Claude Code in VS Code, Visual Studio and GitHub

What each option is, and which to pick

"Claude in your editor" can mean several different products. They share a name, but they run in different places and are paid for in different ways. Here is the short version.

If you live in Visual Studio on Windows rather than VS Code, read Chapter 5 first. Anthropic does not document a Visual Studio extension, so the choice there is different.

Installing Claude Code (CLI) and the VS Code extension

What you need

• A paid Claude subscription (Pro, Max, Team or Enterprise), a Claude Console account, or access through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry.
• For the extension: VS Code 1.94.0 or later.
• On native Windows, Anthropic recommends Git for Windows so Claude Code can use its Bash tool. Without it, Claude Code uses PowerShell instead. WSL does not need it.

Install the CLI

The native installer is the recommended route and updates itself in the background.

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell (prompt shows PS C:\)
irm https://claude.ai/install.ps1 | iex

# Windows CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

Package managers work too, but they do not auto-update:

brew install --cask claude-code        # then: brew upgrade claude-code
winget install Anthropic.ClaudeCode    # then: winget upgrade Anthropic.ClaudeCode

Open a new terminal and check it worked:

claude --version

It should print a version number followed by (Claude Code). If the shell says claude is not found, the install folder is not on your PATH yet (see Common problems).

Install the VS Code extension

1. In VS Code press Cmd+Shift+X (Mac) or Ctrl+Shift+X (Windows/Linux) to open Extensions.
2. Search for "Claude Code" and click Install. The extension ID is anthropic.claude-code.
3. If it does not appear, run Developer: Reload Window from the Command Palette.

Forks such as Cursor can install the same extension, and it is also on the Open VSX registry. The extension's version number matches the Claude Code version it bundles.

Signing in

In the extension: open the panel, click Sign in and finish in your browser. No API key is needed with a subscription or Console account. If you later see Not logged in · Please run /login, the sign-in screen reopens on its own; if it does not, reload the window.

In the CLI: run claude in a project folder and follow the browser prompts. Credentials are stored, so you only do this once. Type /login inside a session to switch accounts.

Account types the quickstart lists: Claude Pro, Max, Team or Enterprise (recommended); Claude Console (prepaid API credits; a "Claude Code" workspace is created for cost tracking); Bedrock, Agent Platform or Foundry; or your company's Claude apps gateway.

Cloud providers in VS Code: turn on the Disable Login Prompt setting, then configure the provider in ~/.claude/settings.json so the extension and CLI share it. Features that need a claude.ai account, such as plan usage bars and the Web tab for cloud sessions, are not available on a third-party provider.

Using the VS Code extension

Opening Claude

The Spark icon marks Claude Code throughout VS Code. Click it in the Editor Toolbar (top right, only when a file is open), in the Activity Bar on the left for the sessions list, or click ✻ Claude Code in the Status Bar, which works with no file open. You can also search "Claude Code" in the Command Palette. Drag the panel to the secondary sidebar, primary sidebar or editor area.

Inline diffs and reviewing changes

The permission mode is shown at the bottom of the prompt box. Click it to switch:

• Manual: Claude asks before file edits and most shell commands. Each edit opens as a side-by-side diff you can accept, reject, or edit before accepting.
• Edit automatically: Claude edits without asking.
• Auto: a classifier reviews most actions instead of asking you.
• Plan: Claude describes what it will do and waits for approval.

In a diff you can review one change at a time with Accept this change and Reject this change, or accept or reject the whole file. Diffs with more than 100 changes open without per-change buttons.

@-mentions and context

• Type @ and a file or folder name. Matching is fuzzy, so @auth finds AuthService.ts. Add a trailing slash for folders: @src/components/.
• Claude sees your selected text automatically. Press Option+K (Mac) or Alt+K (Windows/Linux) to insert a reference with line numbers, such as @app.ts#5-10.
• Reference a terminal's output with @terminal:name, where name is the terminal's title.
• Paste images into the prompt box. Hold Shift while dragging files in.
• Selections from files matched by files.exclude, search.exclude or .gitignore are withheld from the chat panel (the CLI in the terminal does not apply this filter).

Plan review

In Plan mode, VS Code opens the plan as a full Markdown document where you can add inline comments before Claude starts. You can also type /plan, or /plan fix the auth bug to start planning a specific task.

Shortcuts worth learning

Other useful pieces: Session history at the top of the panel to resume old conversations; hover a message to rewind code or fork the conversation from that point (checkpoints); /usage for account and usage; /mcp to manage MCP servers; /plugins for plugins.

Terminal or extension?

The extension offers a subset of commands. The docs list these differences:

To use the CLI without leaving VS Code, open the integrated terminal with Ctrl+` (Windows/Linux) or Cmd+` (Mac) and run claude. It connects to the editor for diffs and diagnostics automatically. From an external terminal, run /ide inside Claude Code to connect. Both share history: run claude --resume to pick up an extension conversation in the terminal. If you prefer the CLI look inside the panel, turn on the Use Terminal setting.

JetBrains users get a similar plugin (IntelliJ IDEA, PyCharm, WebStorm and others). Unlike the VS Code extension, it does not bundle the CLI, so install the CLI first.

Visual Studio (the full Windows IDE)

Visual Studio is not VS Code. Anthropic's Claude Code documentation covers VS Code, its forks, JetBrains IDEs, the desktop app and the terminal. It does not mention Visual Studio at all, and we found no Anthropic-documented Visual Studio extension. So you have two honest options.

Option A: run the Claude Code CLI in Visual Studio's terminal

Visual Studio has an integrated terminal: choose View > Terminal. It can host Developer PowerShell or Developer Command Prompt, and it opens in your solution's folder. Install Claude Code with the PowerShell or CMD command from Chapter 2, then run claude there. Claude will read and edit your files, run builds and use git like it would anywhere.

Option B: use Claude models inside GitHub Copilot for Visual Studio

If you already pay for GitHub Copilot, Copilot Chat in Visual Studio has a Model picker. GitHub's supported-models reference lists Claude models (for example Claude Haiku 4.5 and Claude Opus 4.8) as included in Visual Studio. What you actually see depends on your Copilot plan and Visual Studio version, and for Copilot Business and Enterprise an administrator must enable the relevant policy. This is Copilot's agent using a Claude model, not Claude Code. See Chapter 14.

Bring your own key: Microsoft's docs say that starting in Visual Studio 2026 version 18.10, BYOK is in preview and on by default in Community, Professional and Enterprise, with Anthropic as a supported provider. It works with the new Agent (Preview) experience. Steps: open Chat, open the Model picker, choose Add a model or Manage Models, connect your provider, and work in Agent (Preview). In the older BYOM experience (before 18.10), custom models were not available to Copilot Business or Enterprise users.

Our pick for Visual Studio users: run the Claude Code CLI in View > Terminal for agent work, and keep Copilot for inline completions.

CLAUDE.md and project settings

CLAUDE.md: standing instructions

Every session starts fresh. A CLAUDE.md file is plain Markdown that Claude reads at the start of each session: build commands, conventions, "always do X" rules. Run /init to generate a first draft from your codebase.

• Keep each file under about 200 lines. Be concrete: "Run npm test before committing" beats "test your changes".
• Import other files with @path/to/file.
• Split big rule sets into .claude/rules/*.md, optionally scoped to paths with paths frontmatter.
• If your repo already has an AGENTS.md and no CLAUDE.md, Claude reads AGENTS.md.
• Run /context to see which memory files loaded, and /memory to open them or toggle auto memory (notes Claude saves for itself).
• CLAUDE.md is guidance, not enforcement. For something that must always happen, use a hook.

In the VS Code extension, the command menu's Instructions entry opens your CLAUDE.md files, and Memory manages auto memory.

settings.json: behavior and permissions

Precedence, highest first: managed, command line, project local, shared project, user. Files are strict JSON (no comments or trailing commas). Add the schema line for autocomplete in VS Code. This example, from the docs, lets Claude run lint and tests without asking and blocks it from reading .env files:

{
  "$schema": "https://json.schemastore.org/claude-code-settings.json",
  "permissions": {
    "allow": ["Bash(npm run lint)", "Bash(npm run test *)"],
    "deny":  ["Read(./.env)", "Read(./.env.*)"]
  }
}

Run /status to confirm which settings files loaded. The VS Code extension also has its own settings under Extensions > Claude Code (for example useTerminal, initialPermissionMode, preferredLocation); those control the editor UI, while settings.json is shared with the CLI. Note that /config is a terminal feature; the VS Code chat panel does not open it.

Git basics with Claude: commits, branches, PRs

You do not need to learn special commands. Ask in plain English, and Claude runs git for you:

what files have I changed?
commit my changes with a descriptive message
create a new branch called feature/quickstart
show me the last 5 commits
help me resolve merge conflicts
create a pr for this feature

Pull requests use the GitHub CLI. Install gh and run gh auth login once. Claude then opens PRs with gh pr create and writes the description from the actual changes. Ask it to "enhance the PR description" or "highlight potential risks" before you submit.

To find the session that made a PR later, run claude --from-pr 1234 with your PR number, or paste the PR URL into the /resume picker.

Worktrees for parallel sessions

A git worktree is a second working folder on its own branch that shares the same repository history. Give each Claude session its own worktree and two sessions can edit at once without stepping on each other.

claude --worktree feature-auth     # or -w; creates .claude/worktrees/feature-auth on branch worktree-feature-auth
claude --worktree bugfix-123       # in a second terminal: a separate, isolated session
claude --worktree "#1234"          # start from pull request 1234 (quote the #)

• Add .claude/worktrees/ to your .gitignore.
• A worktree is a fresh checkout: install dependencies there. To copy gitignored files such as .env into every new worktree, list them in a .worktreeinclude file at the project root.
• New worktrees branch from your default branch. To branch from your current work instead, set "worktree": {"baseRef": "head"} in settings.
• When you exit, Claude removes a clean worktree automatically, or asks whether to keep one that has work in it.
• You can also just ask Claude to "work in a worktree" mid-session, or to "use worktrees for your agents".
• The repo needs at least one commit first.

Prefer plain git? git worktree add ../project-feature-a -b feature-a, then cd there and run claude. In the desktop app, turn on Worktree in the project folder control before you start a session. In VS Code, open extra conversations with Open in New Tab; the docs point you to --worktree when you want them isolated.

The Claude GitHub App

One Claude GitHub App powers every Claude feature that talks to GitHub: the GitHub Action, Code Review, auto-fix in cloud sessions, and access to private repos from Claude Code on the web.

When you install it, GitHub makes you accept its whole permission set (you cannot pick a subset). It includes read and write on Contents, Issues, Pull requests, Actions, Checks, Discussions, Repository hooks and Workflows, plus read on Administration, Members, Merge queues, Metadata and Statuses. If your organization wants less, the docs describe making a custom app with only Contents, Issues and Pull requests, but that covers only the GitHub Action; Code Review and web auto-fix still need the official app.

GitHub Actions: mention @claude

The Claude Code GitHub Action (anthropics/claude-code-action) runs Claude Code inside your repository's workflows. Mention @claude in an issue or PR comment and it can answer, change code and push commits.

Quick setup (recommended)

1. Install the GitHub CLI and run gh auth login.
2. In the repository, start claude and run /install-github-app. You need admin access, and it works only with github.com repos.
3. It installs the app, saves a secret (ANTHROPIC_API_KEY for an API key, or CLAUDE_CODE_OAUTH_TOKEN for a subscription token), pushes a branch with the workflow files and opens a PR. Merge it.

Manual setup: install the app, add the secret yourself (generate a subscription token with claude setup-token), and copy examples/claude.yml from the action's repository into .github/workflows/. The core of the official example:

name: Claude Code
on:
  issue_comment:
    types: [created]
  pull_request_review_comment:
    types: [created]
jobs:
  claude:
    if: contains(github.event.comment.body, '@claude')
    runs-on: ubuntu-latest
    permissions:
      contents: write
      pull-requests: write
      issues: write
      id-token: write
      actions: read
    steps:
      - uses: actions/checkout@v6
        with:
          fetch-depth: 1
      - uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}

Then comment, for example: @claude fix the TypeError in the user dashboard component.

• Two modes. With no prompt input, Claude waits for the trigger phrase. With a prompt, it runs automatically on any event, including a cron schedule.
• Who can trigger it. The commenter needs write access, and bots are rejected unless listed in allowed_bots.
• Tuning. Pass CLI flags through claude_args, such as --max-turns, --model and --allowedTools.
• Costs. Each run uses GitHub Actions minutes plus API tokens (or your subscription if you use the OAuth token). Cap work with --max-turns and workflow timeouts.
• Old workflows. If yours say @beta, change to @v1, remove mode, and rename direct_prompt to prompt.
• Put project rules in CLAUDE.md; the action follows it. Never commit keys; use GitHub Secrets.

Code Review and REVIEW.md

Code Review is a managed service, separate from the GitHub Action. Several agents read each pull request, a verification step filters out false positives, and Claude posts inline comments ranked by severity, a summary, and a check run named Claude Code Review. It never approves or blocks a PR; the check never concludes as a failure.

• Who can use it: Team and Enterprise plans with GitHub repos (including GitHub Enterprise Server). Not available with Zero Data Retention, the HIPAA configuration, or customer-managed encryption keys.
• Setup: an Owner of your Claude organization turns on Allow code review in Admin settings > Claude Code, connects GitHub, installs the app and picks repositories. No workflow files.
• Levels: Quick, Standard (default) or Deep. Deeper costs more.
• When it runs, per repo: Once after PR creation, After every push (the default for new repos), or Manual.

Severity markers

🔴 Important: a bug to fix before merging. 🟡 Nit: minor, not blocking. 🟣 Pre-existing: a bug already in the codebase, not introduced by this PR.

Comment commands (top-level PR comment, write access needed)

If the repo also runs the GitHub Action with its default @claude trigger, @claude review starts the Action too. Set the Action's trigger_phrase to something else to keep them apart.

REVIEW.md

Code Review reads your CLAUDE.md files as context. For review-only rules, add a REVIEW.md at the repo root: what counts as Important, a nit cap, paths to skip, checks to always run. The docs' example includes lines like "Report at most five Nits per review" and "Generated files under src/gen/ and any *.lock file" under a "Do not report" heading.

/code-review on your own machine

No Team plan? /code-review works on any plan, inside your own session, and counts toward your normal usage. It runs as a background subagent and reports correctness bugs in your conversation. /review is an alias.

It follows your CLAUDE.md but does not read REVIEW.md, which is only for the managed service. Put rules you want in both places in CLAUDE.md. For a deeper cloud review, /code-review ultra runs ultrareview, which needs a claude.ai sign-in. In the desktop app, findings appear as a Code review card with Apply fixes.

Claude Code on the web with GitHub repos

A cloud session runs Claude Code on a cloud VM (Anthropic-managed by default). Claude clones your GitHub repo, works, and pushes a branch. It keeps going after you close the laptop, and counts toward your plan's usage with no separate VM charge. Available on Pro, Max and Team, and to Enterprise users with premium or Chat + Claude Code seats.

1. Go to claude.ai/code, sign in, and connect GitHub when prompted.
2. Public repos work right away. For private repos, install the Claude GitHub App on the account or organization that owns them.
3. Pick a repo and branch, choose a permission mode (Auto, Accept edits or Plan; cloud sessions have no Manual mode), describe the task and press Enter.
4. Open the diff from the Changes card, leave inline comments, then click Create PR.

From your terminal: if you use gh, run /web-setup inside Claude Code to send your gh token to your Claude account instead (on Team and Enterprise, an Owner must turn on Quick setup first). Then start cloud tasks with claude --cloud "Fix the flaky test in auth.spec.ts", and pull one back to your machine with claude --teleport.

Auto-fix: with the GitHub App installed, Claude can watch a PR and push fixes for CI failures and review comments. Turn it on from the PR's ⋮ menu in the Changes card, or run /autofix-pr on the PR's branch. Its replies on GitHub appear under your username, labeled as Claude Code, which can trigger comment-driven automation, so check that first.

In VS Code: signed in with a claude.ai subscription, open Session history and choose the Web tab to resume a cloud session locally. It downloads a copy; changes do not sync back.

Claude in GitHub Copilot

GitHub Copilot can use Anthropic's models. This is a GitHub product, billed through your Copilot plan, and it is not Claude Code. Two parts matter:

Claude models in the Copilot model picker

GitHub's supported models page lists Claude Haiku 4.5 and 5.5, Opus 4.8, 5 and 5.5, Sonnet 4.6, 5 and 5.5, and Fable 5 and 5.1. Which you see depends on your plan and client (GitHub.com, Copilot CLI, VS Code, Visual Studio, JetBrains, Xcode, Eclipse). GitHub says administrators must enable each Fable model, and that Anthropic retains prompts and outputs for those two models by default. Choosing a larger context window or higher reasoning level uses more AI credits.

The Claude coding agent on GitHub (public preview)

GitHub's third-party coding agents let you hand an issue or prompt to Anthropic's Claude, which works on its own and opens a pull request for your review. GitHub says it is built on the Claude Agent SDK.

• Plans: Copilot Pro, Pro+, Max, Business and Enterprise.
• Turn it on: in your user or organization Copilot cloud agent settings, enable Allow Claude coding agent. This installs a GitHub App called anthropic code agent.
• Start a task: from the Agents tab, by assigning it to an issue, by mentioning it in a PR comment, from GitHub Mobile, or from the chat view in VS Code.
• Models listed: Auto, Claude Sonnet 4.6, Claude Sonnet 5, Claude Opus 4.8, Claude Opus 4.8 (fast mode, preview), Claude Opus 5 and Claude Fable 5.1.
• Cost: sessions use GitHub Actions minutes and AI credits.

In Visual Studio, Copilot's own cloud agent sessions start from the agent picker (version 18.1 or later). Microsoft's Visual Studio agents page does not mention the Claude coding agent specifically, so we can't confirm it appears there.

MCP with GitHub

MCP (Model Context Protocol) servers give Claude extra tools. GitHub runs a remote MCP server that authenticates with a fine-grained personal access token. Create one in your GitHub token settings, then run this in a terminal (not inside a session):

claude mcp add --transport http github https://api.githubcopilot.com/mcp/ \
  --header "Authorization: Bearer YOUR_GITHUB_PAT"

The command does not check the token, so a typo only shows up later. Start a session, type /mcp, and confirm the server shows as connected. Then ask things like "Review PR #456 and suggest improvements" or "Show me all open PRs assigned to me".

• Scopes: the default is local (you, this project). Add --scope user for all your projects, or --scope project to write a shared .mcp.json you can commit. Never commit the token itself.
• In VS Code: type /mcp in the chat panel to add, remove, enable or reconnect servers. It saves to the same config as claude mcp add.
• Check status: claude mcp list. Remove with claude mcp remove github.
• Each server uses some context in every session, so remove ones you don't use.

Many GitHub tasks don't need MCP at all: Claude can already use gh from its shell.

Common problems and fixes

Short glossary

CLI

Command-line interface. Claude Code in a terminal, started with claude.

Extension

The Claude Code add-on for VS Code that gives a chat panel.

Session

One conversation with its own history and context.

Permission mode

How much Claude may do without asking: Manual, Edit automatically / Accept edits, Plan, Auto.

CLAUDE.md

A Markdown file of standing instructions Claude reads every session.

REVIEW.md

Review-only instructions read by the managed Code Review service.

Worktree

A separate git working folder on its own branch, sharing one repository.

Pull request (PR)

A proposed set of changes on GitHub for others to review before merging.

gh

GitHub's official command-line tool. Claude uses it to open PRs.

GitHub App

An installable integration with its own permissions. The Claude GitHub App powers Actions, Code Review and auto-fix.

GitHub Actions

GitHub's built-in automation that runs workflows on events such as comments or pushes.

MCP

Model Context Protocol, a standard way to plug tools and data sources into Claude.

Cloud session

A Claude Code session running on a cloud VM rather than your computer.

BYOK

Bring your own key: using your own model provider account inside another tool, such as Visual Studio.

Copilot

GitHub's AI assistant, which can use Claude models but is a separate product.

Sources

Everything above was checked against these pages in October 2026. Products change quickly; when in doubt, the docs win.

Anthropic (code.claude.com)

1. Use Claude Code in VS Code
2. Quickstart
3. Platforms and integrations
4. JetBrains IDEs
5. Desktop application
6. How Claude remembers your project (CLAUDE.md)
7. Settings files and precedence
8. Common workflows
9. Run parallel sessions with worktrees
10. Claude Code GitHub Actions
11. Code Review
12. Set up Code Review
13. Customize Code Review
14. Read and request reviews on a pull request
15. Review locally with /code-review
16. Use Claude Code in the cloud
17. Get started with Claude Code in the cloud
18. MCP quickstart
19. Connect Claude Code to tools via MCP

GitHub (docs.github.com)

1. Supported AI models in GitHub Copilot
2. Third-party coding agents (Anthropic Claude)

Microsoft (learn.microsoft.com)

1. Using AI models in Copilot Chat (Visual Studio)
2. Use GitHub Copilot agents in Visual Studio
3. Command-line shells in Visual Studio (integrated terminal)