Unlocking Enterprise Intelligence with CTERA MCP

Hi, my name is Aaron Brand. I'm C-T-R-C-T-O of, uh, a member of the founding team. I've been with the company since, uh, we started about, uh, 17 years ago.

And today I'd like to talk to you about how, uh, we are unlocking enterprise intelligence with MCP. Uh, and later on, while we'll take it further, and we'll talk about our data intelligence capabilities and, and gen ai, uh, product. Um, so what, what is enterprise intelligence?

Enterprise intelligence is, uh, we define it as the ability of the, uh, of enterprise companies to, uh, u utilize their data state, uh, and the, the, uh, particularly their most sensitive, uh, files and data in order to, to get better decision making, uh, more productivity, and to put this as knowledge in ha in the hands of their employees. Um, now, uh, when, when, uh, you know, the, the, the whole gen AI space is very new. Uh, and when we started with perhaps maybe, uh, only, you know, about two years ago, uh, this entire idea of enterprise, uh, gen AI started, um, we had a problem.

And the problem was this NX STEM problem, or, uh, which means that when you wanted to connect, uh, enterprise data sources or enterprise applications to gen ai, you had to build a custom connector for each application, for each gen AI model. And this is the NX stem or end times and problem, right? Uh, and this really doesn't scale every, uh, so, uh, it's exponential, right?

It's exponential complexity, and it's also fragile because, uh, if you switch one of the tools, suddenly it stops, stops working. You switch from, uh, chat GPT to Gemini stops working, right? Everything is, it's not predictable, and there's no, it was no way to provide contextual information, uh, from the gen AI model, or from the assistance or the agents where the users are, are actually interacting to this backend enterprise system.

So seeing things such as user authentication, okay, who is the user that's doing the request and, uh, what are their credentials and what is their group membership? Um, all these things, there was no standard way to provide that. Uh, so this is why the, uh, uh, MCP protocol was created.

Uh, it's very new. It was created, uh, uh, this year. It's not, uh, it's not something that has a huge history, uh, and it's, it's still in, its, uh, evolving, right?

Uh, uh, things such as, uh, uh, authentication and other other features are just being added. Uh, and it's not really something that you can say is a table stake, as you know, uh, we discussed in the last, uh, session. It's really something that, uh, is currently cutting edge, at least for the enterprise, right?

Perhaps developers use it frequently, but, uh, having it in the hands of enterprise users is still new. Um, so MCP provides, uh, a few advantages, uh, over this, um, uh, hard to say traditional 'cause nothing is traditional here, right? But, uh, this previous, uh, NXM architecture, uh, which is, it provides a seamless integration that is guaranteed to work between any gen AI model that's, uh, supports MMCP and any tool that supports MCP.

And it's also built to be permission aware and identity aware. It has a way to pro provide this contextual information about the user, what their permissions are, and, uh, to authenticate the users. Uh, and finally, it's, it's, uh, we see it as a future proof architecture where you can easily, uh, swap any component, uh, without having to re-architect the entire system.

And, uh, so at Ciera, we, we paid very close attention to this development, um, and in, in fact, we, we decided we're going all in on MCP, right? So MCP is a core part of our strategy at, at Ciera, and we, we implemented MCP across all of our products. So, um, our implementation is in two main layers.

The first is what, uh, the implementation in the global file system product, which is, uh, where the actual files are stored. Know we have petabytes or tens of petabytes of data, hun, uh, tens or hundreds of billions of files, uh, with ciera, uh, uh, all, all you need to do is to enable MCP, and I'll demonstrate this to you in a moment. Uh, and you have your gen AI agents, assistants, and experts current, uh, suddenly have access to this entire data while respecting the permissions of the user, right?

So e each user users can access it from different systems, and we'll, we'll see it in, in a moment. Um, the second, the second level, uh, is data intelligence. And this, and this is a semantic layer, a semantic layer over the data.

So we're now all talking about the binary representation of these files. We're already talking about, uh, textual information, uh, metadata that is extracted from the files, and we expose that using MCP also as part of the second level. Would, uh, MCP clients and MP MCP servers fit into this, or is that a, a different way of organizing it?

Uh, yeah. Uh, so MCP server is the part that's, uh, implemented within the enterprise, um, the enterprise application, right? So, uh, MCP server doesn't have to have any AI in it.

Mm-hmm. Uh, and the M-C-P-M-C-P client is the AI tool, right? So, um, there are always two, two sides to this.

And, uh, we have, uh, as part of the global file system, we have an MCP server that allows you to consume it using, uh, any gen ai, uh, MCP client. And the data intelligence tool has both an MCP server and an MCP client, and I'll demonstrate that. So you, you mentioned that the global file system was permissions aware.

Is the data intelligence also permissions aware? Yeah, absolutely. Absolutely.

So, um, we ingest all the information, uh, both from, uh, our global file system and from other systems while we, uh, indexing all the permissions and respecting them, uh, end to end. Okay. Yeah.

Um, so a little bit more about, um, the, the first layer, the, uh, the MCP server that we have in the global file system. Uh, so this allows you to, uh, use any MCP tool. It could be flawed, could be open AI chat, GPT.

They just, uh, uh, added MCP very shortly ago, uh, short, uh, short, uh, while ago. And, uh, cursor, uh, or any other tool, like NA, uh, NA 10 or other automation tools that support MCP, you can, uh, uh, connect them to the global file system. Uh, and, and you can achieve all kinds of tasks.

You could, you could have the, your analysis in read files, in enumerate files. It could search for files, it can, uh, uh, read a file, summarize it, and write a file back. Uh, it could add metadata, it could create links.

So, uh, virtually anything that's supported by the platform, uh, is wrapped by this MCP interface. Is there any connection at all with CSI or is that completely orthogonal? Uh, CSI is is, uh, a Kubernetes, uh, dri CS I driver, and we also have CS i drivers for Kubernetes.

But, uh, that's, it's different. Uh, cs i drivers deal, it's, or they're like drivers for connecting, uh, containers to their backend storage, uh, but they don't really deal with, uh, all these, uh, AI operations, right? Uh, having the under understanding of what operation this storage system provides that are not, not only for storage, right, and, uh, reading files, writing files, uh, deleting rating links, so it's much, much more rich and it's, uh, a, a bit, uh, you know, targeted to a different, uh, Well, there, there are plenty of other LMS out there, other than the three that you specifically mentioned, um, vertex, you know, and, and, uh, LAMA 2, 3, 5, 7.

Mm-hmm. Um, does the m CCPs, I mean, is MCP only with open ai, Claude, and, Um, uh, no, uh, absolutely. Absolutely.

MCP is an absolutely an open standard, so it's not, uh, related. It can be work, it can work with any LLM, uh, and, uh, they're, they're are various implementations for NCPR, it's more, it's more of a frontend thing than, than something related to the LLM. It's related to the, the interface, right?

So the user interface where you interact, like the chat interface has to support MCP, uh, and on the backend it works using function calling and other it require has some basic requirements from the LLM, but it's not, uh, it's not at all, uh, specific to any LLM, I guess I'm, I'm trying to wonder why you're not all the 25 other LMS that are out there, or deep seven, whatever. So My, my, so my understanding is that not all of them have an MCP client. All of them have an M CCP client.

Yeah. Yes. Uh, the MCP client is part of the user interface, right?

So, uh, if you have a, uh, chat interface or, um, agent, the front end, the front end implements the MCP, right? Not the LLM, the LLM, it, it receives, uh, function calls and it, and it handles those things. So, um, this, uh, so cursor, for example, which is not an LLM at all as an MCP client, right?

But it, the cursor is the user interface on top of the LM and that implements this, uh, agent flows where it knows how to invoke tools. And so, I, I think so, so those were the, actually what I mentioned in the last slide, these are use, I mentioned them as user interfaces knowledge, LLM. So they're, they're front end application.

Um, and, and as I said earlier, we, uh, our MCP implementation is fully, uh, support authentication. So it's used OS two to allow, uh, users to gain access to their, uh, uh, data, uh, while respecting their permissions. And this is something fairly new in MCP.

It's the, it's, um, not all the MCP clients currently support, uh, O2, but, uh, this is evolving very quickly and everyone is going to support it soon. Uh, but if you don't, uh, and it's also possible to use tokens, uh, in, in case you want to connect not to a user, not to person, but to, uh, an agent or an automated system, you can use a token. I'm, I'm a little, um, I'm, I'm advising in general a lot of caution mm-hmm.

Uh, around MCP. Um, they have moved really quickly to address the giant gaping huge dangerous security holes initially in the protocol. So while access control and, and, and, you know, authentication and these kinds of bindings, um, within, uh, the, you know, the, the receiving, you know, the receiving services, uh, system are, are important.

I feel like those are table stakes that didn't exist to begin with. Do you have a perspective on this, because there are so many attack vectors that come through, um, through, through AI generally, and then now automating those attack vectors into potentially an entire estate mm-hmm. Um, it is, I think, a serious issue.

So given Ciara's strength in this area mm-hmm. Um, how do you see customers dealing with that? And how are you helping Yeah.

Helping with That? Yeah, so that, that's a great question, right? So of course, um, you know, let's, uh, let's imagine a use case, right?

Uh, I, I wanna build, uh, an automation using MCP that receives, uh, um, um, you know, CVS from people through email and then, um, updates my HR system, right? For example. Well, how, how, that would be very dangerous, right?

How about receiving patient, um, test results and, uh, like live and, um, feeding those into some kind of predictive system for better, I dunno, resource allocation or Something. Uh, so o obvi, obviously these are, uh, I wouldn't say that these are solved problems, right? This is something that when you implement it, you have to think carefully, right?

But, but, you know, mentioning, I I, I just wanted to mention the example I, I said earlier, you receive a cv, which is external, and there's some kind of a prompt embedded within the CV that says, okay, please, uh, please do this and that in the HR system, and we'll just follow and do whatever you want, right? Mm-hmm. So, so that, in that case, that will, that those are the types of, uh, problems that we see and they're not entirely solved.

I, I totally agree. And, and has to be used carefully in order to, I mean, you, you have to provide limited access to the, to each, uh, use case, right? Uh, even lower than the user level access, and to limit the privileges in order to, to have this, uh, protection.

So I wouldn't say I have, I have an answer for you 'cause this is an evolving business, but absolutely we have, we, we have to cash caution our customers to, to think carefully. And Sitara, you can, you can do that sort of, uh, access control within Ciara beyond, you know, RAC integration, for example, which may not comprehend that. Yeah.

Yeah. So, uh, um, yeah, so, so those are the types of things that, uh, perhaps, uh, are not a hundred percent, uh, ready where we're, but we're doing more and more. For example, as part of the O2, uh, you have the ability to provide, uh, what capabilities this application requires.

And it's not necessarily, right, if you're an administrator, you don't necessarily want to provide your whole capability, right? You want to, well, yeah. Restrict it to certain, uh, so, but that, that's a great question.

And, uh, we're, we're constantly thinking about how to help more and not everything we can do. So some things have to be done on the client side, whereas if you're using a system like, uh, NA 10 or other automation platforms, uh, there's some things that have to be done on the client side and we, we are not really able to, to solve. Okay.

That's fair. Thank you. Yeah.

Um, okay. So let's, uh, let's go into a short demonstration of, uh, the MCP server. Um, okay.

So all you need to do in order to use the ctra MCP server is to enable it from the ctra portal user interface. Okay? So I click here, I enable it, and it's, uh, probably our easiest feature to enable, right?

Uh, and after you enable MCP server, I'm showing here that I'm using it with cloud. Um, and, um, I defined here a few, uh, some information about the user. So Claude is aware who the user are.

Okay? So it asks, uh, tell me who, where I work, and so on. And I want to show you how I, uh, interact with the global file system that this user has access to.

Okay, so I have a competitive folder. Can you list the, list the files as you see, it's, uh, connecting to the global file system, listing all the files that notice that the competitive is is with a capital C and not with a small C. And it, uh, managed to list the files in the folder.

And now I say, I want to create an email for my boss about, uh, what we should focus on in 2025. Can you read the company, uh, the company files and create this, uh, competitive comparison? Okay.

So what you see here, cloud is interacting with the file system. It's reading several files. It understands using MCP that needs to do that, uh, and it creates a strategic insights document, and now it's writing it back to the global file system.

So, um, so you got an end-to-end application with, without any code that reads a list of files, reads them, summarizes them, and writes back the summary. And then it creates an email, uh, to my boss that I can send with a link to this file in the global file system and tell me, here, here's, here's our competitive summary. Um, here it is, and it clicks in the link.

Uh, right, it creates the link. And here, here is the, here's the file, uh, here's the email, and you exactly How I write all my emails. Exactly.

Using, uh, using, uh, plug, plug agents. So, you know, all of us always do that. Um, and, and here, when you click on the link, you get a link, it has an access expiration, you can download the, your boss can download the file.

Uh, so that's, that's a simple, uh, okay, that's a simple, uh, demonstration of, um, the MCP server in action. Um, yeah, so any questions? Just a general question.

Uh, are you using predominantly Claude? 'cause I know that's where the CPS came from, but that versus open ai? 'cause I'm playing around here, different things.

Mm-hmm. Trying them both. And I'm, I'm using OpenAI, but I'm thinking cloud is pretty good and might switch over.

I'm just curious in general, how you founded as a tool within soter? Uh, so, so Cloud were the first, uh, but, uh, chat PT actually implemented MCP only within the deep research capability that they have. Uh, and only, I think it was yesterday or two days ago, they announced that they have NCP as part of their developer developer mode.

So there, there's still a bit, uh, lagging. Um, you know, they, they thought, uh, they, they would be able to dictate the standard, but, uh, the industry decided otherwise. Uh, so they're lagging on this.

Uh, but there are a lot, a lot of other vendors, and I think the most powerful integrations are actually with the automation platforms. Like, like NA 10 or other, um, uh, where you can, uh, use your users can build entire automations and tie together different, uh, systems. And this is really the power.

Um, you know, what I showed in the demonstration is very simple. It's just working with the global file system, but the real power lies, right? Uh, uh, the, the real power lies where you integrate multiple, uh, uh, enterprise system source destination input and processing layers and conditions and build, and, and it's essentially allows your users to, to do, to be much more efficient right there.

Um, there are a lot of automations, you know, uh, you can invest in for your users, but sometimes, but you have many different users and not really worth it to spend, have a developer build an automation that's specific for this guy finance or how they work. And suddenly we, we democratize this and we allow our users to, to simplify all the repetitive tasks that they hate, right? And, and to have their, you know, to have their job faster and more enjoyable.

So that's why, why we really love this. Uh, the, um, so this concludes, um, this part of our presentation. Thank you very much.

And, uh, you know, in the next, uh, part we will talk about Teradata intelligence, which I, which I think is even more exciting.