Claude Code Right in Your Browser and Other Major Updates

Claude Code Right in Your Browser and Other Major Updates

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
22. 10. 2025
2 minutes reading · 13 views
Claude Code Right in Your Browser and Other Major Updates
Anthropic has just launched Claude Code on the web, allowing developers and everyday users to run coding tasks directly from their browsers. This new feature, available in beta as of October 20, 2025, runs on cloud infrastructure managed by Anthropic. Users can connect their GitHub repositories, describe a task, and let Claude handle the implementation. Each session runs in an isolated environment with real-time progress tracking, where you can provide Claude with additional instructions while it works. This means you can run several bug fixes simultaneously, work on repositories that are not on your laptop, or even code from an iPhone using the Claude iOS app.

Simplified and secure access control

The main innovation is a new sandboxing system that reduces the number of approval prompts by 84%. Instead of constantly approving every action, such as running a command, editing a file, or testing, you define the boundaries in advance. Claude can access only the current project directory and cannot access sensitive system files such as SSH keys or .bashrc. Network isolation allows connections only to approved domains—if someone tries to trick Claude into connecting to a hacker-controlled server, the connection fails at the operating system level. If Claude attempts to access anything outside the sandbox, you receive an immediate notification and can allow it once or adjust the settings.

This approach addresses the dilemma faced by AI coding tools: you want the AI to help by accessing files and the terminal, but without risks such as prompt injection, where hackers trick the AI into performing malicious actions. In Anthropic's internal testing, this not only simplified workflows but also improved security by preventing approval fatigue and the risk of "dangerously bypassing permissions." Each web session runs in a separate secure sandbox where GitHub credentials never enter the environment—instead, Claude uses its own proxy, which verifies every Git operation. Even if the code in the sandbox were compromised, your actual credentials would remain secure.

Claude Code on the web

Availability and additional details

Claude Code on the web is available to users with a Pro subscription for $20 per month (approximately CZK 460) and a Max subscription for $100 or $200 per month (approximately CZK 2,300 or CZK 4,600). Usage has reportedly increased tenfold since May 2025, generating more than $500 million annually for Anthropic (approximately CZK 11.5 billion). Anthropic has openly shared the sandboxing code, allowing other AI tools to adopt it as well. It also integrates with Claude Skills, which Simon Willison says turn Claude into an ideal general-purpose agent, and with McKay Wrigley's tutorial on using Claude Code as a general-purpose agent. To get started, visit claude.com/code, where you can connect your first repository and begin delegating tasks.

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

OpenAI gives Codex reusable cloud workspaces accessible from any deviceOpenAI gives Codex reusable cloud workspaces accessible from any device
Codex gains reusable cloud development environments, alongside voice controls in its CLI, code reviews in the ChatGPT desktop app and cloud-based security tools.
2 min read
2. 10. 2026
Amazon releases Strands Decider 2B for AI workflow decisionsAmazon releases Strands Decider 2B for AI workflow decisions
Strands Decider 2B selects from predefined options and returns a confidence score. The fully open-source model is available now and small enough to run locally.
2 min read
1. 10. 2026
OpenAI says it disrupted a campaign to extract hidden model reasoningOpenAI says it disrupted a campaign to extract hidden model reasoning
OpenAI reported a coordinated effort to extract protected model reasoning and said it closed an extraction pathway. It attributed the main cluster of activity to individuals associated with Moonshot AI, the developer of Kimi.
3 min read
1. 10. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok