# browserbase/skills: A Claude Code Plugin for Browserbase Web Automation > browserbase/skills is a plugin for Claude Code that adds 18 browser automation skills, including web scraping, UI testing, trace capture, and research workflows. It connects Claude Code to Browserbase's cloud browser infrastructure and the browse CLI. **browserbase/skills** — Browserbase's official collection of agent skills to access the web. - Repository: https://github.com/browserbase/skills - Website: https://www.browserbase.com/SKILL.md - Stars: 3,733 · Forks: 240 - Language: JavaScript - License: not declared - Published: 2026-09-23 · Updated: 2026-09-23 · Language: en - Canonical page: https://hysenlabs.com/projects/browserbase-skills ## What browserbase/skills Adds to Claude Code Claude Code supports a plugin system that lets external tools extend the agent's capabilities. browserbase/skills is a first-party plugin from Browserbase, the company behind the Stagehand browser automation framework, that installs 18 named skills into Claude Code. Once installed, Claude Code can use those skills to navigate websites, capture browser traces, generate API specifications from observed HTTP traffic, run UI tests, and perform company research using the web. The plugin connects Claude Code to Browserbase's cloud browser infrastructure. Browserbase provides remote browser sessions with additional capabilities: identity-verified browsers, CAPTCHA solving, and residential proxies. The browse CLI, which the plugin manages, communicates with those sessions. Some skills, such as fetch and search, work without a cloud browser session and do not require a Browserbase account. The README gives representative usage examples for what Claude can do after installation: - Go to Hacker News, get the top post comments, and summarize them. - QA test http://localhost:3000 and fix any bugs you encounter. - Order me a pizza, you're already signed in on Doordash. - Use browse to list my Browserbase projects and show the output as JSON. ## The Eighteen Skills and What Each Does The README documents 18 skills in a table, each with a SKILL.md file in skills/ for full details: - browser: Automate web browser interactions via CLI commands, with Browserbase Identity, verified browsers, CAPTCHA solving, and residential proxies. - functions: Deploy serverless browser automation to Browserbase cloud using the browse CLI. - browser-trace: Capture a full DevTools-protocol trace alongside automation, including CDP events, screenshots, and DOM dumps, then split the stream into per-page searchable buckets. - browser-to-api: Turn a website's observable HTTP traffic into a best-effort OpenAPI 3.1 spec by analyzing a browser-trace capture. - autobrowse: Self-improving browser automation that runs a browsing task, reads the trace, and improves the navigation script until it passes reliably. - optimize-agent-prompt: Optimize Browserbase Agent API system prompts through repeated runs, session logs, and scoring. - safe-browser: Build local browser agents with a CDP-gated safe_browser tool that enforces a domain allowlist. - webmcp-gen: Author, compile, and validate site-specific WebMCP init scripts with the Stagehand WebMCP runtime. - add-webmcp: Analyze an existing web app and add first-party WebMCP tools backed by its routes, forms, actions, and schemas. - cookie-sync: Sync cookies from local Chrome to a Browserbase persistent context so the browse CLI can access authenticated sites. - fetch: Fetch HTML or JSON from static pages without a browser session. - search: Search the web and return structured results without a browser session. - ui-test: AI-powered adversarial UI testing that analyzes git diffs to test changes or explores the full app to find bugs. - browser-use-to-stagehand: Migrate browser-use (Python) automation to Stagehand v3 (TypeScript) on Browserbase. - agent-experience: Audit how agent-friendly a product or documentation site is by dropping Claude subagents at it. - company-research: Discover companies matching an ICP using the Browserbase Search API and deep-research them into a scored report. - event-prospecting: Extract speakers from a conference page, filter against an ICP, and deep-research the best-fit people. - competitor-analysis: Auto-discover competitors via Browserbase Search API and compile a browsable HTML report with a feature matrix and mentions feed. ## Installing via Claude Code's Plugin System or npx Two installation paths are documented. For Claude Code, run the following commands in the terminal: ```bash /plugin marketplace add browserbase/skills ``` Then install the plugin: ```bash /plugin install browse@browserbase ``` For other coding agents, the npx command installs the skills: ```bash npx skills add browserbase/skills ``` The README notes that Claude Code must be restarted after installation for changes to take effect. A manual installation path is also documented in the README: open the /plugin command, select option 3 to add a marketplace, enter browserbase/skills as the marketplace source, select the browse plugin, and install it. For local browser work, the --local flag on the first browser command starts a clean isolated browser. The --auto-connect flag tells the agent to reuse an existing Chrome session with its cookies and login state. The Chrome troubleshooting section of the README notes that on Linux, Chrome can be installed with sudo apt install google-chrome-stable, and on macOS and Windows by downloading from google.com/chrome/. ## AutoBrowse and Prompt Optimization as Self-Improving Workflows Two skills in the collection address a recurring problem in browser automation: brittle scripts that fail when a page changes layout or introduces a new interaction step. The autobrowse skill addresses this by treating automation as an iterative loop. It runs a browsing task, reads the resulting browser-trace, identifies where the automation succeeded and where it failed, and updates the navigation script until the task passes reliably. The README describes this as 'self-improving browser automation.' This is not a one-shot recorder; it is a feedback loop that uses trace data to fix the automation. The optimize-agent-prompt skill applies the same iterative approach to prompt engineering. It runs a Browserbase Agent API system prompt through repeated sessions, captures the agent message traces and session logs, scores each run, and continues iterating until confirmed runs pass without change. This skill is specifically for teams using the Browserbase Agent API who want to tune a system prompt for reliability rather than finding the right prompt manually. Both skills depend on browser-trace as a foundation: the trace capture skill produces the CDP firehose data that these iteration skills read and analyze. ## Local-Mode Skills That Work Without Browserbase Several skills in the collection work without a Browserbase cloud session. The fetch skill retrieves HTML or JSON from static pages, follows redirects, and inspects status codes and headers. The search skill returns structured web results, including titles, URLs, and metadata, without opening a browser session at all. For local development, the --local flag on the browse CLI starts a clean isolated Chromium instance on the developer's machine. The --auto-connect flag connects to an existing Chrome session, which allows the agent to access authenticated state from the developer's own browser without syncing cookies to Browserbase. The cookie-sync skill bridges the two modes: it reads cookies from the local Chrome profile and writes them to a Browserbase persistent context, enabling the browse CLI to access sites where the developer is already logged in. Resetting the local Chrome profile used by the tool is done with: ```bash rm -rf .chrome-profile ``` The safe-browser skill is the security-focused variant: it creates a browser agent that can only visit domains on an explicit allowlist, which is useful when building agents that should not stray outside a defined set of websites. ## Limitations and Dependencies The repository version is 0.0.1 and has no GitHub releases. The package.json marks the version as 0.0.1, and no changelog is documented. Skill behavior may change between commits without a versioned contract. No license is documented. The repository-level license field is listed as unknown, and no LICENSE file appears in the top-level directory listing. For teams considering incorporating the skills into their own tooling, the absence of a license means the default copyright law applies. Most skills require a Browserbase account. The browser, functions, autobrowse, optimize-agent-prompt, browser-trace, and cloud-dependent skills all connect to Browserbase's infrastructure. The company-research, event-prospecting, and competitor-analysis skills use the Browserbase Search API. Costs depend on Browserbase's pricing, which is not documented in the repository. The TypeScript build step runs automatically on install via the postinstall hook in package.json (npm run build), which compiles the TypeScript source to dist/. A working Node.js environment and build tools are required. ## Playwright MCP as the Alternative for Vendor-Neutral Browser Automation Playwright MCP is an open-source Model Context Protocol server maintained by Microsoft that exposes Playwright browser automation as MCP tools. Any MCP-compatible LLM client can use Playwright MCP to navigate pages, take screenshots, click elements, and fill forms. It targets the local browser directly, without a cloud session layer. The key difference from browserbase/skills is scope and infrastructure. Playwright MCP provides the core automation primitives (navigate, click, type, screenshot) for any website, running against a local Chromium, Firefox, or WebKit instance. browserbase/skills builds on Browserbase's cloud infrastructure to add capabilities that local Playwright cannot provide: CAPTCHA solving, residential proxies, session persistence across agents, and identity-verified browsers. For developers who want browser automation without a cloud dependency or a vendor account, Playwright MCP is the appropriate starting point. For teams that specifically need CAPTCHA bypass, authenticated session sharing across agents, or the research and analysis skills in the browserbase collection, the Browserbase-specific infrastructure is what provides those capabilities. ## Conclusion Claude Code users who want browser automation, web research, or adversarial UI testing integrated into their coding agent workflow will find the Browserbase skills plugin a focused solution. The plugin requires a Browserbase account for cloud browser sessions, though local Chrome sessions work for the fetch, search, and local browsing scenarios without a cloud dependency. No license is documented in the repository and the version is 0.0.1. Install with /plugin marketplace add browserbase/skills in Claude Code, then /plugin install browse@browserbase, and restart Claude Code before using any skill. ## FAQ ### How do you install skills in Claude Code from GitHub? Run /plugin marketplace add browserbase/skills in Claude Code to add the marketplace, then /plugin install browse@browserbase to install the plugin. Restart Claude Code after installation. You can also use npx skills add browserbase/skills. ### How do you install browserbase/skills in Claude Code? Add the marketplace with /plugin marketplace add browserbase/skills, then install the plugin with /plugin install browse@browserbase, and restart Claude Code. A manual path is also available through the /plugin menu. ### How do you use browserbase/skills in Claude Code? After installation, ask Claude to browse or use the Browserbase CLI in plain language. Examples from the README include: 'Go to Hacker News, get the top post comments, and summarize them' or 'QA test http://localhost:3000 and fix any bugs you encounter.' ## Sources - [browserbase/skills on GitHub](https://github.com/browserbase/skills) - [Issues](https://github.com/browserbase/skills/issues) - [Project website](https://www.browserbase.com/SKILL.md) - [README](https://github.com/browserbase/skills/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/browserbase-skills