I remember reading a CF blog post about crawler separation and responsible AI bot principles where they argue every bot should have one distinct purpose. Now they're building crawling infrastructure themselves, and their own /crawl endpoint lists "training AI systems" as a use case alongside regular crawling. So not only are they in the crawling business now, they're not following the separation principle. To be fair, there's a business logic here. But it's hard not to notice the irony.
https://blog.cloudflare.com/uk-google-ai-crawler-policy/
I'm surprised that Cloudflare hasn't started hosting a pre-scraped version of websites that use Cloudflare's proxy - something like https://www.example.com/cdn-cgi/cached-contents.json They already have the website content in their cache, so why not just cut out the middle man of scraping services and API's like this and publish it?
Obviously there's good reasons NOT to, but I am surprised they haven't started offering it (as an "on-by-default" option, naturally) yet.
Well, the conversion process into the JSON representation is going to take CPU, and then you have to store the result, in essence doubling your cache footprint.
Doing it on demand still utilizes their cached version, so it saves a trip to the origin, but doesn’t require doubling the cache size. They can still cache the results if the same site is scraped multiple times, but this saves having to cache things that are never going to be requested.
Cache footprint management is a huge factor in the cost and performance for a CDN, you want to get the most out of your storage and you want to serve as many pages from cache as possible.
I know in my experience working for a CDN, we were doing all sorts of things to try to maximize the hit rate for our cache.. in fact, one of the easiest and most effective techniques for increasing cache hit rate is to do the OPPOSITE of what you are suggesting; instead of pre-caching content, you do ‘second hit caching’, where you only store a copy in the cache if a piece of content is requested a second time. The idea is that a lot of content is requested only once by one user, and then never again, so it is a waste to store it in the cache. If you wait until it is requested a second time before you cache it, you avoid those single use pages going into your cache, and don’t hurt overall performance that much, because the content that is most useful to cache is requested a lot, and you only have to make one extra origin request.
> Doing it on demand still utilizes their cached version, so it saves a trip to the origin, but doesn’t require doubling the cache size. They can still cache the results if the same site is scraped multiple times, but this saves having to cache things that are never going to be requested.
Isn't this solving a slightly, but very significantly different problem?
You could serve the very same data in two different ways: One to present to the users and one to hand over to scrapers. Of course, some sites would be too difficult or costly to transform into a common underlying cache format, but people who WANT their sides accessible to scrapers could easily help the process along a bit or serve their site in the necessary format in the first place.
But the key is:
A tool using a "pre-scraped" version of a site has very likely very different requirements of how a CDN caches this site. And this could be easily customizable by those using this endpoint.
Want a free version? Ok, give us the list of all the sites you want, then come back in 10min and grab everything in one go, the data will be kept ready for 60s. Got an API token? 10 free near-real-time request for you and they'll recharge at a rate of 2 per hour. Want to play nice? Ask the CDN to have the requested content ready in 3 hours. Got deep pockets? Pay for just as many real-real-time requests as you need.
What makes this so different is that unless customers are willing to hand over a lot of money, you dont need to cache anything to serve requests at all. Potentially not even later if you got enough capacity to serve the data for scheduled requests from the storage network directly.
You just generate an immediate promise response to the request telling them to come back later. And depending on what you put into that promise, you've got quite a lot of control over the schedule yourself.
- Got a "within 10min" request but your storage network has plenty if capacity in 30s? Just tell them to come back in 30s.
- A customer is pushing new data into your network around 10am and many bots are interested in getting their hands on it as soon as possible, making requests for 10am to 10:05? Just bundle their requests.
- Expected data still not around at 10:05? Unless the bots set an "immediate" flag (or whatever) indicating that they want whatever state the site is in right now, just reply with a second promise when they come back. And a third if necessary... and so on.
Not the same thing, but they have something close (it's not on-by-default, yet) [1]:
> Cloudflare's network now supports real-time content conversion at the source, for enabled zones using content negotiation headers. Now when AI systems request pages from any website that uses Cloudflare and has Markdown for Agents enabled, they can express the preference for text/markdown in the request. Our network will automatically and efficiently convert the HTML to markdown, when possible, on the fly.
They wouldn't, well there's Etag and alike but it still a round trip on level 7 to the origin. However the pattern generally is to say when the content is good to in the Response headers, and cache on that duration, for an example a bitcoin pricing aggregator might say good for 60 seconds (with disclaimers on page that this isn't market data), whilst My Little Town news might say that an article is good for an hour (to allow Updates) and the homepage is good for 5 minutes to allow breaking news article to not appear too far behind.
Based on the post, it seems likely that they'd just delay per the robots.txt policy no matter what, and do a full browser render of the cached page to get the content. Probably overkill for lots and lots of sites. An HTML fetch + readability is really cheap.
Offering wholesale cache dumps blows up every assumption about origin privacy and copyright. Suddenly you are one toggle away from someone else automatically harvesting and reselling your work with Cloudflare as the unwitting middle tier.
You could try to gate this behind access controls but at that point you have reinvented a clunky bespoke CDN API that no site owner asked for, plus a fresh legal mess. Static file caches work because they only ever respond to the original request, not because they claim to own or index your content.
It is a short path from "helpful pre-scraped JSON" to handing an entire site to an AI scraper-for-hire with zero friction. The incentives do not line up unless you think every domain on Cloudflare wants their content wholesale exported by default.
That would prolly work for simple sites, but you still need the dedicated scraping service with a browser to render sites that are more complex (i.e. SPAs)
That was my first thought when I read the headline. It would make perfect sense, and would allow some websites to have best of both worlds: broadcasting content without being crushed by bots. (Not all sites want to broadcast, but many do).
It’s a bit more complicated than that. This is their product Browser Rendering, which runs a real browser that loads the page and executes JavaScript. It’s a bit more involved than a simple curl scraping.
In this case, sure... that said, I've worked on a few sites where more than half the traffic was bots because the content was useful for other sites (classic car classifieds/sales site). The fact that just over half the page requests were actually search query results is what meant a lot of optimization steps in practice... Implementing a "search" database (mongodb and elastic were pretty new at the time), denormalizing a lot of the data structures on the "enterprise" SQL structures for search and display for not logged in users, etc. Heavier caching, donut caching, etc.
It was an interesting and sometimes fun part of my career. Working on a site/application that isn't necessarily a tech site, and that I have a personal interest in was pretty great... some of the pace for sales/commercial features less so, with sales making deals requiring deep integrations on impossible timelines. You learn a lot when a self-hosted site is being kicked while it's down... The cloud migration to get a better use of flexible resources, etc.
That's too funny. If true, really looking forward to the Cloudflare response here. I'm unsure how you would spin that in a way that didn't seem self-serving.
It's very clearly disclosed in the linked docs already, it says that Cloudflare Bot Protection will block it same as all other bots, unless you choose to allow it as an exception. If they didn't do it that way, people would accuse them of either bypassing their own product (possibly anticompetitive) or just having a low quality one.
So it doesn't take any action to work around other bot protections? Feels like that would be on the list of features an AI company wanting to scrape would ask for.
Cloudflare crawl respects robots.txt. It does not attempt to bypass any anti-crawling measures. If the site doesn't want to be crawled -- whether it uses Cloudflare or not -- this product will not help you crawl it.
Some sites actually want crawlers -- e.g. sites that are selling a product, documentation, etc. That's what this product is meant for.
Is this just a way to strong-arm non-cloudflarians into adopting their platform if you don't want your site crawled? It does sound like they are selling the solution to avoid their own content crawler.
fuck firecrawl. they copied my idea by showing interest in my product and then copied it, used their YC money to give it all out for free. fuck nick in particular. I'm still salty over this
"they copied my idea by showing interest in my product and then copied it". What exactly is revolutionary about Firecrawl or your product? Scraping APIs have been around for over a decade.
I was the first to return markdown and use reader mode stuff to strip irrelevant stuff. Theres copying and there's talking to the founder sounding interested to have your team copy what I did in the background. One is fair game, the other is a dick head move.
I think that is a neat idea and it sucks this happened, but how long before somebody simply saw that feature and replicated it? I'm curious, had you considered a deeper moat than that?
This is especially relevant given AI is making this kind of thing easy at an industrial scale. I think we should all be looking for alternative moats.
Sometimes timing is your moat and that's all you need. That being said I'll probably start limiting my public releases to revolve around standards I want implemented.
I'm rethinking the sources of value moats are built around. It seems like the landscape is changing and dimensions such as location, perspective, experience, and attention weigh more than they used to.
> but how long before somebody simply saw that feature and replicated it?
This is a good example. The, idk, "value store" of your org just switched from products and services to the employees who understand your process from a couple angles and can write well.
IMO the under-discussed risk here is that sites will start serving different content to verified crawlers vs real users. You're already seeing it with known search bots getting sanitized views. If your agent's context comes from a crawl the site knows is going to an AI, you have no guarantee it matches what a human sees, and that data quality problem won't surface until your agent starts acting on selectively curated information.
This already has a name in SEO circles: "cloaking" - serving different content to crawlers than to humans. Google has penalized sites for this for years.
The ironic twist is that sites might actually want to serve curated content to AI crawlers - not to deceive, but to control their representation in AI systems. The incentive structure is already there.
The precedent with Google AMP cached content showed what happens: publishers optimized specifically for what the crawler saw, diverging from the actual user experience. The "real" web became secondary.
The real danger isn't fraud - it's gradual optimization pressure. If AI agents increasingly act on crawled data, site operators have every incentive to polish what the crawler sees. You end up with an AI-facing web that diverges from what humans actually experience.
I think the simple explanation is that they weren't selling scraping countermeasures, they were selling web-based denial of service protection (which may be caused by scrapers).
Because the scraper is either impatient, careless or indifferent; and if they scrape for training data they don't plan to come back. If they don't plan to come back they don't care if you tighten up crawling protections after they have moved on. In fact they are probably happy that they got their data and their competition won't
To me the current behavior of those scrapers tells me that "they don't plan", period.
Looks like they hired a bunch of excavators and are digging 2 meters deep on whole fields, looking for nuggets of gold, and pilling the dirt on a huge mountain.
Once they realize the field was bereft of any gold but full of silver?
Or that the gold was actually 2.5 meters deep?
The number of git forges behind Anubis et al and the numerous public announcements should be enough.
Scrappers seem to be exceedingly careless in using public resources. The problem is often not even DDOS (as in overwhelming bandwidth usage) but rather DOS through excessive hits on expensive routes.
Don't need to ask anything i can tell you exactly - because they have no regard for anything but their own profit.
Let me give you an example of this mom and pop shop known as anthropic.
You see they have this thing called claudebot and at least initially it scraped iterating through IP's.
Now you have these things called shared hosting servers, typically running 1000-10000 domains of actual low volume websites on 1-50 or so IPs.
Guess what happens when it is your networks time to bend over? Whole hosting company infrastructure going down as each server has hundreds of claudebots crawling hundreds of vhosts at the same time.
This happened for months. Its the reason they are banned in WAFs by half the hosting industry.
So how would you avoid this specific situation as a web-crawler that tries to be well behaved? You strictly adhere to robots.txt as specified by each domain. The problem is not with any of the sites but the density (1000-10000) by which the hoster packed them. If e.g. the crawler had a 1 sec between page governor even if robots.txt had no rate specified, which to be fair is very reasonable, this packing could still lead to high server load.
That's not the perfect defense you think it is. Plenty of robots.txts[1] technically allow scraping their main content pages as long as your user-agent isn't explicitly disallowed, but in practice they're behind Cloudflare so they still throw up Cloudflare bot check if you actually attempt to crawl.
And forget about crawling. If you have a less reputable IP (basically every IP in third world countries are less reputable, for instance), you can be CAPTCHA'ed to no end by Cloudflare even as a human user, on the default setting, so plenty of site owners with more reputable home/office IPs don't even know what they subject a subset of their users to.
Cloudflare has been trying to mediate publishers & AI companies. If publishers are behind Cloudflare and Cloudflare's bot detection stops scrapers at the request of publishers, the publishers can allow their data to be scraped (via this end point) for a price. It creates market scarcity. I don't believe the target audience is you and me. Unless you own a very popular blog that AI companies would pay you for.
> The /crawl endpoint respects the directives of robots.txt files, including crawl-delay. All URLs that /crawl is directed not to crawl are listed in the response with "status": "disallowed".
You don't need any scraping countermeasures for crawlers like those.
So what’s the user agent for their bot? They don’t seem to specify the default in the docs and it looks like it’s user configurable. So yet another opt out bot which you need your web server to match on special behaviour to block
>So yet another opt out bot which you need your web server to match on special behaviour to block
Given that malicious bots are allegedly spoofing real user agents, "another user agent you have to add to your list" seems like the least of your problems.
Not 'allegedly' - it's just a fact. Even if you're not malicious however it's still sometimes necessary because the server may have different sites for different browsers and check user agents for the experience they deliver. So then even for legitimate purposes you need to at least use the prefix of the user agent that the server expects.
Like they explain in the docs, their crawler will respect the robots.txt dissalowed user-agents, right after the section hat explains how to change your user-agent.
Was it ever not one? They protect a lot of DDoS-for-hire sites from DDoS by their competitors. In return they increase the quantity of DDoS on the internet. They offer you a service for $150, then months later suddenly demand $150k in 24 hours or they shut down your business. If you use them as a DNS registrar they will hold your domain hostage.
yeah, GP completely fails to realize that Cloudflare has always played both sides. that is their entire business model, and it was transparent from the beginning that they would absolutely do the same here.
I think there's some space being absolutely snuffed by the countless bots of everyone, ignoring everything, pulling from residential proxies, and this, supposedly slower, well behavior, smarter bot.
Like there's a difference between dozens of drunk teenagers thrashing the city streets in the illegal street race vs a taxi driver.
If they ever sell or the CEO shifts, yes. For the meantime, they have not given any strong indication that they're trying to bully anybody. I could see things changing drastically if the people in charge are swapped out.
Seems like it was just hours ago they started reaching out to my edge servers from their address space (Me: why is a reverse proxy service banging my servers when I'm not a customer? did some miscreant sign me up somehow?) and it was for Apple, privacy, mom and pie (a VPN service, dressed in noble aspirations). It never quite smelled like pie to me.
If you're doing threat hunting / risk enumeration, Cloudflare is no longer a passive service that miscreants hide behind, they now actively reach out and grab your privates.
The idea of exposing a structured crawl endpoint feels like a natural evolution of robots.txt and sitemaps.
If more sites provided explicit machine-readable entry points for crawlers, indexing could become a lot less wasteful. Right now crawlers spend a lot of effort rediscovering the same structure over and over.
It also raises interesting questions about whether sites will eventually provide different views for humans vs. automated agents in a more formalized way.
I expect that if we still used REST indexing would be even less wasteful.
I've found myself falling pretty hard on the side of making APIs work for humans and expecting LLM providers to optimize around that. I don't need an MCP for a CLI tool, for example, I just need a good man page or `--help` documentation.
I know in practice it no longer is the case, if it ever was.
But semantic HTML is exactly that explicit machine-readable entrypoint. I am firmly entrenched in the opinion that HTML, and the DOM is only for machines to read, it just happens to be also somewhat understandable to some humans. Take an average webpage, have a look at all characters(bytes) in there: often two third won't ever be shown to humans.
Point being: we don't need to invent something new. We just need to realize we already have it and use it correctly. Other than this requiring better understanding of web tech, it has no downsides.
The low hanging fruit being the frameworks out there that should really do a better job of leveraging semantics in their output.
The only ones benefitting from 'wastefull' crawling are the anti-bot solution vendors. Everyone else is incentivized to crawl as efficiently as possible.
I yearn for the days when a single kb get was enough. Now it's endless wastage spawning entire browsers larger than operating systems with mitigations, hacks and proxies. Requesting access directly from webmasters is only met with silence. All of my once simple, hobbyist programs are now bloated beyond belief and less reliable than ever
> It also raises interesting questions about whether sites will eventually provide different views for humans vs. automated agents in a more formalized way.
This question raises an interesting question about if this would exacerbate supply chain injection attacks. Show the innocuous page to the human, another to the bot.
With google covering only 3% I wonder how much people still care and if they should. Funny: I own and know sites that are by far the best resource on the topic but shouldn't have so many links google says. It's like I ask you for a page about cuban chains then you say you don't have it because they had to many links. Or your greengrocer suddenly doesn't have apples because his supplier now offers more than 5 different kinds so he will never buy there again.
If anyone is taking feature requests, could you add an option to return the snapshot as an MHTML with all static assets embedded? (I know this could get inefficient from a storage perspective, but if it really matters you could dedupe assets on your end, which is what my janky homegrown crawler does.)
It seems like there's a missed use case: web archiving. I don't see any mention of WARC as an output format. This could be useful to journalists and academically if they had it.
And while at it, ability to mount the resulting archives at some virtual root in nginx|apache. E.g. serve site-archive.extension at /somepath/site. And standalone simple webserver that can take one or more archives from command line and servers them.
Could they collaborate with the website's creators that have websites behind cloudfare to allow their content to be accessed via an API in exchange of a compensation?.
This could be a way to compensate creators and AI companies be able to access content that's unreachable as it's protected by cloudfare
Oh man, I was hoping I could offer a nicely-crawled version of my site. It would be cool if they offered that for site admins. Then everyone who wanted to crawl would just get a thing they could get for pure transfer cost. I suppose I could build one by submitting a crawl job against myself and then offering a `static.` subdomain on each thing that people could access. Then it's pure HTML instant-load.
I don’t really get the usecase. Is your site static? Then you should just render it to html files and host the static files. And if it’s not static, how would a snapshot of the pages help if they change later? And also why not just add some caching to the site then?
I tried to make exactly this a year ago. Built on Cloudflare using all of their primitives: https://crawlspace.dev -- It didn't work too well (so don't bother trying it).
Worth noting: origin owners can still detect and block CF Browser Rendering requests if needed.
Workers-originated requests include a CF-Worker header identifying the workers subdomain, which distinguishes them from regular CDN proxying. You can match on this in a WAF rule or origin middleware.
The trickier issue: rendered requests originate from Cloudflare ASN 13335 with a low bot score, so if you rely on CF bot scores for content protection, requests through their own crawl product will bypass that check. The practical defense is application-layer rate limiting and behavioral analysis rather than network-level scores -- which is better practice regardless.
The structural conflict is real but similar to search engines offering webmaster tools while running the index. The incentives are misaligned, but the individual products have independent utility. The harder question is whether the combination makes it meaningfully harder to build effective bot protection on top of their platform.
The main win here is abstracting away browser context lifecycle management. Anyone who has run Puppeteer on Workers knows the pain of handling cold starts, context reuse, and timeout cascading across navigation steps. Having crawl() bundle render-then-extract into one call covers maybe 80% of scraping use cases. The remaining 20% where you need request interception or pre-render script injection still needs the full Browser Rendering API, but for pulling structured data from public pages this is a big simplification over managing session state yourself.
Interesting... I built an MCP server for their initial browser render as markdown, and I just tell the LLM to follow reasonable links to relative content, and recurse the tool.
Really hard to understand costs here. What is a reasonable pages per second? Should I assume with politeness that I'm basically at 1 page per second == 3600 pages/hour? Seems painfully slow.
Awesome, so I no longer have to use Firecrawl or my own crawler to scrape entire websites for an agent? Especially when needing residential proxies to do so on Cloudflare protected sites? Why though?
I have tried theirs... they are NOT proxies.. that means majority of the popular sites actually block scraping... even if they are protected by cloudflare itself.
Instead of "should have been an email" this is "should have been a prompt" and can be run locally instead. There are a number of ways to do this from a linux terminal.
```
write a custom crawler that will crawl every page on a site (internal links to the original domain only, scroll down to mimic a human, and save the output as a WebP screenshot, HTML, Markdown, and structured JSON. Make it designed to run locally in a terminal on a linux machine using headless Google Chrome and take advantage of multiple cores to run multiple pages simultaneously while keeping in mind that it might have to throttle if the server gets hit too fast from the same IP.
```
Might use available open source software such as python, playwright, beautifulsoup4, pillow, aiofiles, trafilatura
This presumably is going to be cheap and effective. Its much easier to wrap a prompt round this and know it works that mess around with crawling it all yourself.
You'll still be hand-rolling it if you want to disrespect crawling requirements though.
I’ve actually written a crawler like that before, and still ended up going with Firecrawl for a more recent project. There’s just so many headaches at scale: OOMs from heavy pages, proxies for sites that block cloud IPs, handling nested iframes, etc.
The web was built to be open and available to everyone. Serving static HTML from disk back in the day, nobody could hurt you because there was nothing to hurt.
We need bot protection now because everything is dynamic, straight from the database with some light caching for hot content. When Facebook decides to recrawl your one million pages in the same instant, you're very much up shit creek without a paddle. A bot that crawls the full site doesn't steal anything, but it does take down the origin server. My clients never call me upset that a bot read their blog posts. They call because the bot knocked the site offline for paying customers.
Bot protection protects availability, not secrecy.
And the real bot problem isn't even crawling. It's automated signups. Fake accounts messaging your users. Bots buying out limited drops before a human can load the page. Like-farming. Credential stuffing. That's what bot protection is actually for: preventing fraud, not preventing someone from reading your public website.
Cloudflare's `/crawl` respects robots.txt. Don't want your content crawled, opt out. But if you want it indexed and can't handle the traffic spike, this gets your content out without hammering production.
As for the folks saying Cloudflare should keep blocking all crawlers forever: AI agents already drive real browsers. They click, scroll, render JavaScript. Go look at what browser automation frameworks can do today and then explain to me how you tell a bot from a person. That distinction is already gone. The hot takes are about a version of the internet that doesn't exist anymore.
What % of sites have a content update volume that exceeds what you can get respecting crawl delay?
If your delay is 1s and you publish less than 60 updates a minute on average I can still get 100%. Most crawls are not that latency sensitive, certainly not the ai ones.
HFT bots, now that is an entirely different ballgame.
> Most crawls are not that latency sensitive, certainly not the ai ones.
They certainly behave like they are. We constantly see crawlers trying to do cache busting, for pages that hasn't change in days, if not weeks. It's hard to tell where the bots are coming from theses days, as most have taken to just lie and say that they are Chrome.
I'd agree that the respecting robots.txt makes this a non-starter for the problematic scrapers. These are bots that that will hammer a site into the ground, they don't respect robots.txt, especially if it tells them to go away.
All of this would be much less of a problem if the authors of the scrapers actually knew how to code, understood how the Internet works and had just the slightest bit of respect for others, but they don't so now all scrapers are labeled as hostile, meaning that only the very largest companies, like Google, get special access.
Not really, given that the work we do in that direction isn't exactly public. You can recreate the scenario though. Spin up a wiki of some sort, scrapers love wikis, ideally enable some form of caching, and just sit back and watch scrapers throw random shit in the URL parameters.
Honestly, it feels like cloudflare bullying other sites into using their anti-bot services. great business model by charging owners and devs at the same time. Using AI per page to parse content. its reckless.
The most egregious thing Perplexity did was to straight up ignore robots.txt. Cloudflare promise not to do that, so if we take their word for it, it's a quite different setup.
That said, I'm not fan of letting users forge whatever user agents they please. Instead, AIUI to opt-out of getting crawled I have to look for the existence of certain request headers[1].
Further down they also mention that the requests come from CFs ASN and are branded with identifying headers, so third party filters could easily block them too if they're so inclined. Seems reasonable enough.
If this does bypass their own (and others') anti-AI crawl measures, it'd basically mean that the only people who can't crawl are those without money.
We're creating an internet that is becoming self-reinforcing for those who already have power and harder for anyone else. As crawling becomes difficult and expensive, only those with previously collected datasets get to play. I certainly understand individual sites wanting to limit access, but it seems unlikely that they're limiting access to the big players - and maybe even helping them since others won't be able to compete as well.
The big question here is this a verified-bot on the Cloudflare WAF? Didn't Google get into trouble for using their search engine user agent and IPs to feed Gemini in Europe?
First, the Cloudflare Crawl endpoint does not require the target site to use Cloudflare. It spins up a headless Chrome instance (via the Browser Rendering API) that fetches and renders any publicly accessible URL. You could crawl a site hosted on Hetzner or a bare VPS with the same call.
Second on pricing: Browser Rendering is only available on the Workers Paid plan ($5/month). It is not part of the free tier. Usage is billed per invocation beyond the included quota - the exact limits are in the Cloudflare docs under Browser Rendering pricing, but for archival use cases with moderate crawl rates you are very unlikely to run into meaningful costs.
The practical gotcha for forum archival is pagination and authentication-gated content. If the forum requires a login to see older posts, a headless browser session with saved cookies would help, but that is more complex to orchestrate than a single-shot fetch.
> Browser Rendering is only available on the Workers Paid plan ($5/month). It is not part of the free tier.
The post says it's available for both free and paid plans. According to the pricing page of the Browser Rendering, the free plan will have 10 minutes/day browsing time.
I've used browser rendering at work and it's quite nice. Most solutions in the crawling space are kind of scummy and designed for side-stepping robots.txt and not being a good citizen. A crawl endpoint is a very necessary addition!
All what was expected, first they do a huge campaign to out evil scrapers. We should use their service to ensure your website block LLMs and bots to come scraping them. Look how bad it is.
And once that is well setup, and they have their walled garden, then they can present their own API to scrape websites. All well done to be used by your LLM. But as you know, they are the gate keeper so that the Mafia boss decide what will be the "intermediary" fee that is proper for itself to let you do what you were doing without intermediary before.
That is funny because on this page there is a warning block with the following text:
Refer to Will Browser Rendering bypass Cloudflare's Bot Protection? for instructions on creating a WAF skip rule.
And "Will Browser Rendering bypass Cloudflare's Bot Protection? " is a hash link to the FAQ page, that surprisingly doesn't anything available for this link entry.
Is it because it was removed (/hidden) or because it is not yet available until everyone forget the "we are no evil, we are here to protect the internet"?
most websites, particularly those behind cloudflare, are very restrictive even to crawlers that obey robots. Proof: a ton of my time over the last year, and my crawlers very carefully obey robots.
It's hard to see how this isn't extorting folks by offering a working solution that, oh, cloudflare doesn't block. As long as you pay Cloudflare.
Perhaps I'm overly cynical, but I'd be quite surprised if cloudflare subjected their own headless browsing to the same rules the rest of the internet gets.
>most websites, particularly those behind cloudflare, are very restrictive even to crawlers that obey robots. Proof: a ton of my time over the last year, and my crawlers very carefully obey robots.
The docs are pretty equivocal though:
>If you use Cloudflare products that control or restrict bot traffic such as Bot Management, Web Application Firewall (WAF), or Turnstile, the same rules will apply to the Browser Rendering crawler.
It's not just robots.txt. Most (all?) restrictions that apply to outside bots apply to cloudflare's bot as well, at least that's what they're claiming. If they're being this explicit about it, I'm willing to give them the benefit of the doubt until there's evidence to the contrary, rather than being a cynic and assuming the worst.
I had the idea after buying https://mirror.forum recently (which I talked in discord and archiveteam irc servers) that I wanted to preserve/mirror forums (especially tech) related [Think TinyCoreLinux] since Archive.org is really really great but I would prefer some other efforts as well within this space.
I didn't want to scrape/crawl it myself because I felt like it would feel like yet another scraping effort for AI and strain resources of developers.
And even when you want to crawl, the issue is that you can't crawl cloudflare and sometimes for good measure.
So in my understanding, can I use Cloudflare Crawl to essentially crawl the whole website of a forum and does this only work for forums which use cloudflare ?
Also what is the pricing of this? Is it just a standard cloudflare worker so would I get free 100k requests and 1 Million per the few cents (IIRC) offer for crawling. Considering that Cloudflare is very scalable, It might even make sense more than buying a group of cheap VPS's
Also another point but I was previously thinking that the best way was probably if maintainers of these forums could give me a backup archive of the forum in a periodic manner as my heart believes it to be most cleanest way and discussing it on Linux discord servers and archivers within that community and in general, I couldn't find anyone who maintains such tech forums who can subscribe to the idea of sharing the forum's public data as a quick backup for preservation purposes. So if anyone knows or maintains any forums myself. Feel free to message here in this thread about that too.
I actually don't but it seems that cloudflare caches responses so if anything instead of straining the developer resources, it would strain more cloudflare resources and cloudflare could better handle that more efficiently with their own crawl product.
Also, I am genuinely open to feedback (Like a lot) so just let me know if you know of any other alternative too for the particular thing that I wish to create and I would love to have a discussion about that too! I genuinely wish that there can be other ways and part of the reason why I wrote that comment was wishing that someone who manages forums or knows people who do can comment back and we can have a discussion/something-meaningful!
I am also happy with you also suggesting me any good use cases of the domain in general if there can be made anything useful with it. In fact, I am happy with transferring this domain to you if this is something which is useful to ya or anyone here (Just donate some money preferably 50-100$ to any great charity in date after this comment is made and mail me details and I am absolutely willing to transfer the domain, or if you work in any charity currently and if it could help the charity in any meaningful manner!)
I had actually asked archive team if I could donate the domain to them if it would help archive.org in any meaningful way and they essentially politely declined.
I just bought this domain because someone on HN said mirror.org when they wanted to show someone else mirror and saw the price of the .org domain being so high (150k$ or similar)and I have habit of finding random nice TLD and I found mirror.forum so I bought it
And I was just thinking of hmm what can be a decent idea now that I have bought it and had thought of that. Obviously I have my flaws (many actually) but I genuinely don't wish any harm to anybody especially those people who are passionate about running independent forums in this centralized-web. I'd rather have this domain be expired if its activation meant harm to anybody.
This is used to scrape third-party sites not necessarily behind cloudflare so it has nothing to do with whether cloudflare caches it or not plus when using their browser rendering it doesn't even fetch cached responses anyways....
I didn't know that it doesn't fetch catched responses, my apologies. I had only read through it with a glance and it felt like something that cloudflare might've done. Is there any particular reason that they don't use the cached responses, feels like a missed opportunity but maybe I am missing something?
It's a browser rendering API which means people are paying a premium specifically to have a browser render a live website. If you want to get a cached response of a page and still possibly get blocked by cloudflare you could just make a node script with a simple fetch and save your money.
Off-topic, but I'm having a terrible experience with Cloudflare and would love to know if someone could offer some help.
All of a sudden, about 1/3 of all traffic to our website is being routed via EWR (New York) - me included -, even tough all our users and our origin servers are in Brazil.
We pay for the Pro plan but support has been of no help: after 20 days of 'debugging' and asking for MTRs and traceroutes, they told us to contact Claro (which is the same as telling me to contact Verizon) because 'it's their fault'.
Do you think cloudflare is responsible for all of the network traffic routing in the entire world and can simply fix any problem even if it's on somebody else's network?
No. I do think that Cloudflare is a great company and got where it's at today because they care for this type of issue, and has a much better chance of contacting their peering traffic partner than me because they take care of ~20% of all internet traffic, while I take care of none.
I remember reading a CF blog post about crawler separation and responsible AI bot principles where they argue every bot should have one distinct purpose. Now they're building crawling infrastructure themselves, and their own /crawl endpoint lists "training AI systems" as a use case alongside regular crawling. So not only are they in the crawling business now, they're not following the separation principle. To be fair, there's a business logic here. But it's hard not to notice the irony. https://blog.cloudflare.com/uk-google-ai-crawler-policy/
I'm surprised that Cloudflare hasn't started hosting a pre-scraped version of websites that use Cloudflare's proxy - something like https://www.example.com/cdn-cgi/cached-contents.json They already have the website content in their cache, so why not just cut out the middle man of scraping services and API's like this and publish it?
Obviously there's good reasons NOT to, but I am surprised they haven't started offering it (as an "on-by-default" option, naturally) yet.
Well, the conversion process into the JSON representation is going to take CPU, and then you have to store the result, in essence doubling your cache footprint.
Doing it on demand still utilizes their cached version, so it saves a trip to the origin, but doesn’t require doubling the cache size. They can still cache the results if the same site is scraped multiple times, but this saves having to cache things that are never going to be requested.
Cache footprint management is a huge factor in the cost and performance for a CDN, you want to get the most out of your storage and you want to serve as many pages from cache as possible.
I know in my experience working for a CDN, we were doing all sorts of things to try to maximize the hit rate for our cache.. in fact, one of the easiest and most effective techniques for increasing cache hit rate is to do the OPPOSITE of what you are suggesting; instead of pre-caching content, you do ‘second hit caching’, where you only store a copy in the cache if a piece of content is requested a second time. The idea is that a lot of content is requested only once by one user, and then never again, so it is a waste to store it in the cache. If you wait until it is requested a second time before you cache it, you avoid those single use pages going into your cache, and don’t hurt overall performance that much, because the content that is most useful to cache is requested a lot, and you only have to make one extra origin request.
> Doing it on demand still utilizes their cached version, so it saves a trip to the origin, but doesn’t require doubling the cache size. They can still cache the results if the same site is scraped multiple times, but this saves having to cache things that are never going to be requested.
Isn't this solving a slightly, but very significantly different problem?
You could serve the very same data in two different ways: One to present to the users and one to hand over to scrapers. Of course, some sites would be too difficult or costly to transform into a common underlying cache format, but people who WANT their sides accessible to scrapers could easily help the process along a bit or serve their site in the necessary format in the first place.
But the key is:
A tool using a "pre-scraped" version of a site has very likely very different requirements of how a CDN caches this site. And this could be easily customizable by those using this endpoint.
Want a free version? Ok, give us the list of all the sites you want, then come back in 10min and grab everything in one go, the data will be kept ready for 60s. Got an API token? 10 free near-real-time request for you and they'll recharge at a rate of 2 per hour. Want to play nice? Ask the CDN to have the requested content ready in 3 hours. Got deep pockets? Pay for just as many real-real-time requests as you need.
What makes this so different is that unless customers are willing to hand over a lot of money, you dont need to cache anything to serve requests at all. Potentially not even later if you got enough capacity to serve the data for scheduled requests from the storage network directly.
You just generate an immediate promise response to the request telling them to come back later. And depending on what you put into that promise, you've got quite a lot of control over the schedule yourself.
- Got a "within 10min" request but your storage network has plenty if capacity in 30s? Just tell them to come back in 30s.
- A customer is pushing new data into your network around 10am and many bots are interested in getting their hands on it as soon as possible, making requests for 10am to 10:05? Just bundle their requests.
- Expected data still not around at 10:05? Unless the bots set an "immediate" flag (or whatever) indicating that they want whatever state the site is in right now, just reply with a second promise when they come back. And a third if necessary... and so on.
Not the same thing, but they have something close (it's not on-by-default, yet) [1]:
> Cloudflare's network now supports real-time content conversion at the source, for enabled zones using content negotiation headers. Now when AI systems request pages from any website that uses Cloudflare and has Markdown for Agents enabled, they can express the preference for text/markdown in the request. Our network will automatically and efficiently convert the HTML to markdown, when possible, on the fly.
[1] https://blog.cloudflare.com/markdown-for-agents/
Interesting - its sounds like this could be combined with some creative cache parsing on their side to provide this feature to sites that want it.
> I'm surprised that Cloudflare hasn't started hosting a pre-scraped version of websites that use Cloudflare's proxy
It's entirely possible that they're doing this under the hood for cases where they can clearly identify the content they have cached is public.
How would they know the content hasn’t changed without hitting the website?
They wouldn't, well there's Etag and alike but it still a round trip on level 7 to the origin. However the pattern generally is to say when the content is good to in the Response headers, and cache on that duration, for an example a bitcoin pricing aggregator might say good for 60 seconds (with disclaimers on page that this isn't market data), whilst My Little Town news might say that an article is good for an hour (to allow Updates) and the homepage is good for 5 minutes to allow breaking news article to not appear too far behind.
Keeping track of when content changes is literally the primary function of a CDN.
Caching headers?
(Which, on Akamai, are by default ignored!)
Based on the post, it seems likely that they'd just delay per the robots.txt policy no matter what, and do a full browser render of the cached page to get the content. Probably overkill for lots and lots of sites. An HTML fetch + readability is really cheap.
Offering wholesale cache dumps blows up every assumption about origin privacy and copyright. Suddenly you are one toggle away from someone else automatically harvesting and reselling your work with Cloudflare as the unwitting middle tier.
You could try to gate this behind access controls but at that point you have reinvented a clunky bespoke CDN API that no site owner asked for, plus a fresh legal mess. Static file caches work because they only ever respond to the original request, not because they claim to own or index your content.
It is a short path from "helpful pre-scraped JSON" to handing an entire site to an AI scraper-for-hire with zero friction. The incentives do not line up unless you think every domain on Cloudflare wants their content wholesale exported by default.
I think Common Crawl already offers this, although it's free: https://commoncrawl.org/
That would prolly work for simple sites, but you still need the dedicated scraping service with a browser to render sites that are more complex (i.e. SPAs)
That was my first thought when I read the headline. It would make perfect sense, and would allow some websites to have best of both worlds: broadcasting content without being crushed by bots. (Not all sites want to broadcast, but many do).
It’s a bit more complicated than that. This is their product Browser Rendering, which runs a real browser that loads the page and executes JavaScript. It’s a bit more involved than a simple curl scraping.
So does that mean it can replace serpapi or similar?
But think about poor phishers and malware devs protected by Cloudflare.
Doesn't work for pages protected by cloudflare in my experience. What a shame, they could've produced the problem and sold the solution.
That’s what they are doing. This is a textbook protection racket.
“Buy Cloudflare bot protection, otherwise it would be a shame if your site got scraped and ddos’d.”
Who is doing the scraping and ddosing? Cloudflare.
You can trivially block Cloudflare crawl via robots.txt. You don't need to buy Cloudflare's bot protection -- this is not a malicious bot.
https://x.com/CloudflareDev/status/2031745285517455615
(Disclosure: I work for Cloudflare but not on this product. I get pretty tired of the conspiracy theories TBH.)
In this case, sure... that said, I've worked on a few sites where more than half the traffic was bots because the content was useful for other sites (classic car classifieds/sales site). The fact that just over half the page requests were actually search query results is what meant a lot of optimization steps in practice... Implementing a "search" database (mongodb and elastic were pretty new at the time), denormalizing a lot of the data structures on the "enterprise" SQL structures for search and display for not logged in users, etc. Heavier caching, donut caching, etc.
It was an interesting and sometimes fun part of my career. Working on a site/application that isn't necessarily a tech site, and that I have a personal interest in was pretty great... some of the pace for sales/commercial features less so, with sales making deals requiring deep integrations on impossible timelines. You learn a lot when a self-hosted site is being kicked while it's down... The cloud migration to get a better use of flexible resources, etc.
That's too funny. If true, really looking forward to the Cloudflare response here. I'm unsure how you would spin that in a way that didn't seem self-serving.
It's very clearly disclosed in the linked docs already, it says that Cloudflare Bot Protection will block it same as all other bots, unless you choose to allow it as an exception. If they didn't do it that way, people would accuse them of either bypassing their own product (possibly anticompetitive) or just having a low quality one.
So it doesn't take any action to work around other bot protections? Feels like that would be on the list of features an AI company wanting to scrape would ask for.
No, it does not take any action to work around other bot protections.
https://x.com/CloudflareDev/status/2031745285517455615
(Disclosure: I work for Cloudflare but not on this product.)
Cloudflare crawl respects robots.txt. It does not attempt to bypass any anti-crawling measures. If the site doesn't want to be crawled -- whether it uses Cloudflare or not -- this product will not help you crawl it.
Some sites actually want crawlers -- e.g. sites that are selling a product, documentation, etc. That's what this product is meant for.
https://x.com/CloudflareDev/status/2031745285517455615
(Disclosure: I work for Cloudflare but not on this product.)
I imagine that would cause a backlash from the website owners trusting cloudflare to keep their content 'safe'
Wait. What?
Is this just a way to strong-arm non-cloudflarians into adopting their platform if you don't want your site crawled? It does sound like they are selling the solution to avoid their own content crawler.
As long at it gets past Azure's bot protection ...
Came here to write this. I am getting much better results from Firecrawl (not affiliated with them, just a happy customer).
As someone who helps keep a site online with a lot of content, I have mixed feelings on Firecrawl.
On one hand, their bots seem much more well behaved than others.
However, running a crawler fleet which is deceptive and evasive in its identification and don't honor REP is no way to build a business.
I'd love for you to kick the tires on https://grubcrawler.dev
fuck firecrawl. they copied my idea by showing interest in my product and then copied it, used their YC money to give it all out for free. fuck nick in particular. I'm still salty over this
"they copied my idea by showing interest in my product and then copied it". What exactly is revolutionary about Firecrawl or your product? Scraping APIs have been around for over a decade.
I was the first to return markdown and use reader mode stuff to strip irrelevant stuff. Theres copying and there's talking to the founder sounding interested to have your team copy what I did in the background. One is fair game, the other is a dick head move.
Not sure about the first claim. But yes, talking to the founder, sharing details and having it stolen is not a good look. Sorry that happened to you.
I think that is a neat idea and it sucks this happened, but how long before somebody simply saw that feature and replicated it? I'm curious, had you considered a deeper moat than that?
This is especially relevant given AI is making this kind of thing easy at an industrial scale. I think we should all be looking for alternative moats.
Sometimes timing is your moat and that's all you need. That being said I'll probably start limiting my public releases to revolve around standards I want implemented.
I'm rethinking the sources of value moats are built around. It seems like the landscape is changing and dimensions such as location, perspective, experience, and attention weigh more than they used to.
> but how long before somebody simply saw that feature and replicated it?
This is a good example. The, idk, "value store" of your org just switched from products and services to the employees who understand your process from a couple angles and can write well.
Tell more. Crawling is not a new idea. How did they abuse you?
Please tells me you are joking
IMO the under-discussed risk here is that sites will start serving different content to verified crawlers vs real users. You're already seeing it with known search bots getting sanitized views. If your agent's context comes from a crawl the site knows is going to an AI, you have no guarantee it matches what a human sees, and that data quality problem won't surface until your agent starts acting on selectively curated information.
This could go wrong on same levels.
This already has a name in SEO circles: "cloaking" - serving different content to crawlers than to humans. Google has penalized sites for this for years.
The ironic twist is that sites might actually want to serve curated content to AI crawlers - not to deceive, but to control their representation in AI systems. The incentive structure is already there.
The precedent with Google AMP cached content showed what happens: publishers optimized specifically for what the crawler saw, diverging from the actual user experience. The "real" web became secondary.
The real danger isn't fraud - it's gradual optimization pressure. If AI agents increasingly act on crawled data, site operators have every incentive to polish what the crawler sees. You end up with an AI-facing web that diverges from what humans actually experience.
This already happens in the opposite direction. See: news websites that drop their pay wall for GoogleBot
Is cloudflare becoming a mob outfit? Because they are selling scraping countermeasures but are now selling scraping too.
And they can pull it off because of their reach over the internet with the free DNS.
I think the simple explanation is that they weren't selling scraping countermeasures, they were selling web-based denial of service protection (which may be caused by scrapers).
This was always also sold as bot protection and anti-scraping / crawling features like https://www.cloudflare.com/lp/pg-ai-crawl-control/
Ask yourself, why would a scraper ddos? Why would a ddos-protection vendor ddos?
Because the scraper is either impatient, careless or indifferent; and if they scrape for training data they don't plan to come back. If they don't plan to come back they don't care if you tighten up crawling protections after they have moved on. In fact they are probably happy that they got their data and their competition won't
> they don't plan to come back
To me the current behavior of those scrapers tells me that "they don't plan", period.
Looks like they hired a bunch of excavators and are digging 2 meters deep on whole fields, looking for nuggets of gold, and pilling the dirt on a huge mountain.
Once they realize the field was bereft of any gold but full of silver? Or that the gold was actually 2.5 meters deep?
They have to go through everything again.
The number of git forges behind Anubis et al and the numerous public announcements should be enough.
Scrappers seem to be exceedingly careless in using public resources. The problem is often not even DDOS (as in overwhelming bandwidth usage) but rather DOS through excessive hits on expensive routes.
Ask yourself, why would everyone except you say that they do?
> Ask yourself, why would a scraper ddos?
Don't need to ask anything i can tell you exactly - because they have no regard for anything but their own profit.
Let me give you an example of this mom and pop shop known as anthropic.
You see they have this thing called claudebot and at least initially it scraped iterating through IP's.
Now you have these things called shared hosting servers, typically running 1000-10000 domains of actual low volume websites on 1-50 or so IPs.
Guess what happens when it is your networks time to bend over? Whole hosting company infrastructure going down as each server has hundreds of claudebots crawling hundreds of vhosts at the same time.
This happened for months. Its the reason they are banned in WAFs by half the hosting industry.
So how would you avoid this specific situation as a web-crawler that tries to be well behaved? You strictly adhere to robots.txt as specified by each domain. The problem is not with any of the sites but the density (1000-10000) by which the hoster packed them. If e.g. the crawler had a 1 sec between page governor even if robots.txt had no rate specified, which to be fair is very reasonable, this packing could still lead to high server load.
Their free DNS is only a small piece of the pie.
The fact that 30%+ of the web relies on their caching services, routablility services and DDoS protection services is the main pull.
Their DNS is only really for data collection and to front as "good will"
> The fact that 30%+ of the web relies on their caching services
30% of the web might use their caching services. 'Relies on' implies that it wouldn't work without them, which I doubt is the case.
It might be the case for the biggest 1% of that 30%. But not the whole lot.
>'Relies on' implies that it wouldn't work without them
Last time Cloudflare went down, their dashboard was also unavailable, so you couldn't turn off their proxy service anyway.
No: https://developers.cloudflare.com/browser-rendering/rest-api...
That's not the perfect defense you think it is. Plenty of robots.txts[1] technically allow scraping their main content pages as long as your user-agent isn't explicitly disallowed, but in practice they're behind Cloudflare so they still throw up Cloudflare bot check if you actually attempt to crawl.
And forget about crawling. If you have a less reputable IP (basically every IP in third world countries are less reputable, for instance), you can be CAPTCHA'ed to no end by Cloudflare even as a human user, on the default setting, so plenty of site owners with more reputable home/office IPs don't even know what they subject a subset of their users to.
[1] E.g. https://www.wired.com/robots.txt to pick an example high up on HN front page.
Cloudflare has been trying to mediate publishers & AI companies. If publishers are behind Cloudflare and Cloudflare's bot detection stops scrapers at the request of publishers, the publishers can allow their data to be scraped (via this end point) for a price. It creates market scarcity. I don't believe the target audience is you and me. Unless you own a very popular blog that AI companies would pay you for.
Next step will be their default "free" anti-bot denying all but their own bot. They know full well nearly nobody changes the default.
no? it takes 10 seconds to check:
> The /crawl endpoint respects the directives of robots.txt files, including crawl-delay. All URLs that /crawl is directed not to crawl are listed in the response with "status": "disallowed".
You don't need any scraping countermeasures for crawlers like those.
So what’s the user agent for their bot? They don’t seem to specify the default in the docs and it looks like it’s user configurable. So yet another opt out bot which you need your web server to match on special behaviour to block
Isn't this covered here? https://developers.cloudflare.com/browser-rendering/referenc...
No, hence all their examples using User-Agent: *
>So yet another opt out bot which you need your web server to match on special behaviour to block
Given that malicious bots are allegedly spoofing real user agents, "another user agent you have to add to your list" seems like the least of your problems.
It is cloudflare who made the claim that they are well behaved unlike those other bots and that their behaviour can be controlled by robots.txt
If I need to treat cloudflare bots the same as malicious bots, that undermines their claim.
Not 'allegedly' - it's just a fact. Even if you're not malicious however it's still sometimes necessary because the server may have different sites for different browsers and check user agents for the experience they deliver. So then even for legitimate purposes you need to at least use the prefix of the user agent that the server expects.
Like they explain in the docs, their crawler will respect the robots.txt dissalowed user-agents, right after the section hat explains how to change your user-agent.
Was it ever not one? They protect a lot of DDoS-for-hire sites from DDoS by their competitors. In return they increase the quantity of DDoS on the internet. They offer you a service for $150, then months later suddenly demand $150k in 24 hours or they shut down your business. If you use them as a DNS registrar they will hold your domain hostage.
Where can I learn more about the 150k in 24h?
I imagine it's referencing this story:
https://robindev.substack.com/p/cloudflare-took-down-our-web...
HN Discussion:
https://news.ycombinator.com/item?id=40481808
yeah, GP completely fails to realize that Cloudflare has always played both sides. that is their entire business model, and it was transparent from the beginning that they would absolutely do the same here.
I think there's some space being absolutely snuffed by the countless bots of everyone, ignoring everything, pulling from residential proxies, and this, supposedly slower, well behavior, smarter bot.
Like there's a difference between dozens of drunk teenagers thrashing the city streets in the illegal street race vs a taxi driver.
They always have been.
They also use their dominant position to apply political pressure when they don’t like how a country chooses to run things.
So yeah, we’ve created another mega corp monster that will hurt for years to come.
Well this scraper honours robots.txt so I'm sure most AI crawlers will find it useless.
For a long time cloudflare has proudly protected DDoS-as-a-service sites (but of course, they claim they don't "host" them)
Are you using the word "claim" to call them wrong or for a more confusing reason?
Because I'm pretty sure they are not in fact wrong.
The distinction between a caching proxy and an origin server is pretty meaningless when you're serving static content, if you ask me.
There's a blurry line there, true.
On the other hand when a page is small and static enough that it's basically just a flyer, I also care a lot less about who hosts it.
If they ever sell or the CEO shifts, yes. For the meantime, they have not given any strong indication that they're trying to bully anybody. I could see things changing drastically if the people in charge are swapped out.
Seems like it was just hours ago they started reaching out to my edge servers from their address space (Me: why is a reverse proxy service banging my servers when I'm not a customer? did some miscreant sign me up somehow?) and it was for Apple, privacy, mom and pie (a VPN service, dressed in noble aspirations). It never quite smelled like pie to me.
If you're doing threat hunting / risk enumeration, Cloudflare is no longer a passive service that miscreants hide behind, they now actively reach out and grab your privates.
The idea of exposing a structured crawl endpoint feels like a natural evolution of robots.txt and sitemaps.
If more sites provided explicit machine-readable entry points for crawlers, indexing could become a lot less wasteful. Right now crawlers spend a lot of effort rediscovering the same structure over and over.
It also raises interesting questions about whether sites will eventually provide different views for humans vs. automated agents in a more formalized way.
I expect that if we still used REST indexing would be even less wasteful.
I've found myself falling pretty hard on the side of making APIs work for humans and expecting LLM providers to optimize around that. I don't need an MCP for a CLI tool, for example, I just need a good man page or `--help` documentation.
I know in practice it no longer is the case, if it ever was.
But semantic HTML is exactly that explicit machine-readable entrypoint. I am firmly entrenched in the opinion that HTML, and the DOM is only for machines to read, it just happens to be also somewhat understandable to some humans. Take an average webpage, have a look at all characters(bytes) in there: often two third won't ever be shown to humans.
Point being: we don't need to invent something new. We just need to realize we already have it and use it correctly. Other than this requiring better understanding of web tech, it has no downsides. The low hanging fruit being the frameworks out there that should really do a better job of leveraging semantics in their output.
The only ones benefitting from 'wastefull' crawling are the anti-bot solution vendors. Everyone else is incentivized to crawl as efficiently as possible.
Makes you think, right?
I yearn for the days when a single kb get was enough. Now it's endless wastage spawning entire browsers larger than operating systems with mitigations, hacks and proxies. Requesting access directly from webmasters is only met with silence. All of my once simple, hobbyist programs are now bloated beyond belief and less reliable than ever
> It also raises interesting questions about whether sites will eventually provide different views for humans vs. automated agents in a more formalized way.
This question raises an interesting question about if this would exacerbate supply chain injection attacks. Show the innocuous page to the human, another to the bot.
Apart from the obvious problem: presenting something different to crawlers and humans.
Isn't it already covered by sitemaps and sitemap index files, which are machine readable XML?
I just do a query param to toggle to markdown/text if ?llm=true on a route. Easy pattern that's opt-in.
They already do...
A lot of known crawlers will get a crawler-optimized version of the page
Do they? AFAIK Google forbids that, and they’ll occasionally test that you aren’t doing it.
I haven't checked in a while but I know for a fact that Amazon does or did it
With google covering only 3% I wonder how much people still care and if they should. Funny: I own and know sites that are by far the best resource on the topic but shouldn't have so many links google says. It's like I ask you for a page about cuban chains then you say you don't have it because they had to many links. Or your greengrocer suddenly doesn't have apples because his supplier now offers more than 5 different kinds so he will never buy there again.
If anyone is taking feature requests, could you add an option to return the snapshot as an MHTML with all static assets embedded? (I know this could get inefficient from a storage perspective, but if it really matters you could dedupe assets on your end, which is what my janky homegrown crawler does.)
It seems like there's a missed use case: web archiving. I don't see any mention of WARC as an output format. This could be useful to journalists and academically if they had it.
And while at it, ability to mount the resulting archives at some virtual root in nginx|apache. E.g. serve site-archive.extension at /somepath/site. And standalone simple webserver that can take one or more archives from command line and servers them.
GETOLD /index.html "2026-03-11T10:30:45Z" would be such cool functionality...
Could they collaborate with the website's creators that have websites behind cloudfare to allow their content to be accessed via an API in exchange of a compensation?. This could be a way to compensate creators and AI companies be able to access content that's unreachable as it's protected by cloudfare
They are one step ahead of you: https://blog.cloudflare.com/introducing-pay-per-crawl/
Sort of though. Still private beta since July 2025.
Oh man, I was hoping I could offer a nicely-crawled version of my site. It would be cool if they offered that for site admins. Then everyone who wanted to crawl would just get a thing they could get for pure transfer cost. I suppose I could build one by submitting a crawl job against myself and then offering a `static.` subdomain on each thing that people could access. Then it's pure HTML instant-load.
I don’t really get the usecase. Is your site static? Then you should just render it to html files and host the static files. And if it’s not static, how would a snapshot of the pages help if they change later? And also why not just add some caching to the site then?
Ah the use-case is archive.org but fast. But it's okay. Before I die I will make the static copy of my site myself.
Will this crawler be run behind or infront of their bot blocker logic?
In front: https://developers.cloudflare.com/browser-rendering/rest-api...
I tried to make exactly this a year ago. Built on Cloudflare using all of their primitives: https://crawlspace.dev -- It didn't work too well (so don't bother trying it).
Worth noting: origin owners can still detect and block CF Browser Rendering requests if needed.
Workers-originated requests include a CF-Worker header identifying the workers subdomain, which distinguishes them from regular CDN proxying. You can match on this in a WAF rule or origin middleware.
The trickier issue: rendered requests originate from Cloudflare ASN 13335 with a low bot score, so if you rely on CF bot scores for content protection, requests through their own crawl product will bypass that check. The practical defense is application-layer rate limiting and behavioral analysis rather than network-level scores -- which is better practice regardless.
The structural conflict is real but similar to search engines offering webmaster tools while running the index. The incentives are misaligned, but the individual products have independent utility. The harder question is whether the combination makes it meaningfully harder to build effective bot protection on top of their platform.
LLM generated comment
They say they obey robots.txt - isn’t that the easier way?
The main win here is abstracting away browser context lifecycle management. Anyone who has run Puppeteer on Workers knows the pain of handling cold starts, context reuse, and timeout cascading across navigation steps. Having crawl() bundle render-then-extract into one call covers maybe 80% of scraping use cases. The remaining 20% where you need request interception or pre-render script injection still needs the full Browser Rendering API, but for pulling structured data from public pages this is a big simplification over managing session state yourself.
Queue-It protected pages catch it as well and prevent crawling.
Cloudflare getting all the cool toys. AWS, anyone awake over there?
This is actually really amazing. Cloudflare is just skating to where the puck is going to be on this one.
Interesting... I built an MCP server for their initial browser render as markdown, and I just tell the LLM to follow reasonable links to relative content, and recurse the tool.
Really hard to understand costs here. What is a reasonable pages per second? Should I assume with politeness that I'm basically at 1 page per second == 3600 pages/hour? Seems painfully slow.
They have a Pay Per Crawl option for owners. This plus a /crawl endpoint is genius.
If two customers crawl the same website and it uses crawl-delay, how does it handle that? Are they independent, or does each one run half as fast?
You put a governor on the domain, and you return from the cache instead.
this could be cool to use cloudflare's edge to do some monitoring of endpoints actual content for synthetic monitoring
Awesome, so I no longer have to use Firecrawl or my own crawler to scrape entire websites for an agent? Especially when needing residential proxies to do so on Cloudflare protected sites? Why though?
I have tried theirs... they are NOT proxies.. that means majority of the popular sites actually block scraping... even if they are protected by cloudflare itself.
RIP @FireCrawl or at the very least they were the inspiration for this?
Instead of "should have been an email" this is "should have been a prompt" and can be run locally instead. There are a number of ways to do this from a linux terminal.
``` write a custom crawler that will crawl every page on a site (internal links to the original domain only, scroll down to mimic a human, and save the output as a WebP screenshot, HTML, Markdown, and structured JSON. Make it designed to run locally in a terminal on a linux machine using headless Google Chrome and take advantage of multiple cores to run multiple pages simultaneously while keeping in mind that it might have to throttle if the server gets hit too fast from the same IP. ```
Might use available open source software such as python, playwright, beautifulsoup4, pillow, aiofiles, trafilatura
This presumably is going to be cheap and effective. Its much easier to wrap a prompt round this and know it works that mess around with crawling it all yourself.
You'll still be hand-rolling it if you want to disrespect crawling requirements though.
I’ve actually written a crawler like that before, and still ended up going with Firecrawl for a more recent project. There’s just so many headaches at scale: OOMs from heavy pages, proxies for sites that block cloud IPs, handling nested iframes, etc.
That'd be more like that draw an owl meme. Devil's in the details. Holy shit, there's so many details...
Cloudflare are mafiosos. They create the problem and then sell you the solution to themselves.
"Well-behaved bot - Honors robots.txt directives, including crawl-delay"
From the behaviour of our peers, this seems to be the real headline news.
TIL about the Crawl-delay directive. Although it seems that most honest bots move slower and dishonest bots will learn to.
Interested how this unfolds
"Selling the wall and the ladder."
"Biggest betrayal in tech."
"Protection racket."
These hot takes sound smart but they're not.
The web was built to be open and available to everyone. Serving static HTML from disk back in the day, nobody could hurt you because there was nothing to hurt.
We need bot protection now because everything is dynamic, straight from the database with some light caching for hot content. When Facebook decides to recrawl your one million pages in the same instant, you're very much up shit creek without a paddle. A bot that crawls the full site doesn't steal anything, but it does take down the origin server. My clients never call me upset that a bot read their blog posts. They call because the bot knocked the site offline for paying customers.
Bot protection protects availability, not secrecy.
And the real bot problem isn't even crawling. It's automated signups. Fake accounts messaging your users. Bots buying out limited drops before a human can load the page. Like-farming. Credential stuffing. That's what bot protection is actually for: preventing fraud, not preventing someone from reading your public website.
Cloudflare's `/crawl` respects robots.txt. Don't want your content crawled, opt out. But if you want it indexed and can't handle the traffic spike, this gets your content out without hammering production.
As for the folks saying Cloudflare should keep blocking all crawlers forever: AI agents already drive real browsers. They click, scroll, render JavaScript. Go look at what browser automation frameworks can do today and then explain to me how you tell a bot from a person. That distinction is already gone. The hot takes are about a version of the internet that doesn't exist anymore.
> Honors robots.txt directives, including crawl-delay
Sounds pretty useless for any serious AI company
What % of sites have a content update volume that exceeds what you can get respecting crawl delay?
If your delay is 1s and you publish less than 60 updates a minute on average I can still get 100%. Most crawls are not that latency sensitive, certainly not the ai ones.
HFT bots, now that is an entirely different ballgame.
> Most crawls are not that latency sensitive, certainly not the ai ones.
They certainly behave like they are. We constantly see crawlers trying to do cache busting, for pages that hasn't change in days, if not weeks. It's hard to tell where the bots are coming from theses days, as most have taken to just lie and say that they are Chrome.
I'd agree that the respecting robots.txt makes this a non-starter for the problematic scrapers. These are bots that that will hammer a site into the ground, they don't respect robots.txt, especially if it tells them to go away.
All of this would be much less of a problem if the authors of the scrapers actually knew how to code, understood how the Internet works and had just the slightest bit of respect for others, but they don't so now all scrapers are labeled as hostile, meaning that only the very largest companies, like Google, get special access.
> We constantly see crawlers trying to do cache busting
Do you have a source for this? Not saying you're wrong, I'd just like to know more
Not really, given that the work we do in that direction isn't exactly public. You can recreate the scenario though. Spin up a wiki of some sort, scrapers love wikis, ideally enable some form of caching, and just sit back and watch scrapers throw random shit in the URL parameters.
Honestly, it feels like cloudflare bullying other sites into using their anti-bot services. great business model by charging owners and devs at the same time. Using AI per page to parse content. its reckless.
Didn't they just throw a (very public) fit over Perplexity doing the exact same thing?
The most egregious thing Perplexity did was to straight up ignore robots.txt. Cloudflare promise not to do that, so if we take their word for it, it's a quite different setup.
That said, I'm not fan of letting users forge whatever user agents they please. Instead, AIUI to opt-out of getting crawled I have to look for the existence of certain request headers[1].
[1]: https://developers.cloudflare.com/browser-rendering/referenc...
Does this bypass their own anti-AI crawl measures?
I'll need to test it out, especially with the labyrinth.
They say it doesn't: https://developers.cloudflare.com/browser-rendering/faq/#wil...
Further down they also mention that the requests come from CFs ASN and are branded with identifying headers, so third party filters could easily block them too if they're so inclined. Seems reasonable enough.
Yeah, that'd be huge, like 90% of my search engine results are just cloudflare bot checks if I don't filter it out.
If this does bypass their own (and others') anti-AI crawl measures, it'd basically mean that the only people who can't crawl are those without money.
We're creating an internet that is becoming self-reinforcing for those who already have power and harder for anyone else. As crawling becomes difficult and expensive, only those with previously collected datasets get to play. I certainly understand individual sites wanting to limit access, but it seems unlikely that they're limiting access to the big players - and maybe even helping them since others won't be able to compete as well.
Common Crawl has free egress
I feel there is a conflict of interest here..
I'm split between: Yes! At last something to get CF protected sites! And: Uh! Now the internet is successfully centralized.
The big question here is this a verified-bot on the Cloudflare WAF? Didn't Google get into trouble for using their search engine user agent and IPs to feed Gemini in Europe?
To clarify the two questions raised:
First, the Cloudflare Crawl endpoint does not require the target site to use Cloudflare. It spins up a headless Chrome instance (via the Browser Rendering API) that fetches and renders any publicly accessible URL. You could crawl a site hosted on Hetzner or a bare VPS with the same call.
Second on pricing: Browser Rendering is only available on the Workers Paid plan ($5/month). It is not part of the free tier. Usage is billed per invocation beyond the included quota - the exact limits are in the Cloudflare docs under Browser Rendering pricing, but for archival use cases with moderate crawl rates you are very unlikely to run into meaningful costs.
The practical gotcha for forum archival is pagination and authentication-gated content. If the forum requires a login to see older posts, a headless browser session with saved cookies would help, but that is more complex to orchestrate than a single-shot fetch.
> Browser Rendering is only available on the Workers Paid plan ($5/month). It is not part of the free tier.
The post says it's available for both free and paid plans. According to the pricing page of the Browser Rendering, the free plan will have 10 minutes/day browsing time.
[0] seems to suggest even paid plans are effectively limited to 500 web pages per day, right?
[0] https://developers.cloudflare.com/browser-rendering/limits/#...Can a CDN be a "walled garden"
Fuck Cloudflare.
I've used browser rendering at work and it's quite nice. Most solutions in the crawling space are kind of scummy and designed for side-stepping robots.txt and not being a good citizen. A crawl endpoint is a very necessary addition!
>Honors robots.txt
Is it possible to ignore robot.txt in the case the crawl was triggered by a human?
All what was expected, first they do a huge campaign to out evil scrapers. We should use their service to ensure your website block LLMs and bots to come scraping them. Look how bad it is.
And once that is well setup, and they have their walled garden, then they can present their own API to scrape websites. All well done to be used by your LLM. But as you know, they are the gate keeper so that the Mafia boss decide what will be the "intermediary" fee that is proper for itself to let you do what you were doing without intermediary before.
No: https://developers.cloudflare.com/browser-rendering/rest-api...
That is funny because on this page there is a warning block with the following text:
And "Will Browser Rendering bypass Cloudflare's Bot Protection? " is a hash link to the FAQ page, that surprisingly doesn't anything available for this link entry.Is it because it was removed (/hidden) or because it is not yet available until everyone forget the "we are no evil, we are here to protect the internet"?
most websites, particularly those behind cloudflare, are very restrictive even to crawlers that obey robots. Proof: a ton of my time over the last year, and my crawlers very carefully obey robots.
It's hard to see how this isn't extorting folks by offering a working solution that, oh, cloudflare doesn't block. As long as you pay Cloudflare.
Perhaps I'm overly cynical, but I'd be quite surprised if cloudflare subjected their own headless browsing to the same rules the rest of the internet gets.
>most websites, particularly those behind cloudflare, are very restrictive even to crawlers that obey robots. Proof: a ton of my time over the last year, and my crawlers very carefully obey robots.
The docs are pretty equivocal though:
>If you use Cloudflare products that control or restrict bot traffic such as Bot Management, Web Application Firewall (WAF), or Turnstile, the same rules will apply to the Browser Rendering crawler.
It's not just robots.txt. Most (all?) restrictions that apply to outside bots apply to cloudflare's bot as well, at least that's what they're claiming. If they're being this explicit about it, I'm willing to give them the benefit of the doubt until there's evidence to the contrary, rather than being a cynic and assuming the worst.
Do I have the option to fill it with junk for LLMs?
This might be really great!
I had the idea after buying https://mirror.forum recently (which I talked in discord and archiveteam irc servers) that I wanted to preserve/mirror forums (especially tech) related [Think TinyCoreLinux] since Archive.org is really really great but I would prefer some other efforts as well within this space.
I didn't want to scrape/crawl it myself because I felt like it would feel like yet another scraping effort for AI and strain resources of developers.
And even when you want to crawl, the issue is that you can't crawl cloudflare and sometimes for good measure.
So in my understanding, can I use Cloudflare Crawl to essentially crawl the whole website of a forum and does this only work for forums which use cloudflare ?
Also what is the pricing of this? Is it just a standard cloudflare worker so would I get free 100k requests and 1 Million per the few cents (IIRC) offer for crawling. Considering that Cloudflare is very scalable, It might even make sense more than buying a group of cheap VPS's
Also another point but I was previously thinking that the best way was probably if maintainers of these forums could give me a backup archive of the forum in a periodic manner as my heart believes it to be most cleanest way and discussing it on Linux discord servers and archivers within that community and in general, I couldn't find anyone who maintains such tech forums who can subscribe to the idea of sharing the forum's public data as a quick backup for preservation purposes. So if anyone knows or maintains any forums myself. Feel free to message here in this thread about that too.
"I didn't want to scrape/crawl it myself because I felt like it would feel like yet another scraping effort for AI and strain resources of developers"
You feel better paying someone to do the same thimg?
I actually don't but it seems that cloudflare caches responses so if anything instead of straining the developer resources, it would strain more cloudflare resources and cloudflare could better handle that more efficiently with their own crawl product.
Also, I am genuinely open to feedback (Like a lot) so just let me know if you know of any other alternative too for the particular thing that I wish to create and I would love to have a discussion about that too! I genuinely wish that there can be other ways and part of the reason why I wrote that comment was wishing that someone who manages forums or knows people who do can comment back and we can have a discussion/something-meaningful!
I am also happy with you also suggesting me any good use cases of the domain in general if there can be made anything useful with it. In fact, I am happy with transferring this domain to you if this is something which is useful to ya or anyone here (Just donate some money preferably 50-100$ to any great charity in date after this comment is made and mail me details and I am absolutely willing to transfer the domain, or if you work in any charity currently and if it could help the charity in any meaningful manner!)
I had actually asked archive team if I could donate the domain to them if it would help archive.org in any meaningful way and they essentially politely declined.
I just bought this domain because someone on HN said mirror.org when they wanted to show someone else mirror and saw the price of the .org domain being so high (150k$ or similar)and I have habit of finding random nice TLD and I found mirror.forum so I bought it
And I was just thinking of hmm what can be a decent idea now that I have bought it and had thought of that. Obviously I have my flaws (many actually) but I genuinely don't wish any harm to anybody especially those people who are passionate about running independent forums in this centralized-web. I'd rather have this domain be expired if its activation meant harm to anybody.
looking forward to discussion with ya.
This is used to scrape third-party sites not necessarily behind cloudflare so it has nothing to do with whether cloudflare caches it or not plus when using their browser rendering it doesn't even fetch cached responses anyways....
I didn't know that it doesn't fetch catched responses, my apologies. I had only read through it with a glance and it felt like something that cloudflare might've done. Is there any particular reason that they don't use the cached responses, feels like a missed opportunity but maybe I am missing something?
It's a browser rendering API which means people are paying a premium specifically to have a browser render a live website. If you want to get a cached response of a page and still possibly get blocked by cloudflare you could just make a node script with a simple fetch and save your money.
Selling the cure (DDoS protection) and creating the poison (Authorized AI crawling) against their customers.
I love this from CloudFlare!
Off-topic, but I'm having a terrible experience with Cloudflare and would love to know if someone could offer some help.
All of a sudden, about 1/3 of all traffic to our website is being routed via EWR (New York) - me included -, even tough all our users and our origin servers are in Brazil.
We pay for the Pro plan but support has been of no help: after 20 days of 'debugging' and asking for MTRs and traceroutes, they told us to contact Claro (which is the same as telling me to contact Verizon) because 'it's their fault'.
Do you think cloudflare is responsible for all of the network traffic routing in the entire world and can simply fix any problem even if it's on somebody else's network?
No. I do think that Cloudflare is a great company and got where it's at today because they care for this type of issue, and has a much better chance of contacting their peering traffic partner than me because they take care of ~20% of all internet traffic, while I take care of none.
It is possible that Claro has a bad route that sends all traffic destined for Cloudflare through New York.
Every once and a while we have had Bell Canada route a request that should be going about 6 blocks away across the continent and back.
They are not super helpful fixing it either.