JFrog’s Yonatan Arbel on the Evolution from CLI to Secure Remote MCP Servers | swampUP 2025

Hey, everyone. We're back here with day two coverage of, uh, swamp Up. I'm happy to have my next guest on here.

His name is Yan Arbell. Y Yan. I've, I've interviewed him before.

He's one of the unsung hero stars here at Jfr. He's a frog. Yo, naan, welcome to Text Drunk.

Always a pleasure to be here with you. Alan, Pleasure to have you on Before we jump into, looks like we just had a plane, Passover. Yeah.

Something, huh? Yeah. Before we, uh, jump into what you spoke about today, if you wouldn't mind, talk to the group.

Tell them a little bit about what you do with Jfr, where you've been, how you got here. Right. Uh, okay.

So, first of all, I've been with JF Rog for the last nine years, uh, doing a lot of stuff. I started as a software developer. Uh, afterwards, uh, like few years later, I was a team leader at the infrastructure group, uh, where we kind of, uh, planted the seed for the platform, what we know today as what we know for today.

Okay. Um, and later on, um, you have Jfr CTO approach and say, Hey, Anan, you're so active on the networks. You wanna be, uh, dere and start, you know, communicate, bridging the gap between what Jfr does, you know, outside to the world.

So, you know, kind of try to simplify things. 'cause Jfr, you know, we do a lot of stuff from the most left side to the right side. There is a lot to cover, and we need to somehow order all this building blocks, so it'll be easier to, you know, to grasp what we are doing.

Excellent. Now, so you moved into from software development to Derel, Right? Right.

I moved in from software development to Dere, but, But always a, but yeah. Um, it's, so I'm, I'm, I'm so grateful that I can work with Joab because he is my CDO. But not only that, because, uh, giving me the freedom to also explore and keep coding, you know, once developer always developers.

Right, right. And, and when people ask me, Hey, you moved from a development to be a dev and say, you know, it's like swimming. You, you, once you're a developer, you have the, the, you have the, you know, this infrastructure you can always get back on track and, and, uh, keep coding.

And I never stopped coding, actually. And in fact, jfr MCP server, our first open source experimental MCP server is server that IOP I created. And I, uh, you know, uh, uploaded to GitHub and we, you know, we started seeing some eye rolling on this, uh, on this project.

And, and it got, you know, lots of our customers got really excited about this MCP capabilities. Uh, we had PRS from Nvidia and even, uh, a guy from Cisco approached and say, Hey, we used your MCP. It was so good.

We, we even take it as a reference for our MCP. So it was, first of all, it was really flattered, uh, from it. Um, but yeah, I keep my hands on the, on the keyboard.

We, I keep coding and, you know, even today with AI is so easy. Yes. It, everyone can code Vibe, vibe coding.

Yeah. Vibe coding. I, you know, we, we say vibe coding.

I like the, the word I add sketch coding. Okay. Yeah.

Because vibe is like, you know, when I think about something I, I, I give the prompt and I, I start sketching what I want eventually. If, if I think if it's something that I, I feel that is can, can have some maturity, I will take it to the next level. And by the way, this is some, this is kind of how I build the MCP.

It start with sketching. And then I created, uh, a tool and MP and it opened another door. And I like text, uh, text my manager bragging about each new capability that I just created.

Hey, you know, that now you can ask about the production environment in runtime just from your id. Because I, I, it felt like a magic. It felt That's, you know, that is very close to how I use it too.

'cause I use it more for my stories and stuff like that and articles and scripts. And, um, I also, I, I never called it sketching, but that's what I do, is I, I sketch it out in an outline and then I keep putting more flesh on the bone. Right.

And then I ask it to polish it up. Yeah. Um, I wanted to come back to, and, and talk a little bit about you speaking here.

Well let, before we talk about that, let's go back to MCP server. Yeah. It's crazy.

In January, no one knew what it was. Right. MCP server today, everybody has an m it's become the standard way we're communicating.

Right. And, and you know, at first we heard MCP, like it was everywhere. And now we hear it about it, uh, a little bit less and not because people, uh, are not using it because No, it's built, it became the standard's built.

Yeah. It's built in already. It's done.

It's Built in. Yeah, exactly. Um, and, and it's amazing.

And we, we see the whole ecosystem around, um, you know, how the, it's not, it's not about only, uh, creating this new shiny model the strongest with the biggest context. It's now about how we as, as persons communicate with these AI agents and how AI agents communicate with another AI agents. So that's what it's, it's about the communication and interaction interaction between these AI to ai.

Yeah. AI To ai. Exactly.

Alright, let's turn to your presentation. Yeah. Tell us what you spoke about.

Yeah. So in, in my presentation, uh, I talked about the journey of, uh, what Jfr did from the moment, you know, AI came into our lives and, and like, like we talked yesterday and like gal mentioned today, we see every billboard out there saying how they are working with AI and what they're doing, uh, in order to be an AI company. Um, so ai, it's a tool it come to help us to, to expedite the, the, you know, the processes.

And I started by, by, you know, telling the audience about what we did, uh, at first with the JF, how it's like, it was kind of, uh, an, it started as an internal gig or experiment to see, um, if we can reduce the friction between developers and documentation. So if I'm a developer and I want to know how can I, uh, use the CLI in order to upload a new folder to Artifactory or, or add a property, I'll just do JF how, and then I'll ask my question and hopefully the model will return me, uh, the right command with the right, uh, arguments. And so I can just, you know, alright, yeah, that worked for me.

Let's go. And that's where, um, our, it, it didn't do a lot of, uh, you know, a lot of splash out there. Um, but yet it was a sign for us that, that the intersection between AI and DevOps does make sense.

And we can, uh, you know, get these tools to help our customers and even ourselves, um, by just, you know, having the right models, the right train, uh, train model and, and, you know, flow with it. And then we started develop more things and we see like, uh, and last year when, uh, I was on stage with, uh, with my partner in crime, Brian from GitHub, and we announced the, this, um, copilot extension Yes. For J Rog, I remember.

And, and it was great. It was finally our AI agent got a context of, of external service jfr in that case, uh, it has some caveats like the add jfr that you need to add at the beginning, like telling, listen, now you will take all the information from jfr. Um, it worked and it worked well.

Um, but the world needed something more. Um, I call it like a smoother, uh, process. So, and that's where MCP came in place and kind of, I don't wanna say made it redundant, but made it much more easier.

I don't need, I don't need to tell it anymore. Hey, when I'm speaking with you about, uh, I dunno, build repositories, uh, release bundles, you should understand that you have the tools that's relevant, uh, to do this operation with jfr. So LLM is smart enough today.

It has a context there in the size of a absolutely in the size of a book. It, it, it understands the entire project. Uh, if previously it got nervous from 500 characters, now it can understand a full project.

And you Know what, it's every day it gets better too. Is is the thing about it. And I think, you know, talking, we, so I talk to a lot of companies Yeah.

And talk to a lot of the companies building the LLMs and the models and so forth. Yeah. Even if they stop developing it now, it'll take us seven years to digest Yeah.

What's available to us. Right. Using the MCP servers, using, you know, what we have.

Right. It, it's, it's an amazing time to be involved in this. Yeah, absolutely.

Where can people get more information about the MCP server and about some of the things you're working on? Right. So, uh, alongside my MCP server, and once we see the hype around it, uh, we created a production grade, uh, on in SaaS.

Like we, we, we host it, we maintain it, we update it. 1. So, um, our SaaS customers can already, um, with it.

Yeah. For now it's for SaaS. And they can just, with, with two clicks, they can connect our MCP server with their environment, whether it's their IDE or an agent that they develop themselves.

They can, uh, connect to our MCP and we still have the experimental, uh, MCP server, which just go and type, uh, MCP, GitHub, J Froog, it'll get you directly there, Objective out there as Well. It'll get you directly there. So, Mr.

Devereux, where are you headed to next? Well, you know, we, we always explore and we see we, we saw this, uh, great announcement during, uh, during Swamp Up. That even, even I need to explore and understand more and, and, you know, put my hands and get into these trenches and start work with it.

It looks super exciting and it make lot, lots of sense. The, this, uh, you know, all the announcements about the, the partners and, and everything. It's like, whoa, okay.

We, we need to see. It's, it looks so amazing and we can get, we can do so much things with it. So, um, mainly explore, explore around all how we can, you know, walk with this ai, integrate everything together so everything will fall into place.

I have no doubt it'll happen. Cool. Alright.

Pleasure seeing you. My friend. Jon, you Yan our bell.

Thank you. Dere, but always a developer. Once a developer.

Always a developer. Exactly. Here at Jfr Swamp up.

We're gonna take a quick break. We've got a lot more coming.