Summary

It gets you browser automation without a browser on your machine.

The three agent tools are the same job at three levels of reasoning, and that is a choice you make per task rather than per install: browser_use_agent when the flow is mechanical, claude_computer_use_agent when the page fights back. Persistent profiles are what make the second run cheap — the login you did once is still there.

What it is

Hyperbrowser's own MCP server: 10 tools over their cloud browsers, covering single-page scraping, multi-page crawling, schema-driven extraction, Bing search, three autonomous browser agents, and persistent profiles that carry a logged-in session between runs.

What you get

Scrape one page in the output format you ask for (scrape_webpage), or crawl a site from a starting URL with maxPages, followLinks and ignoreSitemap (crawl_webpages).
Extract structured data against a schema and a prompt, across several URLs in one call (extract_structured_data).
Hand a whole task to a browser agent: browser_use_agent for fast, low-cost runs, openai_computer_use_agent for general-purpose ones, claude_computer_use_agent for the intricate ones — each with maxSteps and optional step-by-step reporting.
Search the web through Bing (search_with_bing).
Keep a persistent profile so a session survives between runs (create_profile, list_profiles, delete_profile).

Requirements

A Hyperbrowser API key — the server reads HYPERBROWSER_API_KEY — and Node.js on the machine. MIT licensed.

Setup effort

One command plus a key — npx hyperbrowser-mcp <YOUR-HYPERBROWSER-API-KEY>, then supply credentials