Developer Sebastian Sosa has published a clean-room reimplementation of Anthropic's official "Claude in Chrome" browser extension, fully open-source under the MIT license, with one key difference: it removes the blocklist that prevents Claude from interacting with 58 domains across categories like banking, dating, and social media.
The official Claude extension, which lets Claude Code automate browser tasks like clicking buttons, filling forms, and reading page content, only works in Chrome and Edge, and blocks automation on sites Anthropic deemed sensitive. The open-source version, called Open Claude in Chrome, works across any Chromium browser (Brave, Arc, Opera, Vivaldi included) and imposes no domain restrictions.
Technically, it replicates all 18 browser automation tools from the official extension: page navigation, screenshot capture, reading the accessibility tree (the structured data browsers use to describe page elements), form input, JavaScript execution, and network monitoring. The architecture uses three components: a Manifest V3 browser extension, a Node.js MCP server (the protocol Claude uses to connect to external tools), and a native messaging bridge linking them.
The project is described as a clean-room implementation, meaning Sosa studied the official extension's behavior and rebuilt the functionality from scratch rather than copying Anthropic's code.
Anthropic put those domain blocks in place for liability and safety reasons, so removing them is a double-edged sword. You get full browser automation without artificial limits, but you also lose the guardrails that prevent an AI agent from accidentally interacting with your bank account or trading platform. Anyone running this should understand that "no restrictions" means exactly that.
The extension is available on GitHub under lukanlabs/lukan. For developers already using Claude Code's computer use capabilities, it's a more flexible alternative, but you're trading Anthropic's safety net for full control.