Hacker News thread reported Copilot co-author trailers landing on commits even where users had disabled AI features. The default had been turned on in PR 310226 on 2026-04-16.
