Hyperbrowser MCP Server – Web Automation Tools

Hyperbrowser MCP Server is Hyperbrowser's official Model Context Protocol server for exposing hosted browser automation and web data tools to MCP-compatible clients. AI agents can use it to scrape webpages, crawl linked pages, extract structured JSON, run Bing-powered web search, and invoke browser-agent workflows backed by Hyperbrowser.

#web-scraping#browser#extraction

Overview

Hyperbrowser MCP Server connects AI agents to Hyperbrowser, a web automation platform for scraping, crawling, structured extraction, and browser-based task execution. The official repository describes the server as Hyperbrowser's Model Context Protocol implementation and notes that the broader Hyperbrowser API supports a superset of the MCP server's features.

What the MCP server enables

The server exposes tools for turning live web pages into agent-usable data. Documented tools include scrape_webpage for extracting formatted content such as Markdown or screenshots from a webpage, crawl_webpages for following linked pages and collecting LLM-friendly content, extract_structured_data for converting HTML into structured JSON, and search_with_bing for web search results. It also exposes browser-agent tools including browser_use_agent, openai_computer_use_agent, and claude_computer_use_agent, plus persistent profile tools to create, list, and delete Hyperbrowser profiles.

When to use it

Use this MCP server when an agent needs current web content rather than static model knowledge. Practical uses include collecting product or competitor information from web pages, extracting tables or page content into structured JSON, crawling documentation or marketing pages, searching the web before analysis, and running browser automation tasks that require a hosted browser environment. It is especially useful when an AI client needs repeatable web actions without embedding custom scraping code in the client itself.

Connection and authentication

Hyperbrowser documents a local stdio MCP server launched with npx. Client examples for Cursor, Windsurf, and Claude Desktop use npx with the hyperbrowser-mcp package and the HYPERBROWSER_API_KEY environment variable. The manual command can also pass the API key as an argument, but environment-based configuration is the documented client pattern and avoids placing credentials in command history or shared config comments.

Key considerations

Keep the API key scoped and stored securely in the MCP client's secret or environment configuration. The MCP server exposes web interaction and profile management tools, so clients should apply normal tool-approval practices before allowing actions that browse, scrape, crawl, or automate third-party websites. The README documents stdio usage; do not configure remote HTTP or SSE endpoints unless Hyperbrowser publishes official support for those transports for this server.

Frequently Asked Questions

When should an AI agent use the Hyperbrowser MCP Server?

Use it when an agent needs to scrape a webpage, crawl linked pages, extract structured JSON from HTML, run web search, or execute browser-agent workflows through Hyperbrowser rather than relying on model memory.

What tools does the Hyperbrowser MCP Server expose?

The official README lists tools for `scrape_webpage`, `crawl_webpages`, `extract_structured_data`, `search_with_bing`, Browser Use automation, OpenAI computer-use automation, Claude computer-use automation, and persistent profile creation, listing, and deletion.

What can an AI agent manage through this server?

An agent can request web content extraction, structured extraction, crawling, search, browser-agent tasks, and Hyperbrowser persistent profile operations that are specifically documented by the server. The broader Hyperbrowser API may support more features than the MCP server.

How is authentication configured for Hyperbrowser MCP Server?

The documented client configurations set `HYPERBROWSER_API_KEY` in the MCP server environment. The key should be referenced from a secret or environment variable, not hard-coded into shared configuration files.

Which transport should be used for Hyperbrowser MCP Server?

Use the documented local stdio transport launched with `npx -y hyperbrowser-mcp`. The official README provides stdio client examples for Cursor, Windsurf, and Claude Desktop and does not document a hosted Streamable HTTP or SSE endpoint for this server.