Browser Use Drops Into Claude Code Letting AI Automate the Web 3x Faster

Browser Use, the open-source web automation framework with over 83,500 GitHub stars, is now available as a plugin directly inside Claude Code. One command drops a full browser agent into your coding session, no separate Python script or MCP server wiring required.

What the plugin system actually is

In Claude Code, a plugin is a package that bundles one or more skills along with optional extensions like hooks and MCP server configurations. Think of skills as reusable instruction sets that tell Claude how to use a specific tool. User-scope installs the plugin to ~/.claude/skills/ so it's available in all your projects, while project-scope installs it into the repo itself, making it shareable with your team through version control.

The opening of the plugin marketplace as a first-class system is the most important update across all of spring 2026. The default Anthropic marketplace has official and verified plugins; to access community plugins, you add third-party marketplaces. Browser Use is one of the first major third-party automation tools to ship through this channel.

Getting it running in 60 seconds

Installing the Browser Use plugin is a two-step process from inside Claude Code:

# Step 1: Register the Browser Use marketplace
claude plugin marketplace add browser-use/plugins
# Step 2: Install the plugin from that marketplace
claude plugin install browser-use@browser-use-plugins

After choosing a scope, the plugin installs immediately. The skills it contains are available in your current session without restarting. Once installed, you can just tell Claude in plain English to open a URL, fill a form, or extract data from a page, and it will figure out the right browser commands to run.

For those who prefer the manual route, you can also install the skill directly: mkdir -p ~/.claude/skills/browser-use and curl the SKILL.md file from the Browser Use GitHub repo.

What Browser Use actually does

Browser Use is a Python framework that wraps Playwright and exposes a high-level agent interface. The agent can navigate pages, click elements, fill forms, take screenshots, and extract structured data, all driven by natural language instructions. From the CLI, you can run commands like browser-use open https://example.com, browser-use click 5, or browser-use screenshot page.png, with the browser staying alive between commands for fast iteration.

The practical use-cases are broad:

• Automated end-to-end testing of web apps during development
• Scraping competitor pricing or public data as part of a research task
• Filling out forms or submitting data to web portals
• Extracting structured content from pages that don't have an API
• Automating repetitive browser workflows like report downloads or dashboard exports

The model question

Browser Use ships its own fine-tuned model, ChatBrowserUse(), purpose-built for browser automation. On average it completes tasks 3-5x faster than other models with state-of-the-art accuracy. You can also use it with Claude, GPT, or Gemini via provider-prefixed model IDs, but the team recommends their hosted bu-* models for best results.

Pricing for the Browser Use hosted model is lean: $0.20 per million input tokens and $2.00 per million output tokens, with cached input at $0.02. The open-source library itself is free under the MIT license, and you can bring your own LLM keys.

Where it gets tricky

Browser automation in an agentic context comes with real friction points worth knowing upfront:

• CAPTCHAs: The open-source version has no built-in CAPTCHA bypass. You need the Browser Use Cloud tier for stealth fingerprinting and proxy rotation.
• Authentication: Managing sessions across agent runs requires either reusing a real Chrome profile or syncing auth state to a remote browser.
• Production scale: Chrome is memory-hungry. Running many parallel agents locally gets messy fast; the cloud offering handles infrastructure and memory management.
• Plugin security: Claude Code's plugin marketplace introduces a new attack surface; plugins can potentially bypass human-in-the-loop protections or exfiltrate files via indirect prompt injection. Stick to verified or well-audited plugins.

Why this pairing matters

The real unlock here is collapsing the gap between writing code and verifying it against the live web. Previously, if Claude Code generated a scraper or a form-submission script, you'd have to exit the session, run the script manually, debug it, and come back. With Browser Use as a plugin, Claude can autonomously write and execute custom automation for testing and validation without ever leaving the coding context.

Since the marketplace opened, both plugins and skills can be filtered in real time from the new /skills and /plugin prompts with built-in type-to-filter. The ecosystem is growing fast, with browser automation now one of the most active categories. Browser Use joining it with 83k+ stars of community momentum is a meaningful signal that the plugin channel is becoming the default distribution path for serious agentic tooling.