Claude Code Desktop: From Code to Merge, All Inside One App

If you have ever written code, opened a browser to check if it works, switched back to your editor to fix something, know how exhausting that loop feels.

 For most developers, this back-and-forth is just accepted as part of the job. You write code in one place, preview it somewhere else, send it for review in another tool, and then wait for CI to pass before you can finally merge. 

Anthropic has been quietly building toward closing this loop, and in February 2026, they shipped a major update to Claude Code on desktop that brings the entire development cycle into a single place. You can now write code, see it running live, review your changes with AI-assisted inline comments, monitor your pull request status, and merge all without leaving the app.

In this article, we will walk through each of these new features in detail, what they do, how they work, and why they matter for anyone getting started with modern software development.

TL;DR 

1. Claude Code on desktop now lets you preview your running app inside the app itself no separate browser needed.
2. Claude watches your live app, reads console errors, and fixes issues automatically without you describing what’s wrong.
3. A “Review code” button lets Claude scan your local diffs and leave inline comments on real bugs before you push anything.
4. After you open a GitHub PR, Claude monitors CI checks, auto-fixes failures, and can auto-merge once everything passes.
5. Sessions are now portable  start in the CLI, continue on desktop, hand off to the web or your phone without losing context.
6. A permission mode system lets you control exactly how much Claude does on its own versus how much it asks you first.

What exactly is Claude Code on Desktop?

A graphical coding assistant that does the heavy lifting

• Claude Code is Anthropic’s AI-powered coding tool. While it started as a command-line interface (CLI) tool, the desktop app brings a graphical interface that makes it easier to use without needing to be comfortable in the terminal. You open the app, point it at a project folder, and start giving it instructions in plain English.
• The desktop version runs on macOS and Windows, and it gives you access to features like visual file diffs, session management, GitHub integration, and now  live app previews and automated PR workflows. It is available to all users with a Claude account, and you can download it from claude.com/download.

Seeing Your App Run Without Leaving the Editor

1. One of the most immediately useful additions is the live app preview. When you ask Claude to make changes to a web app, Claude Code on desktop can now spin up a development server and open an embedded browser right inside the app. You do not have to open a separate browser window, navigate to localhost, and manually describe what you see. Claude sees it too.
2. This matters more than it might sound at first. Previously, if Claude made a UI change and something looked off, you had to describe the problem back to it  “the button is misaligned” or “the text overflows on mobile.” 
3. Now, Claude takes screenshots of the running app, reads the console logs, inspects the DOM, and catches errors on its own. If something breaks, it iterates without you having to be the messenger.
4. You can also click on visual elements inside the preview panel and pass feedback directly to Claude. This makes the iteration loop much tighter. Instead of writing a paragraph explaining what is wrong, you click the element and describe what you want changed. Claude handles the rest.
5. The preview system is powered by a configuration file called .claude/launch.json, which Claude creates automatically based on your project setup. If your app uses a custom dev command or a non-standard port, you can edit this file manually. 
6. It supports multiple server configurations too, which is handy for monorepos where you might have a frontend and a backend running at the same time.

Reviewing Code Before It Leaves Your Machine

Once your changes look good in the preview, the next question is: are they actually correct? Even experienced developers ship bugs that feel obvious in hindsight. That is where the new code review feature comes in.

• Review Comments in Diff View

1. Inside the desktop app’s diff view  which shows you exactly what lines Claude changed  there is a “Review code” button. Clicking it asks Claude to look at those same diffs and leave comments directly in the view. 
2. The review is focused on things that actually matter: logic errors, compile errors, security vulnerabilities, and real bugs. It deliberately skips style issues, formatting problems, and things a linter would catch. The goal is not to nitpick  it is to catch the things that would cause real problems in production.

• Responding to Feedback

1. You can respond to any comment Claude leaves, ask it to address a specific concern, or tell it to make changes.
2.  After Claude acts on the feedback, a new diff appears for you to review again. It is a back-and-forth that mirrors how code reviews work with a human teammate, except it happens in seconds.

• Separate From Team Review

1. This feature is separate from the deeper agent-based Code Review system Anthropic announced for Team and Enterprise plans in March 2026. 
2. That system dispatches a team of agents on every GitHub pull request and is designed for organizations. The desktop review button described here is for individual developers reviewing their own local changes before they even push anything.

Monitoring Pull Requests While You Work on Something Else

• Once your code is pushed and a pull request is open on GitHub, the waiting starts. Will CI pass?  Normally, you would keep a browser tab open, refresh periodically, and come back to fix things when something fails.
• Claude Code on desktop now handles this for you. After you open a PR, a CI status bar appears inside the desktop session. Claude Code uses the GitHub CLI behind the scenes to poll your repository’s check results and surface failures in real time. You will also get a desktop notification when CI finishes  whether it passes or fails.

Auto-Fix and Auto-Merge

• Two toggles make this genuinely powerful. The first is auto-fix. When enabled, Claude automatically attempts to fix any CI check that fails. It reads the failure output, figures out what went wrong, makes the necessary changes, and pushes a fix. The second is auto-merge. 
• When enabled, Claude merges the pull request automatically once all checks pass. The merge method used is squash, which keeps the commit history clean.
• The practical workflow this unlocks is quite significant. You can start a task in a Claude Code session, open a PR, and then immediately move on to a different task in a new session. In the background, Claude is watching the original PR, fixing any failures it finds, and merging it when everything is green. By the time you check back, the work is done.
• One thing to note: auto-merge requires that your GitHub repository has the auto-merge feature enabled in its settings. 
• This is a GitHub-side configuration, not something Claude Code controls. Also, the GitHub CLI needs to be installed and authenticated on your machine. If it is not, the desktop app will prompt you to install it the first time you try to open a PR.

Moving Between Desktop, Web, and Mobile Without Losing Context

• One subtle but genuinely useful change is how sessions now travel with you. If you start a session in the CLI and want to move it into the desktop app, you can type /desktop in the terminal. Claude saves the full session context and opens it in the desktop app, then exits the CLI.
• You can also move a local desktop session to the cloud using a “Continue with Claude Code on the web” button. This pushes your current branch, creates a summary of the conversation, and starts a remote session with everything intact. 
• From there, you can pick it up from the Claude web interface or the Claude mobile app. This is useful for longer-running tasks where you want to start on your laptop and then monitor progress from your phone.
• Remote sessions run on Anthropic’s cloud infrastructure and keep running even if you close the app or your computer goes to sleep. That is a meaningful difference from local sessions, which are tied to your machine being on and the app being open.

To get started, download the Claude desktop app from claude.com/download, update to the latest version, and open the Code tab. The full documentation is available at code.claude.com/docs/en/desktop.

How Permission Modes Control What Claude Does Automatically

• All of these automated features, auto-fix, auto-merge, and preview verification operate within a permission system that you control. When starting a session, you choose a permission mode that determines how much Claude does on its own versus how much it asks for approval.
• In Ask permissions mode, Claude checks with you before editing files or running commands. This is recommended when you are just getting started. Auto-accept edits lets Claude change files freely, but still asks before running terminal commands. 
• Plan mode is useful for complex tasks where you want Claude to map out an approach before touching anything. There is also an Auto mode, currently in research preview for Team and Enterprise plans, that runs with minimal prompts while safety checks run in the background.
• The permission mode does not lock you in; you can switch it at any point during a session. Starting in Plan mode to review Claude’s approach, then switching to Auto 

If you’re learning how Claude Code can simplify modern development workflows, it’s always smart to review your diffs, understand your changes, and then let AI help you merge with confidence. Learn how AI agents are reshaping real‑world coding with HCL GUVI’s Artificial Intelligence & Machine Learning course for hands-on practice in AI‑powered code review, CI automation, and end‑to‑end development workflows!

CONCLUSION

If you are just coming out of college or working on your first real software projects, these updates make Claude Code a much more complete tool. The preview feature removes the need to juggle multiple windows. The code review button gives you a knowledgeable second pair of eyes on your changes before anything gets shared. The PR monitoring means you are not stuck babysitting CI runs.

Taken together, these features represent a meaningful step toward what Anthropic described as “closing the development loop.” The goal is to let developers spend less time on the operational overhead of coding the switching, the waiting, the checking  and more time on the parts that actually require thinking. Whether you are building something new or learning how professional development workflows operate, having a tool that handles this overhead makes a real difference.

FAQs

1. Do I need to pay for Claude to use these features?

The preview, code review button, and PR monitoring are available to all users. However, the Auto permission mode and the deeper agent-based Code Review (for GitHub PRs at the org level) require a Team or Enterprise plan.

2. Does Claude actually push code and merge PRs on its own?

Only if you turn on auto-fix and auto-merge. Both are opt-in toggles. By default, Claude monitors your PR and notifies you it does not take action unless you explicitly enable it.

3. What do I need installed for PR monitoring to work?

You need the GitHub CLI (gh) installed and authenticated on your machine. If it is not set up, the desktop app will prompt you to install it the first time you try to open a pull request.

4. Can I use Claude Code on desktop if I am already using the CLI?

Yes, both can run at the same time on the same project. They share the same configuration files and CLAUDE.md memory. You can even move a CLI session into the desktop app by typing /desktop in your terminal.

5. Is the live app preview available for all types of projects?

It works for frontend web apps and backend servers. Claude can start your dev server, open an embedded browser, test API endpoints, and read server logs. For non-standard setups, you can configure the server manually through the .claude/launch.json file in your project.