Browse.sh - review - 2026

Published: 2026-07-28

browse.sh is an open web skill catalog and CLI tool for browser automation, developed by Browserbase. The project provides a command-line interface to drive Chromium, enabling AI agents to not just read web pages but to fully interact with them: open sites, click buttons, fill forms, monitor network requests, and execute multi-step workflows. The key feature is the open web skill catalog — a collection of ready-made instructions for interacting with specific websites.

Installation

Install via npm:

npm install -g browse

Once installed, the browse command is available, combining browser automation, Browserbase cloud APIs, Functions, templates, and skill management.

CLI Capabilities

The tool provides a broad set of browser primitives:

Browser Automation

browse open https://example.com
browse click "button.login"
browse type "test@example.com"
browse fill "input#name" "John"
browse screenshot
browse snapshot
browse upload "input#file" path
browse wait "div.result"
browse viewport 1280 720

Navigation and Inspection

browse back
browse forward
browse reload
browse highlight
browse refs
browse get "#id"
browse is "#id"
browse eval "document.title"
browse cdp cmd params

Session Controls

browse status
browse stop
browse cursor
browse tab
browse network --tail
browse console --tail
browse mouse

Cloud APIs (Browserbase Cloud)

browse cloud project list
browse cloud session create
browse cloud context list
browse cloud extensions
browse cloud fetch url
browse cloud search query

Skills and Templates

browse skills install
browse skills list
browse skills add amazon.com
browse templates list

Architecture

browse.sh is built on three main components:

CLI (browse) — a command-line interface that controls a local or cloud-hosted Chromium browser. Written in Node.js and distributed via npm.
Skill Catalog (browse.sh) — a website and machine-readable files (llms.txt, llms-full.txt) containing descriptions of ready-made automation scenarios for various websites.
Browserbase Cloud Platform — infrastructure for running browser sessions in the cloud, including proxy pools, session management, fetch and search APIs.

An AI agent (Claude, GPT, or any LLM-based agent) accesses these components through tools. The agent can install a skill for a specific site, open a browser, execute a sequence of actions, and return the result.

Skills

Skills are structured Markdown files that describe how an AI agent should interact with a specific website. Each skill contains:

Metadata: name, description, category, tags, update date, recommended method (API or browser).
Purpose — what the skill does.
When to Use — usage scenarios.
Workflow — step-by-step instructions for the agent: which URLs to open, which APIs to call, which elements to find, which data to extract.
Requirements: whether proxies are needed, whether authentication is required.

Skill Examples

Ticketmaster Find Tickets — search for upcoming events by artist, team, or show. Returns venue, date, on-sale window, presale times, sold-out/cancelled/postponed flags, and the canonical event URL. Uses Ticketmaster's unauthenticated internal API behind a residential proxy.
NASA APOD — fetch NASA's Astronomy Picture of the Day. Supports any date back to June 16, 1995. Uses the public NASA API.
FlightAware Track Flight — track flight status by airline+flight number or tail number. Returns state, origin/destination, times, route, position, delays, and a live-map link.
Glassdoor Company Reviews — extract company ratings, sub-ratings, CEO approval percentages, and employee reviews.
AllTrails Search Trails — search hiking trails by location, difficulty, length, and rating.
SAM.gov Contract Opportunities — search active federal contract opportunities with filters for status, NAICS, place of performance, and more.
ABC7 California Highway Traffic — get real-time MPH data for all California highways from the Sigalert/Total Traffic Network JSON backend.

Each skill has a recommended_method field indicating the preferred approach: direct API (faster and cheaper) or browser (slower but works when the API is unavailable). For example, NASA APOD recommends the API method, with the browser as a fallback.

Skill Categories

Based on available skills, the following categories emerge:

Tickets and Events — Ticketmaster, concert and sports event search
Flights — FlightAware, flight tracking
Government Procurement — SAM.gov, contract opportunity search
Transportation — ABC7, highway traffic data
Outdoor and Nature — AllTrails, trail search; NASA APOD
Careers — Glassdoor, company reviews
Fitness — XCore Studio, class booking

AI Agent Integration

browse.sh provides three surfaces for agent discovery:

llms.txt (https://browse.sh/llms.txt) — a compact index for fast agent discovery. Contains a brief description and links to each skill.
llms-full.txt (https://browse.sh/llms-full.txt) — an expanded catalog with the full content of all SKILL.md files without requiring additional HTTP requests.
Skill Catalog UI (https://browse.sh/) — a human-readable interface for browsing skills.

This system allows AI agents to dynamically discover and load skills for working with specific websites, making browse.sh not just a CLI utility but an entire ecosystem for agentic browsing.

Comparison with Alternatives

Tool	Type	Open Skills	Cloud Sessions	CLI
browse.sh	CLI + Skill Catalog	Yes (open catalog)	Yes (Browserbase Cloud)	Yes
Playwright	Library	No	No (local only)	Yes
Puppeteer	Library	No	No (local only)	No
Selenium	Tool	No	Partial (Selenium Grid)	Yes
Browserbase SDK	SDK + API	No	Yes	No

The main difference between browse.sh and traditional browser automation tools (Playwright, Puppeteer, Selenium) is its focus on AI agents and the skill catalog. Traditional tools require writing code in a programming language, while browse.sh lets agents perform actions through simple CLI commands.

Technical Details

Engine: Chromium (managed via Browserbase or locally)
CLI implementation language: Node.js
Installation: npm (global)
License: Proprietary (Browserbase)
Skill format: Markdown (SKILL.md) with YAML frontmatter
API: Browserbase REST API for cloud sessions
Protocol: Chrome DevTools Protocol (CDP) for local control

Skills can use two execution methods:

API method — sending an HTTP request to a public or undocumented website API. Faster, cheaper, but may be blocked.
Browser method — opening the site in Chromium, performing actions in the DOM. Slower (up to 100× more wall-time by some estimates), but works even when the API is unavailable.

To handle anti-bot systems, skills can use residential proxies (the proxies: true field in metadata).

Browserbase Ecosystem

browse.sh is part of the larger Browserbase platform — the "Browser Agent Platform". The platform includes:

Browserbase Cloud — cloud browser sessions managed via API
Browserbase Fetch API — lightweight page retrieval without launching a browser
Browserbase Search API — search API
Browserbase Functions — serverless functions for browser automation
Model Gateway — LLM gateway
Agent Identity — unified agent identity
1Password Integration — secure authentication via 1Password

Summary

browse.sh is innovative tool at the intersection of browser automation and AI agents. It solves the "hands" problem for AI: allowing agents to not just read text but to fully interact with web interfaces. The open skill catalog makes it especially valuable — the community can create and share skills for any website.

Key advantages:

Simple CLI usage
Open catalog of ready-made skills for popular websites
Both local and cloud (Browserbase) execution
AI agent integration via llms.txt
Flexible method selection (API or browser)

Key limitations:

Dependence on the Browserbase platform for cloud features
Proprietary license
Skills require maintenance and updates when websites change
Some sites require residential proxies

Links

Official website: https://browse.sh/
Browserbase Documentation: https://docs.browserbase.com/
Skill Catalog: https://browse.sh/
llms.txt (for AI agents): https://browse.sh/llms.txt
llms-full.txt: https://browse.sh/llms-full.txt
Browserbase Platform: https://www.browserbase.com/
npm package: https://www.npmjs.com/package/browse

×