Model Context Protocol podcast mentions

Quote

environments and seamless data exchange between workflows, but their existing tools couldn't deliver. That's why Cash App relies on Prefect. Now their ML workflows run on whatever infrastructure each model needs across Google Cloud, AWS, and Databricks. Custom packages stay isolated, model outputs flow seamlessly between workflows. Companies like Whoop and One Password also trust Prefect for their critical workflows, but Prefects Didn't stop there. They just launched FastMCP, production ready infrastructure for the first time. AI tools. You get prefects orchestration plus instant OAuth serverless scaling. Claude, cursor, or any MCP client, no more building auth flows or managing servers. Prefect orchestrates your ML pipeline. Fast MCP handles your AI tool infrastructure. Structure. See what Prefect and FastMCP can do for your AI workflows at AI Engine. Podcast dot com slash prefect today. Bruin is an open source framework that streamlines integration from the command line, allowing you to focus on what matters most. Write Python code for your business logic and let's see. and handle the heavy lifting of data movement, lineage tracking, data quality monitoring, and governance enforced.

Quote

Yeah, that uh code red was interesting. 'Cause then five point two dropped uh last night, last afternoon, whatever it was. Um it's It made a really big improvement in an area that a lot of people will care about, uh which is actually making deck. and spreadsheets um per request, which is really cool. What are you uh kind of a last question or two just before we close things out. Have you um guys as a company? kind of taking advantage of MCPs at all? Um what are your Thoughts on how those are. Uh currently I think there's a lot of really cool efficiency improvements that'll come out of MCPs. And I think the concept of making a person five percent, ten percent 20% more efficient. My answer to that question is teaching someone how to Properly use MCPs. a big question to wrap on, but uh we are seeing MCP uh MCP's pretty critically as a way to be able to expand the capability of our systems for our customers. And part of the reason why is um For legal interest, you can see that the first time we're interact with other systems and other software which those customers use. So

Quote

need and what it should look like and now I can d do it properly and I did. And it's like this super crazy event driven system that like allows you to communicate between the client and the server like easily and it's like so many things at once and Uh even I'm even looking into like adding an M C P server that lets you talk with your Yeah. Actually like literally just like two minutes ago thinking like, oh, an MCP server in the dev tools would be amazing. Because like if you have all of these dev tools that hook into this common interface for helping you. understand your application. That sounds like a treasure trove of information for an LLM when it's trying to de Yes. And that that that's like the the next thing I'm working looking for dev tools. Um it's it's kind of close, it's kind of not Um I I just need to figure out like w where to plug it and how to make it But yeah, I I I I need to like talk with Tanner and the rest of the first time. the team to like figure it out because they have a lot of other stuff for agents that we're working on that has not yet been really revealed so um hopefully he's a little bit more than a little bit Maybe on that note, uh it'd kind of be cool to talk a little bit about ten

Quote

healthier. Uh it really comes down to really strong reasoners that can reason well. over context length and in in general across domains. Building agents that can learn new tools. Uh talk a little bit about We've seen with oh hey, MCP's this great thing. Let's all use MCPs. Whoa, MCP's blow up the context and uh y y use more of it than the question that The agent is trying to solve. Uh, let's instead, you know. delegate it to software programs that the agent can write so it doesn't have all that context. And then oh no, well that is kinda hard. Let's actually make it so that the agent can search. Like if if there's anything on this list that cries for a more foundation foundational solution. Maybe it's like tool use and and an agent's ability to to learn new tools. Is that kind of the thing that you're Yes. So I think what we have taught our models to kind of learn so far is really what we put in conversation.

Quote

So make it as general as possible and kinda let them come up with what Whatever workflows you know works well for them. People use it for all sorts of things outside of coding. Our product marketing guy, Nick Bauman, he uses it to connect to, you know Reddit M C P server scrape content connected to an X M C P server and and MCP kind of lets it function as this everything agent where it can connect to you know whatever services and things like that. And that's really a a side effect of of having very general prompts. a conference in Amsterdam and I built my whole presentation, my whole slide deck. Using this library, it's like a JavaScript library called SlideDev. And I just asked client, like, hey, like here I wrote like a big Klein rules document explaining I told a client like the agenda. I kind of recorded using this other app called Limitless, like transcribe my voice. So you know, Klein really can do anything.

Quote

If we draw that parallel to the agentic world, a lot of people might be creating the their very first agent right now, right? Like, oh, I created an automation That does X, right? And if we look around that automation, there's kind of a There's connections to systems. So that agent has actually a number of things. In and of itself. It's multiple models, maybe an LLM, embeddings, re-ranks. It's a connection to an MCP server or more. It's API calls, it's some workflow code, it's a user interface. That people interact with. And so that's agent one. And then you have to imagine that as as these agents proliferate, right, you have tens of agents, you have hundreds of agents, you have thousands of agents that are all operating within your operational environment, your enterprise. And if you want to do that and manage that complexity, I think those are some of the problems that are really going to be high value in in this space and you know uh we're thinking about some of those as related to like governance and and policy and enforcement and and monitoring, but there's innumerable other problems there that you know how to do that. you manage all those MCP servers? How do you um you know how do you handle

Quote

So t tell us how you do it. All right. So I'm still working through this, but But what I found is useful is to have feedback loops. Oops. So are you familiar with the playwright MCP? No. S so playwright is a um Um a browser use uh agent. So it can it's it's an MCP tool, so it's like the model context protocol, but basically like MCP for those who aren't familiar is um a thing that agents can use as tools. So remember LLMs, like you know, all the models that we're talking about right now. Like, you know, you're you're uh Gemini's, you're open AIs of the world, your anthropics, your mini max, your uh GLM five point. one uh your gem of four um pretty much the only thing they can do is is uh text in and text out, right? Now some of them are multimodal models. where they can analyze some images and whatnot. But at the end of the day, um they're just you know, uh token predictors, right? Um and they can uh well uh you go and then like click around on a browser for you, right? Um so like they can

Quote

For problems worth solving, get started with Claude at Claud.ai slash superdata. Claude.ai slash superdata and check out Claude Pro, which includes access to all of the features mentioned. mentioned in today's episode, claud.ai slash superdata. That makes a lot of sense to me to use Agent Force 360 to accelerate getting agents that are secure and reliable into your customers' hands because I can imagine Imagine if you were using tools like MCP or OpenClaw, if you were kind of putting those kinds of things together, open source options for creating agents, there's all kinds of known security issues with those kinds of specifically the two That I mentioned there M C P and OpenGlaw. Very true. Very true. So I mean these are fast evolving technologies, right? We don't rule this out in the future. But at the same time, these things have to get uh they have to move out of the prototype. and the the uh those stages into more mature technologies where security and other things start to play a very important role, especially in uh enterprise software. software. Right. So at which point we'll be ready to adopt those technologies as well. And so with Agent Force 360 specifically, how what is it like if our Listeners are like, I want to be able to build an enterprise ready agent.

Quote

Don't use code execution or command execution to do so. Do the proper implementation in your code and your tooling and your whatever you need to N C P server if you want to connect to something external, but do it as constrained as possible. That makes sense if you can control all these factors, but as you mentioned, the external M C P server you don't really know what you're getting there. How do you make how do you validate? How do you make How do you make sure that MCP servers you're connecting to are not gonna do weird stuff? Yeah, you'll have to do anything. I mean it depend it what if you control the MCP is easier, right? The MCP server. But if you don't, which most people are not, they're just gonna use whatever is available on uh in the public libraries. or public pro or other products, then yeah, you'll have to assess properly. Uh there's no solution nowadays that I think can do this. There there's tools that allow you to scan and N C P servers, but there's nothing out there that will be for your pipeline of development pipeline to give you that hey if you're gonna implement this MCP server, you're gonna uh run this and uh this gonna scan the S observer and tell you the sort of the tooling uh the the tooling that's available to that server. or through that server, these are the capabilities, these are some of the issues, right? And even the other way around, like how…

Quote

But then also like how should people think about layering this on? check. You know, should we check after generation? Should we check before generations? Should we do both? Uh, you know, should we be Integrating other searches as well. Yeah. So uh on the documentation we do have have a way to connect our MCP server as a connector. to Claude and directions for um ChatGPT as well. Um at generally these large language models, when you ask me why They haven't uh lowered their price per API call for the same. Search. One of the things that's uh pretty well known is that uh the the GROC models, XAI models, um called Brave and uh anthropic calls brave if you're in clawed code it even tells you Hey, I'm calling Brave. So they're kind of uh you know stuck with that pricing. Um you know, and even if they get a discount, you know, it's they are really stuck with the brave pressing and then um the overhead of calling etc. So I think that's one of the reasons why the price hasn't

Quote

Interesting use cases that are very similar across both. see at weights and biases is um a lot of our Customers are more interested in in creating their own air. interfaces, like it seems a lot more possible to kind of vibe code like a quick and dirty, you know, dash That's sort of outside of our own our own product. Like we've been thinking a lot about how do we make like you know APIs, maybe MCP servers or you know, these types of of like interfaces to support customers in that way. Do you also kind of agree with or do you do you feel that trend also? Uh yes, for sure. Um I think this gets to this question around software development and and I think I think I suspect what we might end up doing is separating the nature of software development. from the nature of creating technology. And this gets kind of blurry, right? If you think about what a lot of kids do with their iPhones nowadays, you'd argue they're kind of making technology in a lot of ways. But we don't think of them as software developers. Um vibecoding if we take a a Uh a definition being a largely non-technical person. That can be a product manager that's a little bit technical. It can be a finance

Quote

That access. It's your proxy. That makes sense for a personal assistant. It's just an extension of me. It does. But then you have the second framework, machine identity. This is For fully autonomous agents that run in the background, maybe doing nightly reconciliations or ordering. So they need their own passport. They need their own cryptographic passport to move through the network. Absolutely. This is where things like MCP, the model context protocol come into play. Let's touch on that quickly. MCP's popping up. Think of it like a standardized USB port for AI. It lets a model connect. Different data sources, Google Drive, Slack, the database in a uniform way. But even with that standard connection, you have to verify who is clucking in. Because you don't want your support. supply chain agent accidentally getting access to the payroll system. Exactly. Not if it doesn't have a Distinct verified identity, we're leaving the door wide open otherwise. Okay, so we have the four pillars. Visibility, contextual GPU security, Probe to rails, and non human identity. But how are companies actually? Most aren't building it from scratch. They probably shouldn't. It's too heavy a lift. This is where we see the approach from companies like