Hyperbrowser MCP
Hyperbrowser MCP Server
The official MCP server for Hyperbrowser cloud browsers: page scraping and crawling, structured data, Bing search and three browser agents
- 791 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Browser agents click buttons and fill in forms on sites themselves, and persistent profiles keep logged-in sessions
- All pages, requests and data pass through the Hyperbrowser cloud under your key
- In SSE mode without the `--sse --serve` flags the key check is off
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/hyperbrowser-mcp-serverDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Browser agents click buttons and fill in forms on sites themselves, and persistent profiles keep logged-in sessions
- All pages, requests and data pass through the Hyperbrowser cloud under your key
- In SSE mode without the `--sse --serve` flags the key check is off
This is third-party code. Review the repository files before installing.
What it does
The server connects the agent to Hyperbrowser cloud browsers. It starts as `npx hyperbrowser-mcp` and calls the Hyperbrowser cloud with your `HYPERBROWSER_API_KEY`: pages are opened and processed by a cloud browser session rather than your computer. There are ten tools. `scrape_webpage` extracts a page in `markdown`, `html`, `links` and `screenshot` formats. `crawl_webpages` walks linked pages, up to 100 per call (10 by default), and optionally follows links. `extract_structured_data` turns one or more addresses, wildcards with `*` included, into JSON following your schema and prompt. `search_with_bing` searches the web. Three tools launch browser agents: `browser_use_agent` on Browser Use, `openai_computer_use_agent` on OpenAI's CUA model and `claude_computer_use_agent` on Claude Computer Use. The agent gets a task as text and carries it out in the browser up to a step limit (25 by default, 100 at most). The `create_profile`, `list_profiles` and `delete_profile` tools manage persistent browser profiles that keep sessions between runs. Every tool has an optional `sessionOptions` object with session parameters: proxy, stealth mode, captcha solving, closing accept-cookies popups and a profile. All are off by default, and the description tells the model not to set them without an explicit request. Besides tools, the server serves the Hyperbrowser documentation as MCP resources.
Who it is for. For developers and researchers who collect data from websites and hand an agent multi-step browser tasks without a local Chrome.
Good fit when
- A page is rendered by scripts and a plain request does not get it: `scrape_webpage` extracts it from a cloud session as Markdown
- You need to walk a section of a site and collect text for a knowledge base: `crawl_webpages` takes up to 100 pages per call
- You need data by schema, for example a product list with prices: `extract_structured_data` returns JSON for your JSON Schema
- A task takes many clicks and forms: `browser_use_agent`, `openai_computer_use_agent` or `claude_computer_use_agent` go through it step by step
- A session has to survive between runs: set up a persistent profile and pass its `id` in `sessionOptions`
Not a fit when
- Pages and data must not pass to third-party cloud services: all requests go through Hyperbrowser servers
- You need a local browser on your machine with your session and extensions: the server launches cloud sessions only
- You have no Hyperbrowser account and key: the server does not work without `HYPERBROWSER_API_KEY`
- You need search results from an engine other than Bing: the built-in search is `search_with_bing` only; for other search engines the catalog has separate servers, for example Firecrawl MCP Server
Example request
Scrape the pricing page as Markdown and pull plan names and prices out of it as JSONMCP
- Transport
- stdio, sse
- Authentication
- API key
Environment variables
HYPERBROWSER_API_KEYrequired, secret- Hyperbrowser API key.
Limitations
- Use the tools only on sites and with data you have permission for. This applies to session parameters too: proxy, stealth mode and captcha solving.
- A Hyperbrowser key is required; pass it as the `npx hyperbrowser-mcp <key>` argument or as `HYPERBROWSER_API_KEY` in the client configuration.
- The agents act as intermediaries to third-party models and services: Browser Use, OpenAI CUA and Claude Computer Use. A call is capped at 1 to 100 steps.
- In HTTP mode with the `--sse --serve` flags the key is checked through `https://app.hyperbrowser.ai/api/me`; without both flags the check is off and the `/sse` and `/messages` addresses are open without authentication.
- The author notes that the Hyperbrowser API is broader than the MCP tool set. The last commit of the repository is from November 2025, version 1.1.0.
How to disable. Remove the `hyperbrowser` block from `mcpServers` in `~/.cursor/mcp.json`, the Windsurf settings or `claude_desktop_config.json` and restart the client. Delete persistent browser profiles with the `delete_profile` tool. The key is revoked in the Hyperbrowser console.
FAQ
Which output formats does scrape_webpage support?
The `outputFormat` list accepts `markdown`, `html`, `links` and `screenshot`, at least one. Links return as resources and the screenshot as an image.
How do the three browser agents differ?
`browser_use_agent` is described as fast, lightweight automation on Browser Use, `openai_computer_use_agent` as general-purpose automation on OpenAI's CUA, and `claude_computer_use_agent` as a tool for complex tasks on Claude Computer Use. Their parameters are the same: `task`, `sessionOptions`, `returnStepInfo` and `maxSteps`.
What are profiles for?
A persistent Hyperbrowser profile keeps browser state between sessions. `create_profile` makes one, `sessionOptions.profile` takes its `id`, and `persistChanges` saves changes by default. `list_profiles` lists profiles and `delete_profile` removes them.
Can I run the server without npx?
Yes, the repository has a Dockerfile and a Smithery configuration. For development the sources are built with `npm run build` and started with `node dist/server.js`. Install through Smithery for Claude Desktop: `npx -y @smithery/cli install @hyperbrowserai/mcp --client claude`.
What does the server expose besides tools?
The Hyperbrowser documentation through MCP resources. A client that can list them gets it without separate calls.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that drives your already open Chrome and stops to ask you about a 2FA code, a CAPTCHA, or a choice only you can make
An MCP server and Chrome extension that let the agent drive your already open browser with your logged-in sessions, all locally
A Chrome extension with an MCP server: the agent drives your everyday browser with its sessions, tabs, bookmarks and history
An MCP server that lets an agent drive a real browser: open pages, click and fill forms through Playwright
An MCP server that opens pages in headless Chromium through Playwright and hands the agent the main text as Markdown, even from JavaScript-heavy sites
A browser extension and local proxy that bring MCP tools into web chat UIs like ChatGPT, Perplexity and Grok
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Hyperbrowser MCP (this entry) | yes | API key | 791 | |
| Firecrawl MCP Server | yes | API key | 7,672 | the Firecrawl cloud API: Markdown scraping, site maps, crawling and search; Hyperbrowser focuses on browser agents and persistent profiles |
| Browserbase Skills | no | 3,741 | a Claude Code plugin with the browse CLI and the Browserbase cloud, about two dozen skills; Hyperbrowser gives ten MCP tools |
README in short
The documentation is a short table of contents covering installation with `npx`, configurations for Cursor, Windsurf and Claude Desktop, building from source, the tool list and installation through Smithery. The repository holds TypeScript sources with one file per tool, stdio and SSE transports, static and dynamic documentation resources, a Dockerfile and a Smithery configuration. The tree has no tests.