M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
H

Hyperbrowser MCP Server

Official
Scrape, crawl, and automate the web with Hyperbrowser's cloud browser and agent tools
How to install ↓ Compare GitHub source repository ↗
Category
Browser #17 of 32
Stars
★ 789 Popular
Transport
stdio (local process)
Runtime
Node.js
Credentials
API key / credential required
License
MIT
Last commit
⚠
Tools
10
52FMRS · D

The README is clear and provides concrete config examples for Cursor, Windsurf, and Claude Desktop plus a Smithery one-click install, with an explicit tool list; it does not include a machine-readable manifest or further permission details beyond the API key requirement.

Strongest · Setup experience 13/20 Weakest · Reliability 7/20

Reliability
7/20
Security and permissions
9/20
Maintenance
11/20
Documentation
12/20
Setup experience
13/20
Why each score
Reliability 7/20
The README lists ten tools (scrape_webpage, crawl_webpages, extract_structured_data, search_with_bing, three computer-use/browser-use agents, and a profile management trio), but the manifest was not cached and no test files or CI workflow evidence is present in the supplied material, so there is no way to confirm the declared tools match real behavior or that the MCP init/tool-list handshake succeeds. Per the static calibration rule, reliability cannot exceed 12 without verifiable execution evidence (CI plus committed tests). The stated build steps (npm install / npm run build / node dist/server.js) look plausible on paper but are unverified, and error handling, timeouts, or retry behavior for the external Hyperbrowser API dependency are not documented at all, warranting a low score.
Security and permissions 9/20
The server takes its credential via the HYPERBROWSER_API_KEY environment variable rather than a hardcoded or plaintext example, which is sound practice. However, that credential is used to call an external Hyperbrowser cloud API, involving real network egress and data flow that the README does not describe in terms of data boundaries, logging, or caching. More importantly, three general-purpose browser automation agents (browser_use_agent, openai_computer_use_agent, claude_computer_use_agent) are exposed that can act autonomously on web pages — a high-impact capability — with no mention of pre-action confirmation, scope limits, or cost/rate controls. delete_profile is also a deletion-type operation with no described confirmation step. This does not rise to a red-line violation (no evidence of malware, credential theft, or forced irreversible pay/deploy defaults), but the overall lack of least-privilege scoping and confirmation disclosure justifies a meaningful deduction.
Maintenance 11/20
The repository has 787 stars, 8 open issues, is not archived, and carries an MIT license, indicating basic open-source governance and licensing. However, the README provides no release history, changelog, recent commit activity, or security-response channel, so maintenance cadence and dependency-update practices cannot be verified, resulting in a mid-low score.
Documentation 12/20
Documentation is well structured, with a table of contents, configuration examples for multiple mainstream clients (Cursor, Windsurf, Claude Desktop), a Smithery one-line install option, and a concise tool list, giving basic usability. However, each tool gets only a single-sentence description with no parameter specifications, return formats, rate limits, cost details (whether Hyperbrowser's API is paid is unstated), or a troubleshooting section — users must consult external docs (docs.hyperbrowser.ai) for real usage, placing documentation quality at a middling level.
Setup experience 13/20
Installation steps are clear: a one-line npx run or a from-source build, plus configuration examples for three mainstream clients (Cursor, Windsurf, Claude Desktop) and a Smithery one-click install, making for a short, clear setup path. However, under the static calibration rule, setup cannot exceed 15 without verifiable execution evidence such as CI or tests. Users must also manually register with Hyperbrowser to obtain an API key as a prerequisite step, and there is no evidence the documented flow has actually been verified to work end-to-end, so the score is capped at a moderate-high but limited level.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkChanges third-party account dataControls a browser

Best for

  • Developers who want to scrape or crawl webpages directly from an MCP client
  • Automation pipelines that need messy HTML converted to structured JSON
  • Users who want browser agents (CUA/Claude Computer Use/Browser Use) to perform web tasks

Not for

  • Users who don't want to depend on a third-party cloud browser service or obtain an API key
  • Fully offline or air-gapped environments
  • Cost-sensitive projects unable to cover Hyperbrowser API usage fees

Required permissions

  • Requires a valid HYPERBROWSER_API_KEY to call the Hyperbrowser cloud service
  • The server makes outbound network requests and browser actions against external webpages on the user's behalf
  • Browser agent tools may require additional OpenAI or Anthropic credentials to drive CUA/Claude Computer Use

Risks and side effects

  • A misconfigured or leaked API key could be misused, incurring unexpected charges
  • Browser agents can click and submit forms on live pages, risking unintended actions or exposure to unexpected content
  • Scraping/crawling activity may violate target sites' terms of service or trigger anti-bot defenses

Setup

Before you start

Runtime：Node.js

HYPERBROWSER_API_KEY requiredsecret API key for the Hyperbrowser platform, required to authenticate the server; obtain it from your Hyperbrowser account/dashboard.
  1. Obtain a Hyperbrowser API key.
  2. Run npx hyperbrowser-mcp <YOUR-HYPERBROWSER-API-KEY> directly, or add an mcpServers entry with HYPERBROWSER_API_KEY set to your key in Cursor's ~/.cursor/mcp.json, Windsurf's model_config.json, or Claude Desktop's config.
  3. For development, clone the repo, run npm install, npm run build, then node dist/server.js.
  4. Alternatively install for Claude Desktop via Smithery with npx -y @smithery/cli install @hyperbrowserai/mcp --client claude.
claude_desktop_config.json 
{"mcpServers":{"hyperbrowser":{"command":"npx","args":["-y","hyperbrowser-mcp"],"env":{"HYPERBROWSER_API_KEY":"YOUR-API-KEY"}}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "hyperbrowser": {
      "command": "npx",
      "args": [
        "-y",
        "hyperbrowser-mcp"
      ],
      "env": {
        "HYPERBROWSER_API_KEY": "YOUR-API-KEY"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add hyperbrowser -e HYPERBROWSER_API_KEY=YOUR-API-KEY -- npx -y hyperbrowser-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After adding the config, check that tools like scrape_webpage and search_with_bing appear in your client's tool list, then ask the client to run a simple Bing search; a returned result confirms the connection and API key work.

Troubleshooting

  1. Verify HYPERBROWSER_API_KEY is set correctly as an environment variable
  2. Check that npx can reach the network to download the hyperbrowser-mcp package
  3. For local development, confirm npm install and npm run build completed before running node dist/server.js
  4. Confirm the client's (Cursor/Windsurf/Claude Desktop) mcp config file path and JSON syntax are correct

Things to try

Once connected, you can ask your AI assistant things like:

  • Scrape https://example.com and return the content as markdown
  • Search Bing for 'MCP protocol' and give me the results
  • Extract the product name and price from this page as structured JSON
  • Create a new persistent Hyperbrowser profile for me

Tools 10

scrape_webpage read-only
Extract formatted content (markdown, screenshot, etc.) from any webpage
crawl_webpages read-only
Navigate through multiple linked pages and extract LLM-friendly formatted content
extract_structured_data read-only
Convert messy HTML into structured JSON
search_with_bing read-only
Query the web and get results with Bing search
browser_use_agent read-only
Fast, lightweight browser automation with the Browser Use agent
openai_computer_use_agent read-only
General-purpose automation using OpenAI's CUA model
claude_computer_use_agent read-only
Complex browser tasks using Claude computer use
create_profile writes
Creates a new persistent Hyperbrowser profile
Show 2 more tools
delete_profile destructive
Deletes an existing persistent Hyperbrowser profile
list_profiles read-only
Lists existing persistent Hyperbrowser profiles

Use cases

Scrape a webpage and convert it to markdown for LLM analysis
Crawl a multi-page site and extract content across linked pages
Convert unstructured webpage HTML into structured JSON
Pull live web results via Bing search
Use browser agents to fill forms, click through flows, and automate tasks

Supported clients

Claude Desktop
Cursor
Windsurf

Listed from the project's documentation, not tested by this site.

Overview

Hyperbrowser MCP Server is the official Model Context Protocol server for Hyperbrowser, giving MCP-compatible clients tools to scrape webpages, extract structured data, crawl linked pages, and drive general-purpose browser agents such as OpenAI's CUA, Anthropic's Claude Computer Use, and Browser Use. It requires a Hyperbrowser API key and runs as a stdio process launched via npx or Node.

Similar servers

Compare with other MCP servers

Source revision 7492e83cf182 Data synced 2026-10-12 Read the FMRS scoring method