Published on 27.01.2026
TLDR: Unlike most AI browser extensions that just summarize content, Claude in Chrome takes actions - clicking buttons, navigating forms, extracting data, and running multi-step tasks while you do other work. Combined with Claude Code via MCP, it creates a powerful build-test-verify loop for developers.
The browser automation race is heating up. Dia, Comet, Atlas from OpenAI - everyone is building browsers from scratch. Anthropic took a different path. Instead of replacing Chrome, they built an extension that controls it.
Claude in Chrome launched as a research preview in August 2025. It runs on Haiku 4.5, optimized for fast, efficient multi-step browser tasks without the token cost of Sonnet or Opus. You describe the outcome you want, and it handles the clicking. The extension requires a paid Claude subscription (Pro, Max, Team, or Enterprise) and integrates with your existing Claude context.
Real-world use cases are emerging. Apartment hunting: Claude applies filters on PropertyGuru, scrolls through listings, and summarizes top matches with rent, specs, and contact info - tasks that would take 20+ minutes manually. Customer support negotiations: Claude took over an AT&T support chat, escalated politely, cited service restrictions, secured a higher refund, and prepped an FCC complaint form as backup. Profile research: Point Claude at an X profile and ask what strategies someone uses for making money online - it extracts patterns from 50 tweets without you reading them manually.
The integration with Claude Code via MCP (Model Context Protocol) is where this gets interesting for developers. One user set up a workflow where Claude in Chrome extracts Substack dashboard metrics, then Claude Code processes the patterns and updates a local retrospectives file. The browser automation feeds data to local systems automatically.
For testing, this unlocks a build-test-verify loop. Build in Claude Code with Sonnet or Opus. Test in the browser with Claude in Chrome - it navigates, fills forms, runs UX tests, and reports issues. One developer pointed Claude at a dashboard project and asked it to test the UX journey of all buttons. In 8-10 minutes, it produced a comprehensive report identifying broken features, what works well, placeholder pages, and recommendations for fixes.
Best practices that emerged: Don't micromanage the steps. Bad: "Click the filter dropdown, select 3 bedrooms, then sort by price ascending." Good: "Find 3-bedroom apartments under $4,000/month in Tampa and summarize the top 5." Claude figures out the steps. You verify the result. The extension asks permission before submitting forms, making purchases, or sending messages - you stay in control.
For architects and teams, this represents a shift in how browser-based testing and automation can work. The extension handles repetitive web tasks that normally eat 20 minutes each. But the real value is the Claude Code integration - when context stays synced across terminal and browser sessions via MCP, you can build workflows that combine local development with browser-based verification in ways that were previously fragmented across multiple tools.
Key takeaways:
Tradeoffs:
Link: Claude In Chrome: The Browser Extension That Clicks Buttons While You Sleep