Hyperbrowser
# Hyperbrowser MCP Server
[](https://smithery.ai/server/@hyperbrowserai/mcp)
This is Hyperbrowser's Model Context Protocol (MCP) Server. It provides various tools to scrape, extract structured data, and crawl webpages. It also provides easy access to general purpose browser agents like OpenAI's CUA, Anthropic's Claude Computer Use, and Browser Use.
More information about the Hyperbrowser can be found [here](https://docs.hyperbrowser.ai/). The hyperbrowser API supports a superset of features present in the mcp server.
More information about the Model Context Protocol can be found [here](https://modelcontextprotocol.io/introduction).
## Table of Contents
- [Installation](#installation)
- [Usage](#usage)
- [Tools](#tools)
- [Configuration](#configuration)
- [License](#license)
## Installation
### Manual Installation
To install the server, run:
```bash
npx hyperbrowser-mcp <YOUR-HYPERBROWSER-API-KEY>
```
## Running on Cursor
Add to `~/.cursor/mcp.json` like this:
```json
{
"mcpServers": {
"hyperbrowser": {
"command": "npx",
"args": ["-y", "hyperbrowser-mcp"],
"env": {
"HYPERBROWSER_API_KEY": "YOUR-API-KEY"
}
}
}
}
```
## Running on Windsurf
Add to your `./codeium/windsurf/model_config.json` like this:
```json
{
"mcpServers": {
"hyperbrowser": {
"command": "npx",
"args": ["-y", "hyperbrowser-mcp"],
"env": {
"HYPERBROWSER_API_KEY": "YOUR-API-KEY"
}
}
}
}
```
### Development
For development purposes, you can run the server directly from the source code.
1. Clone the repository:
```sh
git clone git@github.com:hyperbrowserai/mcp.git hyperbrowser-mcp
cd hyperbrowser-mcp
```
2. Install dependencies:
```sh
npm install # or yarn install
npm run build
```
3. Run the server:
```sh
node dist/server.js
```
## Claude Desktop app
This is an example config for the Hyperbrowser MCP server for the Claude Desktop client.
```json
{
"mcpServers": {
"hyperbrowser": {
"command": "npx",
"args": ["--yes", "hyperbrowser-mcp"],
"env": {
"HYPERBROWSER_API_KEY": "your-api-key"
}
}
}
}
```
## Tools
* `scrape_webpage` - Extract formatted (markdown, screenshot etc) content from any webpage
* `crawl_webpages` - Navigate through multiple linked pages and extract LLM-friendly formatted content
* `extract_structured_data` - Convert messy HTML into structured JSON
* `search_with_bing` - Query the web and get results with Bing search
* `browser_use_agent` - Fast, lightweight browser automation with the Browser Use agent
* `openai_computer_use_agent` - General-purpose automation using OpenAI’s CUA model
* `claude_computer_use_agent` - Complex browser tasks using Claude computer use
* `create_profile` - Creates a new persistent Hyperbrowser profile.
* `delete_profile` - Deletes an existing persistent Hyperbrowser profile.
* `list_profiles` - Lists existing persistent Hyperbrowser profiles.
### Installing via Smithery
To install Hyperbrowser MCP Server for Claude Desktop automatically via [Smithery](https://smithery.ai/server/@hyperbrowserai/mcp):
```bash
npx -y @smithery/cli install @hyperbrowserai/mcp --client claude
```
## Resources
The server provides the documentation about hyperbrowser through the `resources` methods. Any client which can do discovery over resources has access to it.
## License
This project is licensed under the MIT License.
TDQS
Scored across 10 tools
The tool set has clear distinctions between core browser automation agents (browser_use_agent, claude_computer_use_agent, openai_computer_use_agent) and utility functions (crawl_webpages, scrape_webpage, extract_structured_data, search_with_bing, profile management). However, there is significant overlap between the three agent tools—all perform browser automation with different model backends—which could cause confusion about which to select for a given task. The descriptions help differentiate their strengths, but the fundamental purpose overlap remains.
Most tools follow a consistent snake_case pattern with clear verb_noun structures (e.g., crawl_webpages, create_profile, extract_structured_data, scrape_webpage, search_with_bing). The three agent tools deviate slightly with longer, descriptive names (e.g., browser_use_agent, claude_computer_use_agent), but they still maintain readability and a similar format. Overall, the naming is mostly predictable with only minor inconsistencies.
With 10 tools, the count is well-scoped for a browser automation server. It covers a range of functionalities from high-level agent-based interactions to lower-level utilities like scraping and searching, without feeling excessive. Each tool appears to serve a distinct role within the domain, making the number appropriate for the server's purpose.
The tool set provides comprehensive coverage for browser automation tasks, including agent-based interactions, web crawling, scraping, data extraction, searching, and profile management. Minor gaps exist, such as the lack of tools for managing browser sessions or handling cookies directly, but these are not critical for core workflows. Agents can likely work around these omissions using the available tools.