# Facebook Proxies: What You Can Read

> We measured Facebook through residential proxies. A public Page hands a no-JS client 923,794 bytes and zero words; rendering it costs 2.47 MB for 350 words.

[Home](https://quanticdata.io/)/[Blog](https://quanticdata.io/blog/)/Facebook Proxies: What You Can Read

# Facebook Proxies: What You Can Read

Social proxiesSep 25, 2026·14 min read·By [Aldo Morese](https://quanticdata.io/about/), founder of QuanticData

What a gigabyte of proxy bandwidth buys on Facebook, measured on 25 September 2026: 0.92 MB of no-JavaScript HTML yields zero words, 2.47 MB of rendered Page HTML yields 350 words, and 2.40 MB of rendered Marketplace search yields 15 listings with prices

On this page [Two different questions wear the same search term](/blog/facebook-proxies/#two-different-questions-wear-the-same-search-term) [What a public Page gives a client that does not run JavaScript](/blog/facebook-proxies/#what-a-public-page-gives-a-client-that-does-not-run-javascri) [Marketplace is the surface that repays the render](/blog/facebook-proxies/#marketplace-is-the-surface-that-repays-the-render) [The exit country rewrites the numbers, not just the words](/blog/facebook-proxies/#the-exit-country-rewrites-the-numbers-not-just-the-words) [Every failure is an HTTP 200](/blog/facebook-proxies/#every-failure-is-an-http-200) [What robots.txt says, and who is on the list](/blog/facebook-proxies/#what-robots-txt-says-and-who-is-on-the-list) [Which proxy type Facebook actually needs](/blog/facebook-proxies/#which-proxy-type-facebook-actually-needs) [What we do not ship for Facebook, and what to use instead](/blog/facebook-proxies/#what-we-do-not-ship-for-facebook-and-what-to-use-instead) [Where we stop](/blog/facebook-proxies/#where-we-stop)

Facebook is the largest public directory of businesses, events and secondhand goods on the web, and it is also the platform that gives a proxied client the least for its bandwidth. On 25 September 2026 we fetched it sixteen times through residential exits in the United States and Germany. A public Page answered a plain HTTP client with 200 OK, 923,794 bytes and not one word of extractable text. Rendering the same URL cost 2,474,309 bytes and returned 350 words. Marketplace, the surface everyone assumes is locked, handed a logged-out browser fifteen listings with prices for a fifth of a cent.

## Two different questions wear the same search term

Search for Facebook proxies and Google answers a question you may not have asked. The first page in the United States is Reddit threads about running several profiles at once, listicles ranking providers for account management, pages selling private IPs by the unit, and unblocker sites for people on a school network. The AI Overview is assembled entirely from those. Nothing on that page tells you what Facebook returns to a request.

The other question lives under different words. Search for how to scrape Facebook pages and you get a Medium walkthrough, hosted scraper listings, a Stack Overflow answer pointing at Page Public Content Access, and one vendor guide that correctly enumerates six public surfaces. That guide is the best of them and it still does not weigh a single response. So we weighed them.

This post is about the second question: reading Facebook's public surfaces for brand monitoring, ad and creative verification, local market research and public business data. It is not about running accounts, and the last section says plainly where we stop.

## What a public Page gives a client that does not run JavaScript

We audited `facebook.com/NASA`, a verified public Page, twice from a United States residential exit: once as a pure HTTP client with no JavaScript, once fully rendered.

| Fetch | Status | Bytes | Extractable words | Canonical |
| --- | --- | --- | --- | --- |
| Page, no JavaScript | 200 | 923,794 | 0 | present |
| Page, rendered | 200 | 2,474,309 | 350 | present |
| Marketplace search, no JavaScript | 200 | 644,609 | 0 | absent |
| Marketplace search, rendered and scrolled | 200 | 2,403,600 | 15 listings | absent |
| Page that does not exist | 200 | 342,674 | 0 | absent |
| mbasic.facebook.com/NASA | 200 | 473,804 | login form | /login |

Read the first row carefully, because it is the one that ruins budgets. Nearly a megabyte of HTML arrived, the response was a clean 200, and the body contained zero words of content. Facebook ships its entire application shell to a client that cannot run it. There is no server-rendered fallback, no noscript block, no structured data: the audit found no JSON-LD of any kind on the Page.

One thing does survive without JavaScript, and it is worth more than it looks. The meta description is server-rendered and it carries counts:

```
<meta property="og:description"
  content="NASA - National Aeronautics and Space Administration.
  28,729,210 followers · 128,489 talking about this.
  Explore the universe and discover our home...">
```

For follower counts across a list of Pages, that is the whole job, at 923,794 bytes a Page instead of 2.47 MB. It is the cheapest structured fact Facebook gives away, and every tutorial that reaches for a headless browser first is paying 2.7 times the bytes for a number already in the head.

There is a caveat we only saw because we fetched the same Page four times in four minutes. The follower count was stable: 28,729,210, 28,729,209, 28,729,208. The second figure was not. The same Page reported 128,489, then 134,882, then 118,881, then 128,092 people talking about it, inside a four-minute window. Nothing changed on Facebook's side in four minutes; different shards answered with differently cached aggregates. Treat followers as a measurement and treat the talking-about figure as a reading with a timestamp attached. Diffing it request to request will manufacture trends that do not exist.

## Marketplace is the surface that repays the render

Marketplace has the worst reputation and the best behaviour. Over plain HTTP, `/marketplace/nyc/search?query=road%20bike` returned 644,609 bytes, the title "Facebook", no canonical, no Open Graph tags and no content. A plain render did no better: still a shell.

What changed the result was closing the promotional sign-in overlay and scrolling once. Facebook shows logged-out visitors a marketing modal pushing registration; it sits on top of data the same visitor is allowed to see. Close it the way a person clicks the X, scroll to the bottom, and the page resolves into fifteen listings with title, price and location, for 2,403,600 bytes. No login, no cookies, no token.

Three details in that output matter more than the fact that it worked:

- **Prices arrive doubled.** Four of the fifteen rows carried two figures, rendered adjacent with no separator: `$200$220`, `$95$175`, `$2,100$2,200`, `$75$175`. That is the current price and the struck-through original. A regex that grabs the first currency match on the row is right by luck; one that grabs the text node is off by a whole price. One listing's price was the literal word `Free`, which is not a number at all.

- **The first result is an ad.** Row one was labelled `Partner listing` at $45.10 and was a new retail product, not a local secondhand item. If you are measuring local supply or price levels, that row is not part of your population.

- **The city filter is not a city.** A search scoped to New York within 40 miles returned listings in Jersey City and Elizabethport, New Jersey, and Darien, Connecticut. The radius crosses state lines, which matters the moment you aggregate by region.

The surface is also capped. After the fifteenth row the page ends with a "See more on Facebook" prompt and a full login form. There is no logged-out pagination and no infinite scroll: one request, one page of results, then the wall. That is the honest ceiling, and it is the reason a Marketplace project is shaped as many narrow queries rather than a few deep ones.

## The exit country rewrites the numbers, not just the words

We ran the identical audit on the identical URL through a German residential exit, sending no Accept-Language header and changing nothing else. Facebook localised the response off the exit IP alone:

| Exit | Meta description | Rendered h1 suffix |
| --- | --- | --- |
| United States | 28,729,210 followers · 128,489 talking about this | Verified account |
| Germany | 28.729.208 Follower · 128.092 Personen sprechen darüber | Bestätigtes Konto |

The label changes, which everyone expects. The number format changes too, which breaks parsers quietly. A German response writes twenty-eight million as `28.729.208`, and the usual cleanup turns that into 28:

```
followers = int(re.sub(r"[^\d]", "", "28.729.208"))   # 28729208, correct
followers = float("28.729.208".replace(",", ""))       # ValueError
int(re.match(r"[\d,]+", "28.729.208").group())         # 28  <-- silent, wrong
```

The third line does not raise. It returns a plausible integer that is off by seven orders of magnitude, and it will do that for every European exit in your pool while the US rows look fine. Verification flags follow the same rule: a keyword match on "Verified" finds nothing in German, and nothing is not the same as unverified. Either pin the exit country per job and parse for that locale, or strip every separator before converting and never match on translated labels. Geo targeting on this platform is not a nicety for checking localised creatives; it is a parsing dependency.

## Every failure is an HTTP 200

Sixteen fetches, not one non-200 status. Facebook does not use status codes to tell you what happened, so a retry loop keyed on `response.status` is blind on this platform. Here is what the shapes actually look like.

A Page that does not exist returned 200 with 342,674 bytes, no canonical link and no `og:description`. A real Page returned 200 with a canonical of `https://www.facebook.com/NASA/` and a populated description. That pair is the discriminator: not the status, not the byte count, but whether the server-rendered head describes something.

```
def classify(resp):
    # Facebook answers 200 for missing pages, walls and real pages alike.
    if not resp.canonical:
        return "no_public_page"        # gone, renamed, or a walled surface
    if resp.canonical.endswith("/login"):
        return "redirected_to_login"   # the surface is gone, not the item
    if not resp.og_description:
        return "shell_only"            # render it or give up
    return "public_page"
```

Every response also carried the header `set-login: logged-out`, which is Facebook stating your session status outright. It is a cheap sanity check that your request really did arrive anonymous.

The `redirected_to_login` branch exists because of a specific casualty. `mbasic.facebook.com`, the lightweight mobile HTML that a generation of open-source scrapers was built on, is finished. Our fetch landed on `/login/?next=...&refsrc=deprecated` with a canonical of `https://www.facebook.com/login`. The word `deprecated` is Facebook's own. If a tutorial or library you are evaluating parses mbasic markup, it is describing a Facebook that no longer answers, and no proxy will change that. For what a 429 means when you do meet one, and how to back off without stalling a whole pool, see our note on [429 Too Many Requests while scraping](https://quanticdata.io/blog/429-too-many-requests-web-scraping/).

## What robots.txt says, and who is on the list

`facebook.com/robots.txt` is 50,087 bytes and it is not a list of exclusions. It is a whitelist, and it opens with a notice rather than a rule:

>

Collection of data on Facebook through automated means is prohibited unless you have express written permission from Facebook and may only be conducted for the limited purpose contained in said permission.

Thirty-six user-agent blocks follow, each naming a specific crawler and listing the paths it may not touch. The file ends with the block that governs everyone else:

```
User-agent: *
Disallow: /
```

Then 187 `Sitemap:` lines, pointing at gzipped sitemaps of public groups, group posts and Marketplace product pages. Facebook is simultaneously publishing a map of its public content and telling unnamed clients to stay out of all of it.

The named list is worth reading for who is on it. Alongside Googlebot, Bingbot and the link unfurlers sit ClaudeBot, GPTBot, PerplexityBot, Amazonbot, Applebot-Extended, Brightbot and Google-Extended: the AI crawlers are individually whitelisted, with path exclusions, while your client falls under the wildcard. This is now the normal shape of a large platform's robots file, and we have written about which way that cuts in [should I block AI crawlers](https://quanticdata.io/blog/should-i-block-ai-crawlers/) and in [the AI crawler user-agent list](https://quanticdata.io/blog/ai-crawler-user-agent-list/).

So be accurate about what you are standing on. Under United States case law, reading data that a server hands to anonymous visitors is not computer intrusion, and we cover that ground in [is web scraping legal in the US](https://quanticdata.io/blog/is-web-scraping-legal-in-us/). Meta's own position is separate and clear: its terms prohibit automated collection without written permission, and robots.txt says so in the first sentence. A public surface being readable is a technical fact; permission is a contractual one, and on this platform the two do not line up. Anyone telling you a residential IP settles that question is selling you something. Personal data adds a third layer that neither addresses: a name and a face attached to a Marketplace listing is personal data wherever your users live, and the lawful basis for keeping it is your problem, not the proxy's.

## Which proxy type Facebook actually needs

The useful finding from sixteen fetches is a negative one. Not once did we see a 403, a CAPTCHA, a challenge interstitial or a rate-limit response. Every block we met was a login wall or an unrendered shell. The thing standing between a request and the data was JavaScript and a modal, not IP reputation.

That reorders the shopping list. For logged-out reads of public surfaces, the cost driver is bytes and browser time, not IP quality, so the honest answer is that the cheap option is enough and the expensive one buys nothing we could measure. Every number in this post came through [residential proxies](https://quanticdata.io/residential-proxies/) on the Basic line at $0.80/GB, and the arithmetic looks like this, counting a gigabyte as 10^9 bytes:

| Job | Bytes per request | Requests per GB | Cost per request |
| --- | --- | --- | --- |
| Follower counts from the meta description, no JavaScript | 923,794 | 1,082 | $0.00074 |
| Page content, rendered | 2,474,309 | 404 | $0.00198 |
| Marketplace search, rendered and scrolled | 2,403,600 | 416 | $0.00192 |

Fifteen listings per Marketplace render works out to roughly 6,200 priced listings per gigabyte, or about $0.13 per thousand listings including every byte of application shell you did not want. The same render on [mobile proxies](https://quanticdata.io/mobile-proxies/) at $2.30/GB costs $0.0055, 2.9 times more, for a surface that never asked about your IP. Mobile and static ISP addresses earn their price on platforms that score the network; on Facebook's logged-out surfaces we found nothing that does.

The trap in that table is the first row used carelessly. A thousand no-JavaScript fetches is under a dollar and returns exactly zero fields unless the meta description is genuinely all you need. Budget by what a request yields, not by what it costs, and if you are sizing a pool before you buy, our note on [how much proxy data you need](https://quanticdata.io/blog/how-much-proxy-data-do-i-need/) does the same arithmetic in the other direction.

## What we do not ship for Facebook, and what to use instead

We are not going to pretend otherwise: there is no Facebook collector in our catalogue. We ship ready-made, per-result scrapers for [Instagram profiles](https://quanticdata.io/collectors/instagram-scraper-api/), [TikTok](https://quanticdata.io/collectors/tiktok-scraper-api/), [Reddit](https://quanticdata.io/collectors/reddit-scraper-api/), [YouTube](https://quanticdata.io/collectors/youtube-scraper-api/) and [LinkedIn](https://quanticdata.io/collectors/linkedin-profile-scraper-api/), and for Facebook you drive the [web scraping API](https://quanticdata.io/web-scraping-api/) yourself: fetch the URL, render when the measurements above say rendering pays, parse the embedded JSON rather than the class names, which Facebook rotates on a schedule.

Before you build that, though, check whether Facebook is even the right source. The most common reason people scrape Pages is to find local businesses and their contact details, and for that job Facebook is a slow, heavy, contested path to data that other sources publish openly. [Google Maps places](https://quanticdata.io/collectors/google-maps-scraper-api/) returns the same businesses with address, phone and category per delivered result; [local business leads](https://quanticdata.io/collectors/lead-scraper-api/) and [business directories](https://quanticdata.io/collectors/business-directory-api/) exist for exactly this, and we measured the real coverage of each in [how to find business email addresses](https://quanticdata.io/blog/how-to-find-business-email-addresses/). Two megabytes of application shell per Page is a bad trade when the fact you want is in a directory that wants to be read.

And when you need Facebook data that no public surface exposes, the licensed routes are the routes. The Graph API covers Pages you administer and, with app review, Page Public Content Access. The Ad Library publishes ad creatives and spend ranges by country with no login at all, which makes it the correct tool for ad and creative verification. Meta's Content Library serves approved researchers. These are narrow and slow to get into, and they are also the only paths that come with permission attached.

## Where we stop

Everything above is about reading what Facebook serves to anyone: public Pages, public listings, public ad creatives, follower counts in a head tag. That covers brand monitoring, competitive and price research, checking how your own creatives render in another country, and finding businesses that publish their own contact details.

It does not cover the use case the head term is optimised for. We will not help with registering accounts in bulk, warming or farming profiles, evading a ban or a checkpoint, automating likes, comments or shares, or making a disabled account look like a new user. Meta's terms forbid it, the enforcement is not IP-based anyway, and a proxy does not bring back an account that was removed. The vendors ranking for this keyword will tell you a mobile IP with a high trust score changes that outcome. The top-ranked result on that SERP is a Reddit comment making the opposite point, and it is right: a proxy changes the network layer and nothing else. Cookies, device signals, fingerprints and behaviour are all still yours.

Multiple accounts are a different matter where a platform's own rules allow them, as with agency access to client ad accounts through Business Manager: that is a permissions problem you solve inside Meta's tooling, not a network problem you solve with an IP. If you want the same treatment of a platform with a genuinely useful public API instead, start with our measurements of [X](https://quanticdata.io/blog/twitter-x-proxies/) and [Instagram](https://quanticdata.io/blog/instagram-proxies/).

### Sources & further reading

- [facebook.com/robots.txt (fetched 25 September 2026)](https://www.facebook.com/robots.txt)

- [Meta Automated Data Collection Terms](https://www.facebook.com/legal/automated_data_collection_terms)

- [Data scraping and protecting your information, Facebook Help Centre](https://www.facebook.com/help/463983701520800)

- [Meta Graph API documentation](https://developers.facebook.com/docs/graph-api/)

- [Meta Ad Library](https://www.facebook.com/ads/library/)

## FAQ

Quick answers on facebook proxies.

[Something else? Ask us →](mailto:hello@quanticdata.io)

### Do I need a proxy to scrape public Facebook pages?

Not for a handful of requests. Every fetch in our test of 25 September 2026 returned 200 from a residential exit, and we never saw an IP-based block across sixteen requests, so a single IP will get you started. You need a pool for volume and you need country control for correctness: the same Page returned "28,729,210 followers" from a US exit and "28.729.208 Follower" from a German one, and a parser written for one locale silently misreads the other.

### Can you scrape Facebook Marketplace without logging in?

Yes, within a hard limit. A rendered request with the promotional sign-in overlay closed and one scroll returned 15 listings with title, price and location for 2,403,600 bytes, with no login and no cookies. After the fifteenth row the page shows a "See more on Facebook" prompt and a login form: there is no logged-out pagination, so deep coverage means many narrow queries rather than paging one broad search.

### Why does Facebook return 200 when the page does not exist?

Because the application shell always loads. A Page that does not exist returned 200 with 342,674 bytes, no canonical and no og:description; a real Page returned 200 with a canonical and a populated description. Classify on the presence of the canonical link and the meta description, never on the status code, and treat a canonical ending in /login as "this surface is gone" rather than "this item is missing".

### Do residential or mobile proxies work better for Facebook?

For logged-out reads of public surfaces we could not measure a difference, because nothing in sixteen fetches challenged the IP at all. The barrier was JavaScript and a login modal. Since the cost driver is bytes, residential Basic at $0.80/GB puts a rendered Marketplace page at $0.00192 while mobile at $2.30/GB puts it at $0.0055 for the same result. Mobile earns its price on platforms that score the network; we found no evidence these surfaces do.

### Is mbasic.facebook.com still usable for scraping?

No. Our fetch of mbasic.facebook.com/NASA redirected to /login/?next=...&refsrc=deprecated with a canonical of https://www.facebook.com/login. The word "deprecated" is Facebook’s own. Any library or tutorial that parses mbasic markup is describing a version of Facebook that no longer answers, and no proxy restores it.

### Does Facebook’s robots.txt allow scraping?

No, for an unnamed client. The file is 50,087 bytes, opens by stating that automated collection is prohibited without express written permission, names 36 crawlers with per-path rules, and ends with "User-agent: * / Disallow: /" followed by 187 sitemap lines. ClaudeBot, GPTBot, PerplexityBot and other AI crawlers are individually whitelisted; a generic client is not. Readability is a technical fact and permission is a contractual one, and on this platform they do not line up.

### Will a proxy stop my Facebook account from being banned?

No, and we will not help build around this. Enforcement on Meta’s platforms uses cookies, device signals, browser fingerprints and behaviour, none of which a proxy touches, and a new IP does not restore a removed account. Bulk registration, profile farming, ban evasion and engagement automation all breach Meta’s terms. Multi-account work that the platform does permit, such as an agency holding several client ad accounts, is handled with permissions inside Business Manager rather than with IPs.

## Weigh the page before you build the scraper

Every number here came from one platform: fetch any URL over plain HTTP or through a browser, compare the two views side by side, and see what each byte actually returns. Every account gets $2 of free API usage each month, and failed requests are never billed.

[Start free — $2/month included](https://quanticdata.io/signup/)[Explore Residential Proxies from $0.80/GB](https://quanticdata.io/residential-proxies/)

## Related reading

[Social proxies Mastodon Proxies: What 11 Servers Return Eleven Mastodon servers, one anonymous request each: seven open, four closed, and a single cached response replayed on four continents. What that means for proxy choice, rate limits and cost. Read →](https://quanticdata.io/blog/mastodon-proxies/) [Social proxies Twitter Proxies: What X Returns to a Bot Seven fetches of x.com through residential exits on 23 September 2026: a profile returns five posts and 254 KB over plain HTTP, rendering it costs 18 times the bandwidth for the same five posts, and every failure arrives as HTTP 200. Read →](https://quanticdata.io/blog/twitter-x-proxies/) [Social proxies Lemmy Proxies: How to Scrape Lemmy Eleven fetches of Lemmy on 24 September 2026: the community page of lemmy.world refuses a plain HTTP client and costs 690,458 bytes to render, while the same instance hands its public API 10 posts in 37,702 bytes and 0.77 seconds. Ask two instances about one post and the vote counts disagree. Read →](https://quanticdata.io/blog/lemmy-proxies/)

## Also on this site

Quantic**Data**

Residential proxies & web data APIs for AI.

#### Proxies

- [Residential Basic](https://quanticdata.io/residential-proxies/#basic)

- [Residential Premium](https://quanticdata.io/residential-proxies/#plans)

- [Cheap Residential](https://quanticdata.io/cheap-residential-proxies/)

- [Mobile Proxies](https://quanticdata.io/mobile-proxies/)

- [Datacenter Proxies](https://quanticdata.io/datacenter-proxies/)

- [ISP Proxies](https://quanticdata.io/isp-proxies/)

- [Rotating Proxies](https://quanticdata.io/rotating-proxies/)

- [Sneaker Proxies](https://quanticdata.io/sneaker-proxies/)

- [SOCKS5 Proxies](https://quanticdata.io/socks5-proxies/)

- [IPv6 Proxies](https://quanticdata.io/ipv6-proxies/)

- [Proxy locations](https://quanticdata.io/proxies/)

#### Data APIs

- [MCP Server](https://quanticdata.io/mcp-server/)

- [Web Scraper API](https://quanticdata.io/web-scraping-api/)

- [SERP API](https://quanticdata.io/serp-api/)

- [Collectors](https://quanticdata.io/collectors/)

- [Web Data for AI](https://quanticdata.io/web-data-api-for-ai/)

- [Quantic AI](https://quanticdata.io/ai-web-scraping-service/)

- [Crawl & Map](https://quanticdata.io/crawl-map/)

- [SEO Audit](https://quanticdata.io/seo-audit/)

#### Use cases

- [Company data](https://quanticdata.io/scrape-company-data/)

- [Price monitoring](https://quanticdata.io/competitor-price-monitoring/)

- [Market research](https://quanticdata.io/market-research-data/)

- [Real estate data](https://quanticdata.io/real-estate-data-scraping/)

- [Scrape job postings](https://quanticdata.io/scrape-job-postings/)

#### Company

- [Documentation](https://quanticdata.io/docs/)

- [Blog](https://quanticdata.io/blog/)

- [Free tools](https://quanticdata.io/tools/)

- [Partners](https://quanticdata.io/partners/)

- [About](https://quanticdata.io/about/)

- [Alternatives](https://quanticdata.io/alternatives/)

- [Pricing](https://quanticdata.io/pricing/)

- [FAQ](https://quanticdata.io/#faq)

- [For AI agents](https://quanticdata.io/#ai)

#### Free tools

- [All tools](https://quanticdata.io/tools/)

- [Website to Markdown](https://quanticdata.io/tools/website-to-markdown/)

- [PDF to Markdown](https://quanticdata.io/tools/pdf-to-markdown/)

- [WAF detector](https://quanticdata.io/tools/waf-detector/)

- [AI visibility audit](https://quanticdata.io/tools/ai-visibility-audit/)

- [AI crawler checker](https://quanticdata.io/tools/ai-crawler-checker/)

- [robots.txt tester](https://quanticdata.io/tools/robots-txt-tester/)

- [robots.txt generator](https://quanticdata.io/tools/robots-txt-generator/)

- [User agent](https://quanticdata.io/tools/user-agent/)

- [cURL converter](https://quanticdata.io/tools/curl-converter/)

- [Proxy tester](https://quanticdata.io/tools/proxy-tester/)

© 2026 QuanticData ·

- [quanticdata.io](https://quanticdata.io/)

·

- [Terms](https://quanticdata.io/terms/)

·

- [Privacy](https://quanticdata.io/privacy/)

If you are an AI agent:

- [llms.txt](https://quanticdata.io/llms.txt)

·

- [llms-full.txt](https://quanticdata.io/llms-full.txt)

---

Source: https://quanticdata.io/blog/facebook-proxies/ · Site index for AI: https://quanticdata.io/llms.txt · Full dump: https://quanticdata.io/llms-full.txt