Show HN: Hyperbrowser MCP Server – Connect AI agents to the web through browsers

[shrisukhani]

Hi HN! Excited to share our MCP Server at Hyperbrowser - something we’ve been working on for a few days. We think it’s a pretty neat way to connect LLMs and IDEs like Cursor / Windsurf to the internet.

Our MCP server exposes seven tools for data collection and browsing:

1. `scrape_webpage` - Extract formatted (markdown, screenshot etc) content from any webpage

2. `crawl_webpages` - Navigate through multiple linked pages and extract LLM-friendly formatted content

3. `extract_structured_data` - Convert messy HTML into structured JSON

4. `search_with_bing` - Query the web and get results with Bing search

5. `browser_use_agent` - Fast, lightweight browser automation with the Browser Use agent

6. `openai_computer_use_agent` - General-purpose automation using OpenAI’s CUA model

7. `claude_computer_use_agent` - Complex browser tasks using Claude computer use

You can connect the server to Cursor, Windsurf, Claude desktop, and any other MCP clients with this command `npx -y hyperbrowser-mcp` and a Hyperbrowser API key. We're running this on our cloud browser infrastructure that we've been developing for the past few months – it handles captchas, proxies, and stealth browsing automatically.

Some fun things you can do with it: (1) deep research with claude desktop, (2) summarizing the latest HN posts, (3) creating full applications from short gists in Cursor, (3) automating code review in cursor, (4) generating llms.txt for any website with windsurf, (5) ordering sushi from windsurf (admittedly, this is just for fun - probably not actually going to do this myself).

We're building this server in the open and would love feedback from anyone building agents or working with web automation. If you find bugs or have feature requests, please let us know! One big issue with MCPs in general is that the installation UX sucks and auth credentials have to be hardcoded. We don’t have a solution to this right now but Anthropic seems to be working on something here so excited for that to come out. Love to hear any other complaints / thoughts you have about the server itself, Hyperbrowser, or the installation experience.

You can check us out at https://hyperbrowser.ai or check out the source code at https://github.com/hyperbrowserai/mcp

[xena]

Is there support for robots.txt so service operators can opt out of your mass scraping?

[olivia-l]

Not only do they not respect robots.txt, but they publish an entire page[1] in their docs dedicated to circumventing scraping countermeasures.

I pointed their scraper at a url on my server to test it's behavior. It made four separate requests to the same page, three with the UA "udici"[2], and one with the UA "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/134.0.0.0 Safari/537.36" and a different IP. The first three requests were all made within the span of 1 second, and the fourth 27 seconds later.

I emailed their published support address asking for an IP range and UA. They gave me the entire IP range of google cloud, and ignored the UA question.

This goes well beyond the "it's up to our users to implement responsible scraping practices" implication from the developer's other comment[3]. Instead, their service behaves maliciously by default, and they have implemented and documented switches that users can toggle for additional malicious scraping behavior. As far as I can tell, it is not even possible to implement a robots.txt-respecting scraper on top of this, because I couldn't find any mechanism for users to set a specific UA string.

[1]: https://docs.hyperbrowser.ai/sessions/advanced-privacy-and-a... (archived: https://web.archive.org/web/20250322045952/https://docs.hype..., http://archive.today/2025.03.22-050029/https://docs.hyperbro...)

[2]: https://github.com/nodejs/undici

[3]: https://news.ycombinator.com/item?id=43442116

[soulofmischief]

Would you like to explain how directing your user agent to use the internet just as you would in order to complete a task or solve a problem is "mass scraping"?

[shrisukhani]

No, we don't enforce any robots.txt restrictions ourselves. We also don't do any scraping ourselves. We provide browser infrastructure that operates like any normal browser would - what users choose to do with it is up to them. We're building tools that give AI agents the same web access capabilities that humans have, don't think it's our place to impose any additional limitations.

[TheTaytay]

This looks cool.

1) I looked at the pricing. Is search included in the price - (you just pay credits/browser time?)

2) Can you tell me more about the source of your residential proxies? I am new to this space, so don’t know how people source these legitimately.

Thanks!

[shrisukhani]

Thanks!

1) Yep, you just pay from browser time and proxy usage

2) We use a handful of proxy providers under the hood ourselves. There’s a lot of shady ones but we only work with ones where we’ve vetted the source of. Different providers source proxies in different ways - directly from ISPs, paying end sources for proxies etc

[pizzafeelsright]

Fantastic.

MCPs are showing promise.

[shrisukhani]

Thanks!

And yeah MCP is super promising. We announced this on X and LinkedIn yesterday and the response has been really good. A lot of people with a bunch of use cases.

One surprising thing is there’s also a bunch of semi/non-technical people using our MCP server and the installation experience for them rn just absolutely sucks.

I think once auth and 1-click install are solved, MCP could become the standard way to integrate tools with LLMs

[fosterfriends]

++ love that folks are trying to build companies on MCP. Good luck!

[dennisaxu]

dope