Overview

The Hyperbrowser MCP Server implements the Model Context Protocol (MCP) to expose a set of web-crawling, scraping, and browser-automation tools as an MCP-capable server. Run it locally (via npx) or embed it into an MCP-enabled platform to provide LLMs and agents with structured context from webpages, screenshots, and extracted JSON data. It also exposes browser agents such as OpenAI CUA and Anthropic Claude Computer Use to perform automated browser workflows when a higher level of interaction is required.

This server is useful when you need production-ready webpage extraction or when you want agents to reliably operate on the web and return LLM-friendly outputs. Typical workflows include transforming messy HTML into structured JSON, traversing multi-page document sets, scraping product or news pages into markdown + metadata, and delegating complex browser actions to Claude/OpenAI agent implementations.

Features

Webpage scraping with formatted outputs (markdown, text, screenshots)
Multi-page crawling that follows links and aggregates content for LLM consumption
HTML-to-JSON structured data extraction (schema-first or inferred)
Web search via Bing integration for discovery + scraping
Lightweight browser automation via Browser Use agent
General-purpose automation via OpenAI CUA and Anthropic Claude Computer Use agents
Profile management (create, list, delete persistent Hyperbrowser profiles)
Exposes hyperbrowser documentation through MCP resources for discovery

Installation / Configuration

Quick run (no global install):

# Run the MCP server with your Hyperbrowser API key
npx hyperbrowser-mcp
# or explicitly provide your API key via env:
HYPERBROWSER_API_KEY="YOUR_API_KEY" npx hyperbrowser-mcp

Development (clone and run locally):

git clone [email protected]:hyperbrowserai/mcp.git hyperbrowser-mcp
cd hyperbrowser-mcp
npm install
npm run build
node dist/server.js

Cursor integration (add to ~/.cursor/mcp.json):

{
  "mcpServers": {
    "hyperbrowser": {
      "command": "npx",
      "args": ["-y", "hyperbrowser-mcp"],
      "env": {
        "HYPERBROWSER_API_KEY": "YOUR-API-KEY"
      }
    }
  }
}

Windsurf / Codeium configuration (./codeium/windsurf/model_config.json):

{
  "mcpServers": {
    "hyperbrowser": {
      "command": "npx",
      "args": ["-y", "hyperbrowser-mcp"],
      "env": {
        "HYPERBROWSER_API_KEY": "YOUR-API-KEY"
      }
    }
  }
}

Claude Desktop client example:

{
  "mcpServers": {
    "hyperbrowser": {
      "command": "npx",
      "args": ["--yes", "hyperbrowser-mcp"],
      "env": {
        "HYPERBROWSER_API_KEY": "your-api-key"
      }
    }
  }
}

Available Tools / Resources

The MCP server exposes a set of tools (methods) that clients can call. Below is a concise reference:

Tool name	Purpose
scrape_webpage	Extract content from a single URL into markdown, text, and optional screenshot
crawl_webpages	Follow links and aggregate content across multiple pages into LLM-friendly format
extract_structured_data	Transform unstructured HTML into structured JSON (tables, entities, objects)
search_with_bing	Run a Bing web search and return results for follow-up scraping
browser_use_agent	Fast, programmatic browser actions (navigation, clicks, captures)
openai_computer_use_agent	Use OpenAI CUA for complex browser interactions and task orchestration
claude_computer_use_agent	Use Anthropic Claude Computer Use for longer-running or complex browser tasks
create_profile	Create a persistent Hyperbrowser profile for automation contexts
delete_profile	Delete an existing persistent profile
list_profiles	List existing persistent profiles
resources	Discovery endpoint that returns server documentation and capability descriptions

Clients discover these tools via MCP resource discovery and can call them as MCP tool actions.

Use Cases

Product catalog scraping: Use scrape_webpage to extract product details and extract_structured_data to normalize price, SKU, images, and specs into JSON for downstream ingestion.
Documentation harvesting: Run crawl_webpages on a docs site to collect pages, generate markdown summaries, and feed them into a retrieval-augmented generation (RAG) pipeline.
Competitive monitoring: Combine search_with_bing to find mentions, then scrape_webpage to capture article snapshots and metadata.
Browser automation workflows: Use openai_computer_use_agent or claude_computer_use_agent to automate sign-in, navigation, or multi-step form submission, and capture final state (DOM + screenshot).
Structured data recovery: When a site exposes messy HTML tables or inconsistent markup, run extract_structured_data to convert into normalized JSON objects for analytics.

Example workflow (conceptual):

Call search_with_bing(“site:example.com pricing”) to discover candidate pages.
For each result, call scrape_webpage(url) to get markdown + screenshot.
Call extract_structured_data(html) to produce JSON records for each product page.
Store outputs in your database or feed into an LLM toolchain for summarization.

Resources & Links

GitHub: https://github.com/hyperbrowserai/mcp
Hyperbrowser docs: https://docs.hyperbrowser.ai/
Model Context Protocol: https://modelcontextprotocol.io/introduction

License

This project is distributed under the MIT License.

Hyperbrowser MCP Server: Web Crawling & Data Extraction