blog/markdown/Create-a-Mastodon-bot-to-forward-Twitter-and-RSS-feeds-to-your-timeline.md

33 KiB

This article is divided in three sections:

  1. A first section where I share some of my thoughts on the Fediverse, on the trade-offs between centralized and decentralized social networks, and go over a brief history of the protocols behind platforms like Mastodon.

  2. A second section where I show with a practical example that leverages Platypush how to set up a bot that brings your favorite Twitter profiles and RSS feeds to your Fediverse timeline, even if they don't have an account there.

  3. Some final observations on the current drawbacks of the Fediverse, with a particular focus on Mastodon and the current state of relaying.

If you are just here for the code, feel free to skip to the Creating a cross-posting bot section and skip the last section. Otherwise, grab a coffee while I go over some techno/philosophical analysis of social media in 2022, how we got here and what the future may hold.

Searching for a social safe harbor

My interest into the Fediverse and its ideas, protocols and products dates back to more than a decade.

I've had an account on the centralized Diaspora instance more or less since the service was spawned in 2010 until it shut down, even though I haven't updated it for the last couple of years.

And I've been running a Mastodon instance mainly dedicated to Platypush for a while. However, I haven't advertised it much so far, since I haven't been spending much time on it myself until recently.

My interest in the Fediverse used to be quite sporadic until recently. Yes, I would rant a lot about Facebook/Meta, about the irresponsibility and greediness rooted deep in its culture, their very hostile and opaque approach against external researchers and auditors and the deeply flawed thirst for further centralization that motivates each of its decisions. And, whenever I got too sick of Facebook, I would just move my social tents to Twitter for a while. Which is far from perfect, but it probably used to be the least poisonous between the two necessary evils. As somebody how had been on alternative social networks for more than a decade, I know way too well the feeling of excitement when a new shiny toy comes in town, quickly followed by the rolling tumbleweeds.

That applies until recently.

I don't feel comfortable anymore sharing my thoughts and communications on a platform owned by the richest man on earth, which also so happens to be a chief troll with distorted ideas about the balance between freedom of speech and responsibilities for one's words.

So, just like many other users did after Musk's takeover, I also rushed (back) to the Fediverse as a safe and uncompromising solution. But, unlike the majority of them, instead of rushing to mastodon.online (I don't like the idea of moving from a centralized platform/instance to another), I rushed to upgrade and prepare my dusty social.platypush.tech instance.

Give me back the old web

The whole idea of a Fediverse is as old as Facebook and Twitter themselves.

identi.ca, launched in 2008, was probably the first usable implementation of an open-source social network based on Activity Streams, an open syndacation format drafted by the W3C to represent entities, accounts, media, posts and more across several social platforms. Considering the time when it was born, it was a lot influenced by the ideas of the semantic web that were popular at the time (it's about that pre-crypto Web 3.0 that didn't happen, at least not in this universe's timeline).

GNU Social followed in 2009 (and it's still active today), then Diaspora in 2010 brought the world of alternative open-source social networks into the spotlight for a while.

A lot of progress has happened since then. ActivityPub, another open protocol drafted by the W3C, has become a de-facto standard when it comes to sharing content across different instances and platforms. And tens of platforms (including Mastodon itself, Pleroma, PeerTube, Pubcast, Hubzilla, NextCloud Social, Friendica) currently support ActivityPub, making it possible for users to follow, interact and share content regardless of where it is hosted.

Anybody can install and run a public instance using one of these platforms, and anybody on that instance can follow and interact with other users, even if they are on other platforms, as long as those instances are publicly searchable. This is possible because the underlying protocols are the same, no matter who runs the server or what application the server runs. If I have an account on a Mastodon instance, I can use it to follow a video channel on a PeerTube instance and comment on it. Even if they run on different machines and they run different applications, the platforms are able to share content and ensure federated authentication with one another, just like your web browser can be used to render content from different web servers: as long as they speak the same protocol (in this case, HTTP), a browser can render any content, regardless if it comes from an Apache or a Tomcat server.

This is the way social networks should have been implemented from the very beginning. Anybody can run one, it's up to admins of instances to decide which other instances they want to federate with (therefore importing traffic from other instances into a unique federated timeline), and it's up to individual users to decide who they want to follow and therefore be part of their home timeline, regardless of who runs the servers where those accounts are hosted.

It's an idea that sits somewhere between email (you can exchange emails with anyone as long as you have their email address, even if you have a @gmail.com account and they have a @hotmail.com account, even if you use Thunderbird as a client and they use a web app) and RSS feeds (you can aggregate links from any source under the same interface, as long as that source provides an RSS/Atom feed).

And that's indeed the trajectory that social networks were projected to follow until the early 2010s. The W3C and ISO had worked feverishly on open protocols that could make the social network experience open and distributed, like the whole Internet had been designed to run up to that date. And implementations such as identi.ca, GNU Social and Diaspora were quickly popping up to showcase those implementations.

But that's not how history went in this universe, as we all know.

Facebook underwent an exponential growth through aggressive centralization and controversial data collection practices and monetization practices. Most of the other social networks also followed the Facebook model.

Open chat protocols like XMPP were gradually replaced by centralized apps with nearly no integrations with the outside world.

Open syndacation protocols like RSS and Atom were replaced by closed timelines curated by centralized and closely guarded algorithms. This was in part also due to Google killing Reader, the most used interface for feeds, because it was in the way of their idea of web content monetization: without a major player like Google who had interest in the development of those open protocols, innovation on RSS/Atom largely stalled.

Open activity pub/sub algorithms were replaced by a handful of walled gardens, whose concept of "data portability" often involved manually downloading a heavy, unsorted and often unusable zip dump of all of your data.

Transparent, machine-readable data access was replaced by proprietary user interfaces, and a few half-heartedly implemented APIs that cover only part of the features, and can be deprecated with nearly no notice depending on whatever objective a private company decides to pursue on the short term.

I would argue that the aggressive push towards centralization, closed protocols and walled gardens of the 2010s has only benefited a handful of private companies, while throwing a wrench in a machinery that was already working well, replacing it with a vision of the Web that created way more problems that the ones that it aimed to solve. All in all, the 5-6 companies behind that disaster named Web 2.0 are responsible for pushing the Web back by at least a decade.

The wave however, as it always happens in that eternal swing between centralization and decentralization that propels our industry, is changing. The drawbacks of the centralized social network model have been under everyone's for the past few years. The "you can check out any time you like, but you can never leave, because all of your friends and relatives are here" blackmail strategy starts to be less effective, because alternatives are popping up, they are starting to gain traction, and the bleeding of active users on Facebook and Twitter has been a fact for at least the past two years.

Facebook is aware of it, but some reason they believe that the solution to the problems of centralized social networks is a creepy clone of SecondLife that they call Metaverse. Twitter is much more aware of the issue, and they have in fact decided to speed up things with their Bluesky project.

They have recently published a Github repo with a simple MVP consisting of a server, an in-memory database and a command-line interface, and a (still quite vague) architecture document that resembles a lot the ActivityPub implementation, except with a more centralized and hierarchical control chain with a (still vaguely defined) consortium/committee sitting at its top, and a Blockchain-like append-only ledger to manage information.

I see Twitter's announcement as a reflex reaction to the bleeding of users towards decentralized platforms that happened shortly after Musk's takeover. It almost feels as if an engineer was rushed to push some MVP on their laptop to show that they have a carrot they can give to their users. But it's too little, too late.

There are nearly two decades of work behind ActivityPub. A lot of smart people have already figured out the (open) solutions to most of the problems. I don't see the value of reinventing the wheel through a solution owned by a private company, with a private consortium behind it, that proposes a solution that is largely incompatible with what the W3C has been working on since the mid 2000s.

And I don't trust the sincerity of Twitter and the BlueSky investors. If Twitter was that interested in building a decentralized social network, then where have they been for the past 15 years, and why haven't they contributed more to open protocols like ActivityPub? What's the need of yet another closed-access committee to design the future of social media when we already have the W3C?

It sounds like they have preferred instead to milk their centralized, closed-source and closed-protocol cow as long as they could (even when it was clear that it wasn't profitable). They have built some hype around BlueSky for the past two years that was all marketing talk and no architecture document (let alone a usable codebase), and they have rushed to push a half-baked MVP after the richest man on earth bought them and thousands of users opened accounts somewhere else - and, most of all, a lot of people realized that almost anybody can set up a social network server. The sudden Twitter❤️open-source and Twitter❤️open-protocols shift is quite familiar. Whenever it happens, it's because a company in a monopoly/oligopoly-like market has stopped growing, and the closed+centralized approach that made their fortunes (and allowed them to make profits without innovating much) has become too hard to maintain and scale. Whenever this happens, the company usually display a sudden burst of love for the open-source community, and it turns to them for new ideas (and to write code for their products so their engineers don't have to). They usually admit that the solutions proposed by the community and the committees for standards were right all the time, but they usually don't take responsibility for slowing down innovation by years while they dragged their feet and milked their cows. However, they still want a chance of running the show. They still want to lead the discussions around the new platforms and protocols, or at least have a majority stake in them, so they can more easily prepare the ground for the next step of the embrace-extend-extinguish cycle. Needless to say, we should play our roles so that such strategies stop being successful.

Is there anybody out there?

The open-source alternatives and the open protocols haven't succeeded in the past decade not because their proposed solutions were technically inferior to those provided by Facebook or Twitter. On the contrary, they had figured out the solutions to the problems of distributed moderation, federated authentication and cross-platform data exchange long before them.

They didn't succeed because it's hard to replicate the exponential snowball of a true network effect once all the people are already using a certain platform. Even if you pour a lot of time, money and resources into building an alternative (like Google+ tried to do for a while), people are naturally resistant to change, and it's just too hard to move them once all of their contacts are on a single platform. Especially when social networks are owned by private businesses that keep the barriers towards data portability artificially high.

So, even with all the advantages of a federated network of instances, the two titans still outweighed in an industry where the winner takes it all, and for a long time Mastodon and Diaspora instances were deserts comparable to Google+ - except for few enthusiastic niches, and for a few active instances run from places with strict social media limitations.

The wind has started to change in April 2022. And the EU has also recently announced further steps in enforcing their vision for greater digital interoperability.

After the early April diaspora I picked up my instance again, started following some new interesting accounts and federating with some relays, and there's now enough activity for me to use my Mastodon instance as my daily social driver. Even if the scale of the Mastodon network (around 3-4 million users) still pales in comparison to that of Facebook's empire, it starts to be a considerable fraction of Twitter's active (human) user base.

However, even if many influential accounts have moved to Mastodon (or at least they cross-post to Mastodon), such as The Guardian, Hacker News and the official EU News channel, there is still a big gap in terms of accounts and content that are only available on Twitter/Facebook.

So I took some initiative, and decided that if the mountain doesn't come to me, then I'll move it to me myself.

Creating a cross-posting bot

There are a lot of amazing profiles to follow on the Fediverse, but you also still miss a lot of the "official" accounts that make a timeline actually stimulating. In my case, it's accounts of publications like the MIT Technology Review, Quanta Magazine, Scientific American, IoT-4-All, The Gradient and The Economist that really give me food for thought and make my social media experience worth the effort of scrolling through memes and rants.

Those accounts are only on Twitter and Facebook for now, or maybe on some RSS feed. But Platypush also provides integrations for RSS feeds and Mastodon. So a bot that brings our social newspaper to our new doormat is just a few lines of code away.

Let's start by creating a new account on any Mastodon instance we like (if you don't host one yourself, just make sure that you are aligned with the instance admins and rules when it comes to bot activity). You can probably start your adventure with a bot hosted on one of the largest platforms - e.g. mastodon.social/mastodon.online. Specify username, email address and password for your bot, confirm the email address, login with the bot account, navigate to PreferencesDevelopment ⇛ Create a New Application, give it full access (read+write+follow+push) to the account, and copy the provided Access Token - you'll need it soon.

New application screenshot

It's also advised to navigate to Profile and tick the This is a bot account box, so people on the network know that there's not a human behind it. You can also provide a brief description of what profiles/feeds it mirrors so people know what to expect.

Bot account flag

The Platypush automation part

You can install and run the Platypush bot on any device, including a Raspberry Pi or an old Android phone running Termux, as long as it can run a UNIX-like system and it has HTTP access to the instance that hosts your bot.

Install Python 3 and pip if they aren't installed already. Then install Platypush with the rss integration:

[sudo] pip3 install 'platypush[rss]'

Now create a configuration file under ~/.config/platypush/config.yaml that enables both the integrations:

mastodon:
  base_url: https://some.mastodon.instance
  access_token: YOUR-BOT-API-ACCESS-TOKEN

rss:
  poll_seconds: 300
  subscriptions:
    - https://blog.platypush.tech/rss
    - https://nitter.net/hackernoon/rss
    - https://nitter.net/TheHackersNews/rss
    - https://nitter.net/QuantaMagazine/rss
    - https://nitter.net/gradientpub/rss
    - https://nitter.net/IEEEorg/rss
    - https://nitter.net/ComputerSociety/rss
    - https://nitter.net/physorg_com/rss

Twitter no longer supports RSS feeds for profiles or lists (so much again for the "Twitter❤️open protocols" narrative), and there's a multitude of (mostly paid or freemium) services out there that currently bridge that gap. Fortunately, the admins of nitter.net still do a good job in bridging Twitter timelines to RSS feeds, so in rss.subscriptions we use nitter.net URLs as a proxy to Twitter timelines.

UPDATE: nitter.net has got a lot of traffic lately, especially after the recent events at Twitter. So keep in mind that the main instance may not always be accessible. You can consider using other nitter instances, or, even better, run one yourself (Nitter is open-source and light enough to run on a Raspberry Pi).

Now create a script under ~/.config/platypush/scripts named e.g. mastodon_bot.py. Its content can be something like the following:

import logging
import re
import requests

from platypush.event.hook import hook
from platypush.message.event.rss import NewFeedEntryEvent
from platypush.utils import run

logger = logging.getLogger('rss2mastodon')
url_regex = re.compile(r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\(\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+')


# Utility function to parse bit.ly links content
def parse_bitly_link(link):
    rs = requests.get(link, allow_redirects=False)
    return rs.headers.get('Location', link)


# Run this hook when the application receives a `NewFeedEntryEvent`
@hook(NewFeedEntryEvent)
def sync_feeds_to_mastodon(event, **context):
    item_url = event.url or ''
    content = event.title or ''
    source_name = event.feed_title or item_url

    # Find and expand the shortened links
    bitly_links = set(re.findall(r'https?://bit.ly/[a-zA-Z0-9]+', content))
    for link in bitly_links:
        expanded_link = parse_bitly_link(link)
        content = content.replace(link, expanded_link)

    # Find all the referenced URLs
    referenced_urls = url_regex.findall(content)

    # Replace nitter.net prefixes with twitter.com
    if '/nitter.net/' in item_url:
        item_url = item_url.replace('/nitter.net/', '/twitter.com/')
        source_name += '@twitter.com'

    if item_url and content:
        content = f'Originally posted by {source_name}: {item_url}\n\n{content}'
        if referenced_urls:
            content = f'Referenced link: {referenced_urls[-1]}\n{content}'

        # Publish the status to Mastodon
        run(
            'mastodon.publish_status',
            status=content,
            visibility='public',
        )

        logger.info(f'The URL has been successfully cross-posted: {item_url}')

Now just start platypush with your local user:

platypush

The service will poll the configured RSS sources every five minutes (the interval is configurable through rss.poll_seconds in config.yaml). When a feed contains new items, a NewFeedEntryEvent is fired and your automation will be triggered, resulting in a new toot from your bot account.

Some cross-posts from a bot timeline

If you like, you can follow crossbot, a Platypush-based bot that uses the automation described in this article to cross-post several Twitter accounts and RSS feeds to the platypush.tech Mastodon instance.

Some performance considerations

Note that on the first execution the bot will start from an empty backlog, and depending on the number of items in your feeds you may end up with lots of API requests made to the instance. Depending on how large (and how bot-friendly) the instance is, this may result either in a (small) DoS against the instance, or your bot account being flagged/banned. A good idea may be to throttle the amount of posts that the bot publishes on every scan, especially on the first one. A few solutions (and common sense considerations) can work:

  • Start a [Python Timer](https://www.section.io/engineering education/how to perform threading timer in python/) when a new item is received, if a timer is not already running. Every time a NewFeedEntryEvent is received, you can append the event to the queue, and upon a selected timeout the queue will be flushed and the most recent n items synchronized to Mastodon.
from queue import Queue
from threading import Timer, RLock
from time import time

from platypush.event.hook import hook
from platypush.message.event.rss import NewFeedEntryEvent

# How often we should synchronize the feeds
flush_interval = 30

# Maximum number of items to be flushed per iteration
batch_size = 10

# Shared events cache
events_cache = []

# Current timer and its creation lock
feed_proc_timer = None
feed_proc_lock = RLock()


def feed_entries_publisher():
    # Only pick the most recent events
    events = sorted(
        filter(lambda e: e.published, events_cache),
        key=lambda e: e.published,
        reverse=True
    )[:batch_size]

    for event in events:
        # Your event conversion and `mastodon.publish_status`
        # logic goes here
        try:
            ...
        except:
            ...

    # Reset the events cache
    events_cache.clear()


@hook(NewFeedEntryEvent)
def push_feed_item_to_queue(event, **context):
    global feed_proc_timer

    # Create and start a timer if it's not already running
    with feed_proc_lock:
        if (
            not feed_proc_timer or
            feed_proc_timer.finished.is_set()
        ):
            feed_proc_timer = Timer(
                flush_interval, feed_entries_publisher
            )

            feed_proc_timer.start()

    # Push the event to the cache
    events_cache.append(event)
  • A producer/consumer solution can also work. Create a new hook upon ApplicationStartedEvent that starts a thread that reads feed item events from a queue and synchronizes them to your bot:
from queue import Queue, Empty
from threading import Thread
from time import time

from platypush.event.hook import hook
from platypush.message.event.application import ApplicationStartedEvent
from platypush.message.event.rss import NewFeedEntryEvent

# How often the events should be flushed, in seconds
flush_interval = 30

# Maximum number of items to be flushed per iteration
batch_size = 10

# Shared events queue
events_queue = Queue()


def feed_entries_publisher():
    events_cache = []

    while True:
        # Read an event from the queue
        try:
            events_cache.append(
                events_queue.get(timeout=0.5)
            )
        except Empty:
            continue

        # Only pick the most recent events
        events = sorted(
            filter(lambda e: e.published, events_cache),
            key=lambda e: e.published,
            reverse=True
        )[:batch_size]

        for event in events:
            # Your event conversion and `mastodon.publish_status`
            # logic goes here
            try:
                ...
            except:
                ...

        # Reset the events cache
        events_cache.clear()


@hook(ApplicationStartedEvent)
def on_application_started(*_, **__):
    # Start the feed processing thread
    Thread(target=feed_entries_publisher).start()


@hook(NewFeedEntryEvent)
def push_feed_item_to_queue(event, **context):
    # Just push the event to the processor
    events_queue.put(event)
  • A workaround for bootstrapping your bot could be to perform a slow boot. Add one feed at the time to the configuration, and restart the service when the latest feed has been synchronized, until all the items have been published.

After the first run the feeds' latest timestamps are updated and they won't be reprocessed entirely upon restart. However, it's generally a good idea to keep your bot light. If it posts too much, it may end up polluting many timelines, as well as fill up a lot of storage space on many instances. So apply some common sense: don't cross-post the whole Twitter, or your cross-posting bot will not add much value.

The advantages of a cross-posting bot

If used and configured responsibly, a cross-posting bot can vastly improve the social experience on the Fediverse.

It brings relevant content shared on other platforms to the Fediverse, spinning off discussions and interactions outside of the mainstream centralized platforms.

It's also a quick and efficient way to bootstrap your new instance. Many new administrators are faced with a dilemma when it comes to kickstarting their instances. Either they go the conventional slow way (advertise their instance to increase their user base, and manually discover and follow accounts on other instances in order to slowly populate the federated timeline, hoping that users won't leave in the meantime), or they associate to one or more relays (some kind of instance aggregators that bring traffic from multiple instances to the federated timeline), just to be overwhelmed by an endless torrent of mostly irrelevant toots that will quickly fill up their disk storage. Such a bot is an efficient way in between: it populates your instance with the content that you want, it brings in some hashtags and links from Twitter that you may decide or not to boost on your instance, and it attracts people that are looking for curated lists of content on the Fediverse.

...but the Fediverse isn't all that rosy either...

After so many praises of ActivityPub, Mastodon and its brothers, the time has come to highlight some of their drawbacks.

I briefly mentioned relays in the article, and that's not a coincidence. Relays, if implemented, maintained and adopted properly, can be the killing feature of the Fediverse. No more cold bootstrapping would be required for new instances: as long as they share common interests and adhere to similar rules as other instances, they can easily federate with one another by joining a relay.

A relay is basically a server with a list of instance URLs. It subscribes to the local timelines of the instances and it broadcasts their activities over ActivityPub. Therefore, all the instances that are part of the same relay can see all the public posts published on all the other instances in their federated timeline.

Amazing, isn't it? Except that, as of today, the experience with relays is far from this vision of a curated and manageable aggregator of instance. There are only a few usable open-source relay projects, and most of them are still in a beta/pre-production stage. Most of the URLs you find on Reddit or on forums are no longer working. An up-to-date list of active relays is available here, it includes about 40 nodes as of today, and after trying most of them I can tell that they fall into three categories:

  • About half of them will turn your timeline into an endless torrent of spam and saturate your database. Most of them automatically accept any relay requests, and with no inbound filter spammers can easily take over. Also, with no clear mission/purpose/shared interests or languages, and poor filtering by topics and languages provided by the platform, after relaying you can expected your federated timeline to turn into a Babylon with all the languages and topics in this world. My database storage inflated by ~40 MB just a couple of minutes after joining the most populated relay.

  • A third of the URLs points to servers that no longer seem to accept relay requests, or with nearly no content.

  • The remaining ~15% points to a couple of relays that actually push not-so-spammy content in a manageable way.

At the time being I have joined those relays, but there's really no concept of curation/aggregation yet at the current stage. To me, relays should be to Fediverse instances what OPML is to RSS feeds and podcasts: a curated way to aggregate sources that share common traits, not a chaotic party where everybody is allowed to join. We don't seem to be at that stage yet.

It also doesn't help that the two main instances (mastodon.online and mastodon.social) aren't part of any relays. The only way to get posts from the largest instances pumped into yours is to follow individual accounts. I understand the challenges of having to moderate large-scale relays involving the two official instances, but I also think that if we keep the largest instances out of the relay game then we can't expect relaying to improve much.

On the contrary, I see the risk for things to evolve in a direction where large instances don't have any incentives in joining a relay, while relays are mostly run by hobbyists and end up attracting a long tail of unfiltered and non-curated traffic from all the other small instances. In such a scenario, most of the people will simply open their accounts on the largest instances, because that's where most of the things happen anyway. And then things will just swing back towards centralization. That's why I don't get those who praise decentralized social networks and then simply move to one of the two main Mastodon instances. Supporting decentralization isn't just about migrating from a large centralized platform to a smaller one. It's a much better idea to support a smaller instance: it'll still act as a gateway to follow and interact with anyone on the Fediverse anyway, while keeping the content really decentralized.

All in all, however, I still believe that the Fediverse is the only possible future for social media that is both scalable, portable and transparent. The current immature state of the relaying technology will probably be fixed one iteration at the time. And, even if Mastodon turns out to be a new centralized titan in the future, we can simply move our data and accounts to another instance running another server, just like we would move a website from a hosting service to another. Because, after all, data portability and interoperability is all the web was supposed to be about.