Swarm Community Call — July 2025

Hello everybody and welcome to July's Swarm community call. Um broadcasting live to all the corners of the interweb. Um that includes streaming on X. So hello there and you are very welcome to join the convers the conversation in discord. Thank you for coming.

My name is Antonio and I'm very glad to be your host today wherever you are within this pale blue dot. All right so what can you expect on the call today? Here is the agenda. We will start with core development updates with Nikki Paparapo and Callum Toner. Uh begin uh following the community talk.

We will speak about swarm MCP the bridge to a decentralized AI with Andra Saranji and Nugan and we will conclude the call today with community MA and open space for debate. Feel free to talk and to ask any questions here on the stage chat or in any of the forums that you have received if you're subscribed to newsletter. So let's begin with the core development update and for this for this month we will focus on the recently released B version 2.6.0 which landed on July 22nd after resolving issues and uncovered during testing and improving internal QA processes.

The team has successfully shipped a release that delivers a major boost in usability and performance. Um, and to talk about this, we have Nikki and Kellum to walk us through the key stages and what's coming up next. I will invite Nikki to the stage. One second, please. Hello, Nikki.

Hello. Hi, Antonio. Can you hear me? Loud and clear.

Stage is yours.

Nice. Nice. Nice. So, hey Antonio, hello everyone. Welcome.

Uh, glad to have you here for July's Storm community call. Let's dive immediately into what's uh new and what's coming next. So, on the 21st of July, we had the B2.6 released. And if you haven't updated yet, now it's the time.

This release brings significant performance gains, especially in f in how uh fast nose get up and running. One of the highlights is a noticeable reduction in warm-up time uh which makes uh syncing much faster and improves the overall uh startup experience. If you're running a node uh you will definitely feel the difference especially if you're uh managing uh multi instances or restarting um your nodes uh spinning up new nodes etc. So uh a quick uh but important reminder stake migration is uh required for those taking their nodes. This is part of the protocol changes uh introduced in this release.

So please check the release blog post for instructions. Migrate your stake and make sure that you're running the latest version. So while 2.6 six is settling nicely. Uh we are already working uh hard on the next release the B uh 2.

7 and it's shaping uh up to be another meaningful step forward. We have a big list of improvements and here what's in progress at the moment for the next release. So first of all as part of the in browser node initiative uh we are implementing um automatic TLS handling uh to simplify secure communication uh without manual setup. This is key to making um browser native nodes more accessible. In the aspect of uh the pulsync improvement suite, we are rolling out new metrics for pulsing.

This will help us better understand a protocol's behavior and uh guide the next round of tuning and optimization based on real world data. Also uh 2.7 will come with fixes on erasure coding and it's actually corrections to further stabil stabilize how uh data is broking broken into chunks and reconstructed a core part of uh SHM's uh availability which is the eraser codes. At the same time we're working on improving uh fits especially targeting uh better update handling and responsiveness. Um it's important to mention here that a small but important uh bug introduced in 2.

6 uh affecting um affecting postads uh snapshot syncing on ultral light nodes. However, uh this has already been fixed and will be included in 2.7. But if in the meanwhile this affects you directly, please contact us um as soon as possible. Um in 2.

7 we will also see improvements on how transaction fees uh are managed aiming for more efficiency and lower overhead across the board. And last but not least uh we are also bringing fixes on the sampling. The current current implementation uh of sampling puts too much load on game logic uh local DB and overall um load performance. Uh with the fixes coming, we expect to see big improvements in uh sampling efficiency. Um a significant step forward and most probably more uh fixes will be coming uh next.

Uh in parallel of course we are discussing our sweeps, discussing them and considering what's next. So please provide your input about uh our sweeps. Come with your suggestions or anything any feedback you have. So the BA 2.6 X already brings me uh measurable improvements especially around startup time and uh with 2.

7 we are continuing to polish optimize and improve the overall experience of running and building on swarm. Your feedback always helps drive uh all this. Uh so if you have tried uh 2.6 let us know how is it going. If you haven't uh now it's the time.

again. Thanks for being here and for being part of this journey. Let's keep building together. Back to you, Antonio.

Thank you so much, Nikki. Exciting times. All right. So, if you have any questions about it, you can always drop it in the chat. We Nikki and others will probably potentially answer to you.

Okay. Uh moving on to the next section of the call, we have community talk token. will speak about an exciting feature. It's called Swarm MCP, the bridge to decentralized AI. And in this section, we'll introduce a brand new initiative.

This swarm model context protocol, which is in proof of concept stage that is designed to position swarm as a core infrastructure for agentic AI. To talk about this, we will have Andres and and or Nug here. Let me invite them to the chat to the stage. Hello Andrash, can you hear me?

Hi. Yes, I can.

Okay. And will be coming in a second.

All right.

So guys, stage is yours. What is this MCP?

Yes. Thanks. Give me a moment. I prepared the presentation. I will share it.

Yeah, it works from my side. So I presume at least a percentage of the viewers can see it properly.

Uh yes, maybe um just let me know if someone can can see it and I can share the uh the presentation link as well.

Uh so yeah, so let's what what is Swarm MCP uh and I guess the subtitle says a little bit about it. So it's the bridge to decentralized AI and it's kind of like exciting um kind of like chapter I think for swarm and we trying to prove that swarm is more than suitable to um kind of like um solve a lot of um kind of ongoing um problems or like open up AI to a decentralized future essentially. But first of all, what is uh MCP? So if you don't know uh it's a abbreviation for model context protocol. So this is a kind of a standard an open standard that entropic uh the creators of code uh AI models have come up with um I guess recently and technically it's kind of like a universal adapter for AI tools.

So kind of like a plug-andplay system. So it solves a problem where um you have a kind of standardized interface for your AI models and agents to be able to access certain tools and um this external tools can be anything. So in this case um swarm is an external tool. Um there are already some a lot of MCP uh implementations um connecting um agents to APIs and remote DB, local DB actually uh you can have multiple MCP servers um and then through those you can your AI can um you can empower your AI technically to to use a lot of tools and how it actually works just in a nutshell. So kind of there's a discovery phase uh when the MCP server returns a list of tools.

So in our case um it's formulated functions. I guess an simple example is like um give me the weather in a city. So it's like an API in in a very simple case in and we with swarm we try to do something similar but with swarm functions like um upload download but uh you will see it in a moment. So once the AI gen kind of knows about these tools and have some definitions what they can do there is the reasoning part. So the AI kind of can autonomously um deduce or reason about what tools to actually use based on the prompts or the kind of the task it is given or it has to fulfill.

And then uh when the AI reasoned about it and selected the tool uh it has to execute it. And then at that point technically just means that the AI uh calls the MCP server through the specific interface with the specific input and uh the MCP server executes uh the function and it will um return the uh result. In this case for our example it's it will return most of the time either a swarm hash or or like a sw downloaded something data from swarm to the agent and the MCP server will do that and how it actually works and why actually we want to build a swarm MCP. So the main problem is actually that agents um also need persistent memory that's one thing but if you want to kind of uh implement that then of course like with any kind of data currently the most of the solutions are centralized and uh of course they will always be a single point of failure. uh also private data comes into the picture uh because uh of course like uh one strength of swarm as well that uh you know it can be very private resilient and accessibility is always there.

So uh the solution uh kind of a no-brainer for this that uh you know AI can use swarm for resident private and decentralized storage uh like like a lot of use cases as well and and then technically if we do this swarm MCP uh server bridge then the MCP can handle a lot of complexities uh that uh that are involved when you use swarm. uh so like you know like post systems and uh even just to know the a the API the B client API so this can be all kind of abstracted away with the MCP so the all kinds of agents and AI LMS can use swarm directly without kind of any even like without any prior knowledge um about the swarm infrastructure and so how it works actually uh so there is actually P which is already um live and running and um technically you can try it out as well and we'll have a demo shortly. So uh it has a few um kind of core functions you know it can upload data it can download data. So in this example like let's say we want agent A wants to upload something to swarm kind of like keeping it in memory like maybe to share it with another agent then it can upload it to swarm using uh the upload uh function from the MCP uh and then it will the upload function will return a swarm hash. So the agent will will have access to the swarm hash and then either the agent can use this to download the uh data at a later time like maybe the agent has multiple subtask with have to or maybe give it to other agents and then it comes the hash comes into the picture in an easy way because then um you know the agent uh can share the hash with another agent or just with the user but let's say uh like another agent and another agent can call the download function from the MCP server and then uh it will receive the data from swarm directly from swarm.

So there are no intermediaries there uh other than MCP server that of course you can run locally as well and then uh yeah so technically this is a very simple use case but the POC can already do much more like uh uh uploading files even entire folder structures actually and then it can even use a feeds uh for kind of like a sequential memory. So um let's say um a good use case here is to have maybe like stages on completing a task where the agent can update the feed with maybe another agent can um check that feed and uh see the last result on the last index on the feed or it can just use the feed as a kind of a logs of steps of results. And so there are many use cases there because so kind of like whatever swarm can do the agent can uh use all those features as long as it's implementing the MCP as a function. And so what are the future plans with this? So we are working uh the swarm foundation and solar punk together working on a road map which is actually goes beyond the MCP.

So um MCP is kind of like the big cornerstone of it. But we are looking into how to use swarm in agent to agent communication uh to agent discovery on the swarm network. So imagine a kind of registry of agents where you can directly access uh agent information on swarm and uh kind of browse um based on capabilities of agents and then you can find or kind of discover the right agent for you. So technically you can um build uh fully autonomous agents that they can uh discover each other or share uh data with with each other for example using even act. So uh with ACT the access control try which uh agents can even share data privately.

Uh so technically you can create like a closed loop of agent economy. Uh and we also technically looking into how to make an actual economy. You know, imagine like um agents um performing tasks and then they get paid for it uh either autonomously or with some human involvement. That's also something that swarm can um help in even the whole communication agent to agent can technically go through swarm. So these are all the things we are looking into and we also of course want to um aid wider adoption.

So we're thinking to integrate into major AI frameworks like langchain or any partner that are interested to have swarm MCP integrate into kind of the capabilities uh or or rely on swarm as a as a kind of unstoppable memory layer as well. And um of course uh so since MCP is just a PC for now, we have plans to make it production ready solution um uh I would say relatively soon. And with this uh I guess you're all eager to see what how it actually looks or what it can actually do. So it's a it's a bit of a magic. So I I would ask now to to show you how it works because we have a little demo to show how how um MCP can already be a very powerful solution in even in the PC phase.

So please thank you. Uh okay. Um I share my screen and um hope you see it. So this is the uh the we of course it's open source so you can access it from the interspercp URL. As you can see here, there is a um bit was um read me description documentation about it, but don't be scared about that because it's quite easy to fire it up.

So, first uh it's a it's a NodeJS server uh in the background. So, need a NodeJS environment to to start and it's on like like on mentioned it's a it's a proof of concept state. So, um it's it's not published anywhere. So you need to clone this repository and build. Um so here it is and you need first of course a u developer environment that uh supports the MCP communication.

Uh for the demo purposes I'm using this sort now but we also tried it out with cloud desktop. It works. So that's the building process is nothing else just executing npmci it will install the dependencies and then you just build the application with npm run build after that you uh can configure your endp server now I just have only one this uh that I built it's configuration uh you can see it here Sorry.

Yeah. Apparently, nobody can really see the screen, the shared screen. So, could you re like just try it again because Yeah.

Yeah.

We can't follow up follow.

Um,

and it's a shame not to follow because it's such a cool demonstration. I saw it before, so I know.

But I guess majority of

I played the game and now do you see something?

Oh, yeah. We see discord. Okay. All right.

Okay. And now

nice.

Okay. Okay. Cool.

Cool.

So there were two comments so far. This mpmci and mpm ram build. And this is the beans serve environment. This is basically a VS code fork um having this MTP communication um feature. And um so here is the configuration of the MTP.

Uh as you see it's it's really short. You need to define the full path for the node execution environment and and pass your build uh path for the MCP server and there are also environment variables but below that also there's I mean by default you going to use the gf of spawn but uh of course it's always better for uh more advanced users to use their own local node for privacy and stuff and So um let's let's see like uh it already recognized five tools um in upload text download text upload file folder download folder and uh also it comes with the feed function didn't mention let's uh just try it out like let's say upload I will beautiful bird the swarm and the agent already like listed from the MCP tool like what capabilities it has. It's going to use the proper one that shows like uh um it it can it can execute and and perform this uh oh well uh I I don't know what can be the problem at the moment but I never got this error before and like the factory Um let's let's try it again by um restarting and uh let's see uh the maybe the uh just stopped because I changed my further. Sorry about that. So let's try it again.

successfully uploaded to Swarm. You can access it by an URL, but you can use this swarm reference hash. Uh and and in another uh session you can download this text. Oh, that was nice. Thank you so much, No.

Thank you, Andra. Uh, if anyone had any questions, please pop it up in the stage chat and we will promptly get through it. Thank you for that for the demo. Uh, despite the problems, but uh, you managed to pull it through. So that's always successful.

Thank you. And let's move on then.

Maybe back to Anton.

Can you hear me?

One, two, one, two. Uh, this there's a

hear you.

Okay. So, I guess it was new who couldn't hear me. Yeah, I got a

Sorry. All good. All good. No. Thanks.

Thanks, man. Hello. All right. So, uh let's move on to the next section. If so, as I said, if you have any question, uh please pop it up in the state chat where Nugan and Andreas will promptly answer it.

And let's move on to our our third section of today, which is the AMA where we have received uh qu not so many questions. Maybe it's because it's summer and people are a bit lazy with all the hot. But we have received one which is a bit recurrent, but it talks about uh and I'm going to just read it out bluntly. It says, "Are there any plans to stimulate the price of PCZ?" Well, that's that that's it.

There's no no more questions. So um I just would like to say that for the sake of transparency which is always a goal and the place where we see the foundation going uh any movement that there to stimulate the price or to move it will be properly communicated. So as long as you keep around and you keep subscribing and participating in the calls, you will you will hear all the action that will potentially take take place because uh this is it. This is a this is a commitment with transparency. So yeah, don't worry about it.

Just stay around and you will hear what is going on. And I don't know if there's any other question in the state chat. Uh so that's it. I am waiting for Mina to tell me. No, it's not.

That's not it. But uh I think I think it is.

It actually looks like that's it. We can mention that the next call will be uh of course last Thursday of the month as always which falls on 28th of August.

Awesome. All right. So see you there.

Um thank you Mina. Uh so that's it. Thank you very much to the speakers, presenters and all participants of the call. It's been great to hear your voice and of course thank you the audience. This has been the last call that I host.

It's been a pleasure to share these moments with you. Thank you for being there and see you on chain. We are all swarm.