Welcome, everyone. Today we're focusing on one of our AI defense pillars MCP scanning. Think of this as our pre flight check. When we point our scanner at an MCP server, we aren't just checking if it's online.00:05We are performing a deep dive audit.00:20I'll begin by setting the stage, looking at the new era of AI action and the MCP architecture. We'll examine the risks that emerge when AI agents connect to enterprise tools and why standard security just isn't enough anymore. Then I'm going to show you how we solve that with MCP scanning. We're going to walk.00:24Look through how we proactively map this new attack surface.00:44Identify every connection and build the visibility we need to stay secure. Let's dive in. We've officially moved beyond the era of simple AI chatbots. We are now in what we call the era of AI action.00:47AI agents are NO longer just answering questions. They are doing work. They are interacting directly with your enterprise tools, your databases, and your code repositories. They are executing commands and pulling sensitive data to complete complex tasks on your behalf.01:04This is a massive win for productivity, but it introduces a set of unique security challenges that traditional enterprise environments just weren't built to handle. When an AI has the keys to your code repo or your internal tools, the risk isn't just a bad prompt. It's a real world security breach.01:21Think of it this way. If an LLM is the brain of your AI, the MCP server is the bridge that connects it to the real world. It's how your AI accesses the tools, data, and services it needs to get work done. But that bridge is also a new attack surface.01:41If an MCP server is poorly configured, an AI application could inadvertently gain access to sensitive internal documents, databases or even cloud resources. The risk here isn't just bad prompts. It's real world data exposure.02:00That's why we built MCP scanning in Cisco AI defense. Cisco AI Defense scans these servers to show us exactly what AI tools are connected, what access they expose, and whether those connections create a security risk.02:16We're giving you total visibility and posture assessment for the tool layer of your AI. This is the slide I really want to sit on for a minute because it's the heart of what we're talking about today. When teams connect AI to their real enterprise tools and data, they're opening up a brand new attack surface, the tool layer.02:31Most security stacks were never built to see it. Cisco AI Defense closes that gap in two moves. 1st, scanning. Think of this as a discovery phase for your AI ecosystem. It's not just about general visibility. We're talking about deep code scanning of your GitHub repositories.02:52This gives you a clear picture of what's actually connected, what access those tools expose, and whether any of it creates risk surfacing things like prompt injection, exposure or excessive permissions. Second, guardrails. Once you know the risk.03:10You enforce against it in real time. This part we will cover in a separate video. So the story is simple. Scanning shows you the risk. Guardrails neutralize it. To give you a clear picture of what we're about to cover in the live environment.03:26I've broken the demo down into the two phases you see here on the screen. This is our roadmap for securing the AI tool layer. 1st, we are going to define MCP server scanning profile and set the scanning cadence. Then we're going to register and scan.03:43I'll show you how simple it is to bring an MCP server into the Cisco Security Cloud Control console and trigger an immediate scan. This is where we eliminate the unknowns and bring your AI tools under one security umbrella. Next, we'll move into findings and observability.04:00This is the AHA. Moment of the demo, we'll look at the findings from that scan, specifically identifying high risk threats like direct prompt injection, and I'll show you how we use risk profiles to categorize those threats automatically so you aren't stuck doing manual analysis.04:16So, let's stop talking about the how and actually go do it. Let's jump into the console. Once you're in the AI defense console, navigate to secure, then choose supply chain profiles under profiles.04:33Then navigate to MCP server scanning profile. As you can see here in our MCP server scanning profile, we're doing a deep dive into the MCP server implementation code and tool metadata. We aren't just looking for basic issues.04:50We're detecting everything from prompt injection and jailbreak attempts to malicious code execution, data exfiltration, and even tool impersonation. And the best part is that this is completely tailored to your organization. You have the flexibility to enable or disable any of these scanning rules based on your specific business needs and risk.05:08Appetite.05:28You also have full control over your scan cadence. Whether your security policy requires daily, weekly or monthly checks, you can simply set the frequency that works best for your team. You're in the driver's seat. Once you set the MCP server scanning profile, you can go ahead and register.05:31Register the MCP server, onboarding a new MCP server is incredibly straightforward. Let's walk through the 1st step of adding one to your environment. I'm navigating here to the MCP section under AI defense. Then click on add MCP server.05:51As you can see, this side panel makes the registration process clean and intuitive. You start by giving your server a name and an optional description to keep your inventory organized. Next, we move to the server details. We've made this flexible. You can either connect a remote MCP server by providing the URL and choosing.06:10In your authentication type, whether that's an API key or OAuth two.zero.06:30Or select source repository. When you select this, you're essentially connecting your GitHub repository directly to our security stack. It's a very streamlined process. 1st, you provide the repository URL and branch.06:38This tells the system exactly where your code lives. Next, you define the discovery path. This helps the system pinpoint exactly where your MCP packages are located. Now, this is a feature I really want to highlight.06:53The scan during registration toggle. I always encourage our customers to keep this enabled. It ensures that as soon as the server is registered, we're already identifying potential risks, vulnerabilities, and compliance issues, and it automatically triggers the scanning cadence you've already defined in your MCP server scanning profile.07:10It's a huge win for security posture. By being an advocate for this workflow, you're helping your teams adopt a secure by design approach ensuring that NO server enters your network without being vetted 1st. In my demo, I'm registering the Arna MCP server right now.07:30I've entered the name as Anna MCP server and added a quick description. Next, I chose remote MCP server option. I've input the server URL and selected none for authentication, as this is an open source server for our demonstration purposes.07:47Crucially, I've made sure the scan during registration toggle is enabled. Once you have your details filled out, you just hit register and the platform takes it from there. It's that simple to get started with robust AI defense for MCP scanning.08:03When I click register, the platform instantly begins the scan, and just like that, the server is registered and the scan is underway, and there it is, the Arna MCP server is now successfully registered and has been fully scanned. You can see it right here in our list.08:21Showing the latest scanned results from just a few moments ago, it's great to see the platform working in real time. We've already identified 19 threats, including that direct prompt injection risk we were looking for. To understand exactly what's happening.08:38Let's click on the server name to dive into the details. Now that we've clicked into the ARNA MCP server, we're looking at the heart of our AI defense intelligence. This isn't just a list of risks, it's a full diagnostic report.08:53Right in the center, you can see the threats by severity dashboard. We've identified 19 high severity threats. This is the kind of visibility that turns a potential supply chain disaster into a manageable security task. Down here, we have the top attacks breakdown. It's showing that.09:09That's exactly which tools are being targeted, like get_features_range and export_data. This is the level of detail that makes our platform so powerful. And on the right, we have our actions and insights. This is where we move from detection to remediation.09:29We can immediately review risk profiles, create new security policies to block high risk tools or even rerun a scan to verify our fixes. This is why I'm such an advocate for this tool, especially when talking to our customers. It doesn't just show them the problem, it gives them the direct path to secure their AI deployments.09:49It's a huge win for any team looking to move fast without sacrificing safety. Everything we need from the server URL to the scan status is right here. It's a clean, centralized view that keeps the entire team aligned. It's been a great success seeing how this simplifies the security life cycle.10:09Now let's take a look at view deep dive. This is a fantastic example of proactive defense. Now that we're inside, we can start investigating these specific risks. Let's take a closer look at what the scanner picked up by clicking on view deep dive.10:30This is where the real magic happens. We've moved beyond the high level summary to see exactly what's driving those 19 high risk findings. At a glance, we can see that all 19 of our high risk findings are tied directly to tools. The platform has mapped these findings against the Cisco AI security and safety framework.10:48Which gives us a standardized way to understand the impact of these risks. Down here in the findings evidence section, we get the granular detail we need. You can see tools like export_data and get_available_symbols flagged as high risk.11:08Seeing the specific tool and the associated risk right here is incredible. This level of transparency is a major win for our teams by why showing them exactly where their vulnerabilities lie. And from this view, we can jump straight from identifying these tools/risks.11:27To creating policies that protect them. It's a seamless workflow that has really streamlined how AI defense handles MCP scans, and there you have it. In just a few minutes, we've gone from registering an unknown server to having a complete map of our AI attack surface.11:45We now have the deep visibility we need to build a rock solid security strategy. I hope this walkthrough showed you just how powerful this discovery process is. We're really setting the bar for what AI visibility looks like and we've laid the perfect foundation for our defense.12:02Thanks for sticking with me through this deep dive into scanning. Stay tuned for the next part where we put this intelligence into action.12:19
