Transcript
Oh Thanks for joining in and welcome to expert insight series and there's not a series, there's a single session, however, so.00:00We today we are going to present about unlocking the power of NSO six.seven.00:20And I'm your host for the day today and my name is and so I'm excited to guide you through this session for an so six.seven which can transform your network automation capabilities with some.00:25Practical insights of AI MCP and and kind of a great knowledge of whole bundled network automation part. And, yeah, so now let's get started. So today I will take you to the sessions best practices. I have around 15 years.00:41Of experience in telecom industry and specializing in various technologies such as IP mobility and virtualization.01:01Currently my focus is on network automation, assurance and mostly AI, and I own multiple industry recognized certificates. With me, I have my colleagues and panelists today.01:10Yeah, hero san, and Patrick with me, thanks Patrick and Hero for joining in. And before I start the presentation, let me cover few logistical items. So during today's session, you will have an opportunity to ask some questions you might have.01:25About NSO six.seven are pertaining to that.01:45Just type in your questions into the Q and A panel. My colleagues and my panelists would be happy to help you in that. And in case you have a question, you can raise your hand also. We'll try to respond. You can lower the hand when your question has been answered.01:49So for any unanswered questions, we will get back to you via email or any following presentations. If you're having any audio issues, use the chat panel and one of our panelists will assist you. And with that, let's begin our.02:06Presentations for today. So here is the roadmap for our discussion today. We have broken this sessions into three major topics and 4th is the demo.02:241st, in the 1st topic, we will introduce Cisco NSO six.seven reviewing the major platform updates, its architectural shifts and its system requirement with its long lived release. Second is the model context protocol.02:41Quick recap we'll be going through establishing a clear understanding of what actually MCP is and why anthropic created it and how it revolutionizes the.03:01AI integration. 3rd will dive into the NSO adaptive server, MCP server, detailing on how NSO exposes its schema and capabilities to AI hosts while preserving your core.03:12And finally, we'll then move into the demo where you can see the AI assistants interacting directly with the NSO. Okay, so we are all set for the 1st topic that is.03:27Introducing NSO six.seven. We'll cover what makes this release unique and the performance and enhancements under the hood and critical upgrades or prerequisites which you need to plan for.03:43So NSO six.seven is a long lived release, making it an ideal target for a long term production deployments as the name says long lived, right? So long lived releases are basically recommended for deployments that require stability and.04:01Longer maintenance windows. And as they provide extended support, the security and security updates which are com compared to short lived releases and the support for this release lasts to about 42 months after the end of the life announcement.04:19So going on to the highlights of six.seven, the 1st major highlight which I would like to highlight is that add up to MCP server. So this is the flagship flagship feature of this release. It enables you with native schema driven and interaction between NSO and external AI.04:38Distance via the MCP protocol that is the model context protocol that will see how it has worked upon in the demo video later in the end.04:57Next is the optimistic upgrade. So introduced in this optimistic upgrade, we introduced incrementally in six.two.two and then fully it got matured in that in our six.seven. So this minimizes your operational downtime by allowing a read active read.05:08Then write transactions basically to continue while packages are being.05:28Reloaded. So if I can say that in the meantime means while packages getting reload usually in earlier releases before six.two, all the transactions were stopped. So using this optimistic upgrade features, what we did is while the packages are.05:32Reloaded or packages if you are working on any of the packages also.05:52So your transactions basically are not affected. Next is the XML compliance templates. So you can now use the standalone XML files for compliance auditing, providing self contained flexible compliance verification.05:55Without any heavy overhead. So in nines in in NSO six.seven, then this new feature XML compliance templates allows you to use standalone XML files for compliance auditing. This means that you have self contained xmls files that defines your compliance rules and checks.06:15But making it very easier and more flexible to verify device configurations specific and without adding any heavy processing overhead. So if I can say instead of embedding compliance hold inside a complex script or a code, you write them into the next files.06:35So these XML files describes what configuration should or should not be present on the devices. So for an example, if I can say imagine you want to ensure that all your routers have a secure NTVR.06:54Configuration which you need to configure it on most of the routers or all of the routers. So you create an XML compliance template that specifies the required NTP settings such as enabling the authentication keys or restricting access to with an Excel access list or something like that.07:10And then NSO will use the XML file to audit your devices and report if any device is missing these NTP settings. So if the device is not compliant, you can make, you can then remediate using the NSO automation capabilities.07:28Okay, so yeah, next is the web UI redesign. So we have refreshed this web interface with the redesigned transaction views and a streamlined configuration editor for an improved.07:45Operator experience basically. The next one is on demand V one becomes the default in NSO six.seven on demand V one. So in NSO six.seven, the CDB persistence mode. CDB is your configuration database.08:01Of NSO where all your data and the transactions are saved basically and has changed to on demand V one replacing your older in memory V one mode, which is now deprecated.08:20So, so CDB is where your NSO stores information previously, we were using the in memory mode that kept all the data in the ramp, which would consume most of the memory in a large scale deployment.08:31So with the change basically what we did is in NSO six.seven, we came up with an on demand V one board which loads and keeps the data in memory only and when it is only needed, so it's an on demand and offloads your less used data to the disk.08:47So this process basically this approach significantly reduces very high memory footprint making NSO more efficient and scalable for managing many devices or a large scale configuration.09:04The last thing is the system requirements. So, the critical prerequisite for NSO six.seven has been changed to a minimum configuration of Java to 21. So all previous Java versions used to all previous Java including 1117 basically are not.09:20Supported, so for NSO six.seven, we have to use Java 21. Okay.09:41Yeah, so, yeah. So though in the last in the last slides basically we covered all of our NSO six.seven features now basically we'll do do a quick recap on MCP protocol, so what does it do and.09:48And and how does it function? So what exactly is the MCP? So model context protocol basically is an open standard protocol which is was introduced by anthropic in November 2024.10:06So it standardizes how your LLM is powered, our applications, how applications get connect to an external data sources, enterprise tools, and context providers. So under the hood basically MCP uses JSON RPC two.zero protocol.10:22So you can think MCP as a USB C type of connector for all the AI applications. Every peripheral has a proprietary cable, right? So earlier we had printers which had one cable cameras, which had another external drive, which had another type of cable. So.10:41Right now also I have many old cables which is just lying out there. So, so yeah, so this is very good kind of an innovation in where we can use a single kind of an application or or a protocol basically to connect out the similar devices or the applications.10:58And MCB does the same thing for AI, so basically instead of building a custom or integration between every AI model, what they did is every, they came up with an MCB compatible AI client and which can immediately plug in and play with this protocol.11:17Work and interact with your platform.11:37So let's define what an MCP server is. So an MCP server is simply an application or a service that exposes specific capabilities such as data files.11:40Or executable functions over the MCP standard. So it communicates using as you know, JSON RPC two.zero and it's a defined transport layer. And MCP servers can be stateless or stateful. It can be either local or say remote.11:58And most importantly MCV server is completely model agnostic, that is the server doesn't know or care whether an AI querying is a cloud or say a streamlit or any GBT forum or in any any kind of a model. So it simply presents capabilities in a standardized format.12:17So here's a very good thing that is MCP server over here is our NSO and if you are using the clients such as cloud GPT, that that would be the clients, ok? So how do MCP clients and servers actually pass your messages back and forth?12:36So this specification defines two primary transports that is 1st is the SDIO, stdio, that is standard input output. Here the host application spawns the MCP server as a local sub process passing JSON RPC messages across STD in a NESTED I out.12:56So this is a common local developer tool and a command line agents which basically refer to the next, ok? So second is the streamable HTTP HTTP. So this is the current standard for remote network accessible MCP servers that is form which was formalized in.13:162025 and it enables a robust scalable communication over standard web infrastructure so the MCP specification establishes three distinct architectural roles if you can see out there. So for.13:36Next is the LLM, next is the host with which the MCP client and 3rd is the MCP server. So LLM, so what is LLM? It's the reasoning engine that processes your natural language and decides it.13:56When it needs more data or needs to take an action. So the decision factor with lies with the then the host application that is called as an MCP client which is in the middle, it is a container application, a small application such as an ID or a desktop assistant or a Python.14:12Script like streamlate or something like that cloud code or cursor that manages the conversation at all the the on which you give your prompts, right? Ambits the MCP client and orchestrates the calls.14:31Next is the MCP server. So this back, this is a back end service providing with you with all the tools, resources and templates. And the critical part of this is that the LML at the LLM itself never speaks to MCP server directly.14:47The host application acts as a coordinator and it discovers what MCP server offers to it, and then it presents those capabilities to the LLM, and when the LLM requests a tool called the host invokes it up across the MCP connection. So it's like that.15:04MCP capabilities are structured around three fundamental perimeters. 1st is the tools, next is the resources and the prompts. What are tools basically, those are executable functions that allows to.15:30To model or to take action on the.15:44Dynamic data. So e.g. running a device sync check on NSO or querying an interface status or executing a dry run commit. So these are all types of tools. So what kind of action you are thinking to do? So it comes from the tools.15:48Next is the resources. Resources is the addressable readable context object, thinking like as a file or database record schema or definition or any API response that provides static or semi static background data. So it, so how resources are the data basically on which the tools will work on.16:06The next is the prompts. So predefined parameters prompt templates and workflows similar to structured or commands like that. So it helps guide users through complex operational tasks consistently.16:27So usually we give prompts on chat GBD, right? So those are basically a reusable template prompts basically to now to appreciate the value of MCB.16:42Look what happens means we will just study how we if you if you do it without MCP server, how, how it would look like. So in in a classic LLM loop basically you have a direct exchange between the user and the model. So the model relies entirely on the what.16:57It's in the training data set, so LLM, it will have its own in inbuilt training data set and whatever context it usually makes, it spaces you it spaces into the chat window. So this model has a major limitation in our enterprise networking or any kind of networking feature across the.17:16LLM has zero knowledge of your live network state. So that is the greatest thing the MCP here is solving is that we as an LLM, it doesn't have any kind of a live network state knowledge. So however, it has to query somewhere to the another on about your live network status.17:36So it cannot access your CDB configurations, right? So it cannot also execute real world operations or by or it or making a dryer run or something like that. So it's very difficult without an MCP model. So while it's great for general questions like.17:55Just you have an Excel sheet and and organize data into it. You put in your prompts, it will give it. So if it it's not like that, right? So we require it as a dynamic for dynamic data. So that is why.18:13Without MCP, it would be really a NO go for us. Now let's see how MCP changes this dynamics. So in the step one basically the discovery process.18:29And when a session starts, the host application or the MCP client connects to the MCP server and query this capabilities. So it will ask that, ok, what tools resources and prompts do you have available?18:43It will ask the and the MCP client will ask the MCP server what all tools you have, please give that information. Then the server will reply with a structured list of capabilities and the parameters. So it will, in the next step, it will reply with all the tools and the schema, whatever it has.18:59Then the host injects this definition into the system prompt of the LLM. Now before the user even types the command, the LLM is now fully aware of every network tool and data resources, it has at its disposal. So.19:19So that is how basically LLM gets all the information from the MCP client. The step two is the invocation. So when the user asks the network related question.19:36The LLM realizes that it needs a real time data. So e.g., if an operator asks, check if the device core router is in sync or not. The LLM formulates a structured tool called. It will do, ok, what I need to do is check sync. The device will, isn't and the device name is core router zero one.19:53The example. So it will send this request back to the host application and the host executes the tool called across the MCP server and receives a live output from the NSO. And it then passes that results back to the LLM and LLM interprets then the output and provides a clear net.20:11Natural language application explanation to the engineer.20:31The final answer. I hope it makes sense here. So, and a vitals, what, what are the primitives and who can control basically these are just a summary slide for MCP that is tools are controlled by the model, the LLM autonom autonomous.20:35Mostly decides when a tool should be invoked.20:55Based on the context of the user request, then the resources are the controlled by the application host. The host application chooses when to load and inject those resources into a conversation context, and prompts are the controlled by the user ourselves and the human engineer selects a pre built prompt template or.20:58Workflow to trigger a structured process and this clear division of control ensures that AI behaviors remain structured, predictable and aligned with our operational rules. Why should our organization consider MCP or adopt MCP?21:19This is the question asked usually. So, 1st thing is that it is standardized, right? So you write your MCP server integration once and you can immediately connect it to MCP compliant client or any of the hosts.21:40Composable. It is a single host application can connect to multiple MCP servers simultaneously. So when I have one MCP server then NSO, you might have a different application which has a different MCP server. So you can connect those MCP servers simultaneously with one.21:55Composable host application. So it's some kind of a dashboard, a simple dashboard or prompt client or a client. So AI assistant can connect to any of your MCP server, either it can be a Jira MCP NSO MCP or a Splunk MCP whatever for telemetry. So all these MCPs.22:13Servers can be.22:32Multiple MCP servers can connect to one of the host. 2nd is the separation of concerns. It is the LLM handles natural language reasoning and the MCP server provides the data and execution boundaries and the host manages the.22:34Application states. And for auditable, that is it's always auditable, why? Because every single interaction is a structured process and inspectable via JSON RPC call. So that is the more advantages why it is because it is auditable.22:53And which eliminates your opec or ad hoc integrations? So now what MCP you think is not actually.23:11So it is to set an expectation properly, we should be also clear about what MCP cannot do or it's not supposed to do. The MCP is not an LLM or an LLM API replacement or something like that. So you still need an interface provider like opening.23:28Yeah or an internal enterprise model.23:48To connect to the host. So it is not a security sandbox also. MCB does not replace your network authentication, authorization or role based access controls, any kind of. So it's not also tied to any of the vendor, it's completely open source and model neutral.23:52And 3rd important is that it's not magic, so we usually believe that when it gives answers, I also think that it's magic, but it is not a magic. The host application and the the line systems still do the heavy lifting of the orchestration and security enforcement.24:09Okay, now let's look at how Cisco has implemented the standard in its NSO adaptive MCP server capabilities. So, now let's.24:30Make it clear boundary about MCP NSO MCP server, which it can do and which it cannot do, ok? So what it can do is it is a native NSO package. It's a simple package, we need to run it on the NSO, which bridges the MCP protocols directly into the NSO features. It connects.24:48Both these things. It strictly inherits and enforces all existing NSO authentication such ASA and NSCM policies. Also, it exposes the secure MCP endpoint directly into the NSO, meaning you don't need to do or you don't need to deploy or manage an external bridge service.25:08Or something like that. Now, what it cannot do, it is not an AI assistant by itself. It does not generate any text or reason on its own. So, so if you assume that if you just bring in a package of NSO and NSO MCP server, it will not do anything.25:27So we need some brain besides that, right? And a host besides that. Cisco does not bundle 3rd party AI models inside Cisco NSO, so you need to choose and connect your own preferred enterprise LLM client, ok?25:46So, to be, to be just think that we use as an extremely as an NSO client over here as an NSO host client. You can use either cursor or cloud code or any of the LLM client which are.26:01Which the organization has been approved to use. So it cannot force an external AI to follow operational steps correctly. So there is NO control over AI of following some correct steps or something. It's not predefined as such.26:17Okay. Architecturally, the NSO MCP server is simply another northbound interface. So it sits just alongside your existing interfaces like CLI, web UI or netconf, resconf, whatever the interfaces we have.26:34Just like those interfaces MCP leverages the exact same core NSO engine underneath it and the configurations data with the CDB and the network elementar such as nets, triple AAA authentication, then the same access rules or transaction engines.26:55All this basically commit queues, rollback logs, everything is considered under it. Now the key consideration out here is because AI clients can generate configurations rapidly based on natural language prompts having transaction level rollbacks and strict NSM will.27:14NSCM policies, it is more critical than ever to guard against its unintended changes. To run the NSO adaptive MCP server, you need to ensure your environment meets with the following baseline.27:33Requirements, so NSO version 1st of all, it should have six.seven or later version. As we discussed earlier on the 1st slide itself Java runtime should have a Java 21 or later as the MCP package logic is built entirely on that modern java.27:53Run time then MCB protocol specification, it supports the standard 2025 spec release in March through the latest revisions in 2026 which we had.28:10So ensure your build environment CICD pipelines, runtime and VM environments are all upgraded to Java 21 before deploying your NSO six.seven. So this is an hybrid architecture within with an proxy layer in between and.28:27A Java MCP backend in a two year design out here. So the external endpoint is exposed through NSO web UI and HTTPS licenser, so on the top we have an MCP client or AI assistant which you have. It.28:47Goes to the NSO web UI HTTP listeners, external MCP endpoint client which is over here. Next is the Earlyn proxy layer. The only proxy layer handles all the authentication side of the request and processing and forwarded inward basically.29:05It then goes to the Java application component, which implements the MCP server logic. So all your logic is at the Java application component, and it discovers all the capabilities from the running NSO schema and packages.29:21So the main element out here is your Java application component which is which holds all your MCP logic. Now that this Java backend is never exposed to the outside world.29:37So this two layer design is the reason the logs are splitted across two places. So in a link proxy layer and at the Java application. So if you can see MCP server log and java VM log, so due to this basically one is the logic and one.29:54Yep a link proxy basically authent for the authentication and all the stuffs. So this two layer approach, so that's why we have to basically for any kind of troubleshooting or internal verification, these two logs which are mentioned we have to analyze those.30:13So here is what an end to an operational setup will look like. So, 1st thing is the MCP server. So in the on the on the right hand side, the right side over here in the cisco's MCP server, we have tools resources, and prompts.30:37So Cisco NSO six.seven exposes all your models, service definitions, then operational actions, all as a tool and its resources. Then the next part comes over here is the host and the MCP client.30:57So the host and MCP client is the one in the middle, the stream lit bit for it consider the streamlit UI and the MCP client. It's a lightweight custom host application featured in clean streamlit web UI.31:12And the, and the 3rd part is about the Cisco MCP server is the Cisco LLM. It's the reasoning engine, the brain of all this. So in our enterprise environment, this connects to our circuit. So the circuit is Cisco secure internal LLM, which ensures all the network L telemetry.31:28And data remain completely private and protected within our corporate boundaries.31:48Yeah. So let's trace a complete end to end request here. So example, you can say that an engineer.31:55Types into a streamblet UI, show me all the interfaces on the edge router with all the input errors. This then the streamlid host sends a prompt and the tool descriptions to the LLM. The LLM recognizes its need of live data and returns the tool called request for.32:08Getting the device interface matrix.32:29Then the host invokes the tool called our HTTP post to the NSO the MCP endpoint. Then NSO validates the user session via the NSCM policies and queries the device through the NET.32:33Retrieves all the operational metrics from the CDB and returns the structured JSON RPC output. Then the host passes the response back to the LLM, which formats the raw metric into a clean and easy to understand summary for the engineering.32:49Okay. Now let's summarize the key benefits of adopting the NSO adaptive MCP server. So it's in AI ready without any core disruptions. You don't have to re architect your NSO deployment or rewrite the services to take advantage of the AI.33:08That is the number one benefit. Second is that live state visibility, your AI assistance move beyond static knowledge basis to interact with the live real world network state and validate the operational health. Thirdly it is thin and maintainable hosts, so your UI.33:26And client applications remain thin wrapped, remain thin wrappers focus solely on the UX and orchestration while NSO handles all your complex data models. And it's a future proof of standard, so by standardizing or using MCP, you can easily swap the underlying LLM upgrade.33:46The models or connection MCB servers as your AI ecosystem evolves. So now let's go for the demo. Hello and welcome to demo one in this demo, we'll be going through the four step procedure.34:071st we'll install NSO six.seven using custom ansible playbook and then install the MCP server package on the same server and by and 3rd step is enabling the MCP server on the NSO. And 4th step is installing the custom MCP client, which will communicate to the NSO enterpris.34:26Devices via the prompts.34:46So now let's begin. So in this, 1st we'll go to the package, the NSO package which we downloaded. Inside this NSO package, we have a nSO.yaml file which we have created.34:48So we will run this YAML package using Ansible playbook. So you have to note that this is basically a custom made, and this will install NSO six.seven. So the steps have been minimized due to the time constraint of installation.35:07We'll go to now the CLI and verify whether the devices have been attached or not. Yeah, so we can see the devices are attached to now, which you can note that the MCB commands are not working, so for that we have to install the packages. So there is NO package.35:23So we are now installing 2nd step that is installing the MCP package. So in this MCP package, resides in the work sub directory of the NSO folder. So we have to transfer basically, the NSO.35:43Package or the MCB package to the running NSO package out here to basically activate the package inside the NSO. So we are right now copying the NSO NSO adaptive MCP server package into the NSO run.36:03Packages, and then after copying it, we'll basically go to the CLI prompt and perform package, status and then we'll reload the package.36:21So after reloading the package, it, it would take time, but it is, as you can see, it is minimized. So it is completed successfully, if you can see out here the MCP package has been installed and it's up and running. Now we'll go to the MCP server. So if you can see MCP server.36:38Your configuration now has been enabled.36:57So what we'll do is we'll enable the MCP server using the command MCP server enabled command, we'll commit that configuration and see whether what's the MCP server configuration details. So if you can see out there MCP server is enabled and all the policies which are according to the MCP are in the.36:59Disabled state, so all the policies are in the restricted state as of now. So we'll go to again to that and we'll enter it. So basically we need to now test whether all my tools of the MCP are active or not.37:20So in this under under the, under the tar or the zip folder, we have this test.shell script, test shell script which tests all the things about the MCP server and the MCP tools which we have. So until and unless it all are passed.37:37We cannot move ahead. So right now if you can see out here all the tests has been failed. So this is the usual process because MCP takes time and also it requires a little a kind of a second reload, a package reload after it is reloaded.37:58So this is a caveat out here. So you can just basically rerun the package or reload the package out here while going onto the NCS client and then see the package details and observe whether.38:16The packages have been under operational state or not. So now we'll again go to the, the where the test script is and we'll rerun the test again.38:36So if you can see out here we out of which four has passed, we'll try to run it again and see, ok, now it is five passed. So all has been passed. So over all our three of the steps is good. So now we'll run the MCP client out here.38:55So we have clients such as called as a stream let and a backend called as a flask. So we'll open a new terminal window, we'll run the backend script of flask, and the front end script of stream debt. So in case you are using it.39:11Where means you can use a different scripts also or different MCP client according to your need. We have cursor or say claudicode. You can use that, but we have designed a kind of an MCP client internally for this and it is called the streamed. So we will kind of enable that.39:28Out here, so this is the step. We'll go to the web UI part under the NSO, so web UI subfolder. In this we will activate using the Python script and let this window be open.39:49Similarly, the, the back end, once the back end is up, we will, we are kind of enabling the stream lead run, so stream lead run front end is a front end process, so once that is enabled, you can see automatically now.40:07Both the things is up, you open the web browser at the local host 8501, you'll be able to see this network assistance or the MCP client. So that's the end of demo one. Now let's proceed to demo two. So in.40:23Demo two, we will have four steps again, check the MCP client command output with some restricted tools, then allow MCP client for live status tools, install BGP MGR package for BGP configuration and deleting the configuration from backend and adding new BGP neighbor wires.40:41MCP client.41:00So if you can see out here, we have this MCP client running the flask at the backend and available MCP tools out here. So these are the available list of tools on the left hand side which you can see out here. These are the very short MCP tools or you can say very less.41:02Amount of tools which you are able to see. So now we'll start with one of the prompts that is show how many devices are attached to this NSO. So it will get the output. So there are two XR devices which are attached, so it will also give the information regarding which tools or MCB tools it has used to.41:21Fetch this information from. So it is used read configuration read configuration tools that has been using, then show command output of BGP summary.41:40So we'll see what doesn't get the reply. So we'll see show for show BGP summary of the reply we get. So it doesn't have any kind of information regarding that or NO tools is being used for BGP configuration out here.41:55So due to which what we'll do is we will see the details. So right now the MCP server policies default action configured is restricted if you can see up here. So now what we'll do is we'll permit the default action that is permit all the tools which we are using under the.42:11Or or all the live status tools for getting the live status of the devices and the live status of the configuration will do an MCP server. We we change the configuration of the MCP server policies to the permit now. So let's run the same show being.42:30GP summary command again or.42:50The prompt for that. So here we get the prompt. So if you can see it, it displays the whole output out there. So it uses the live status MCP tool out there. So currently left hand side, you have to refresh the tools to see the long list of.42:52Rules.43:13Which are now permitted to use. So earlier it was under the restricted mode. Now for the only read only things and this is now the live status, so it is now permitted. Now for doing the BGP configuration, we need to install the BGP MGR package out. So for installing.43:13The BGPMDR packet, we have just we have created this internally custom made package, so we'll download it from the GitHub under the NCS run packages folder.43:33Once that is downloaded, we will basically make clean and make the package and then we'll go on to the, the NCS CLI.43:43Because we directly went to the NCS package folder now. So directly the GitHub has already got all the information from the NCS CLI and the package from GitHub. So we just need to go into the CLI now and zero the packages out there.43:59To basically, to load the package of BGPMGR into the NSO. So if you can see out there, four, four packages are already informs already loaded out here. We'll do package reload.44:15And see if you can observe that BGPMGR package is also now running and active in the operating status, the 1st package. So now we'll switch back again to the NSO MCP client. We'll refresh the tools.44:30After refreshing, you will see the BGP MGR MCP tools available out there.44:51So create, update, delete, all those things would be possible for the BGP configuration. Now next what we'll do is we'll delete the device configuration or NO router BGP will perform on only both the devices XR one and XR two from the backend. So we'll go to the NSO CLI.44:57Devices XR one config, NO router BGP and device XR two under the config board NO router BGP. We have done both the route, NO router BGP and then the commit from both the devices. Now we'll get back to the NSO client.45:16Now for configuring the BGP parameters, there are some parameters which we need to give. So what parameter input we need to give that information is shown over here. So now we'll see the BGP output or the BGP summary output, whether.45:36Is there any BGP configuration present on both the device or not? So there is NO kind of BGP configuration present out there if you can see.45:52So now we'll input the prompt using the parameters which we saw earlier that the BGP configuration for BGP config on XR one and XR two BGP name, what the AS number is on the BGP one and BGP two.46:05And it's in the word devices we are basically using for the configuration. So just by providing that information, your BGP configuration has been done using the BGP MGR MCP tool, which is.46:24Activate the package into the NSO. So this is how basically these things becomes easy using the MCV client. So if you can see out there, we are also doing show BGP summary on both XR one and XR two devices.46:41And to verify that the configuration has been done correctly. So, yeah. So BGP neighbors has been configured perfectly. Thank you for watching. The demo two is ended. Thank you so much.46:58Okay, so coming back from the demo, thank you so much for watching the demo, next is that as we close for today's webinar.47:14You'll see a brief survey pop up from Cisco customer experience. So it takes less than 2 min to fill out and your direct feedback helps us to tailor our upcoming expert inside series to cover all your topics you care about, ok? So on behalf of entire Cisco CX team.47:30And our panelists, Hero and Patrick, thank you for your time, our engagement and our continued partnership, wish you have a wonderful rest of your day. Thank you. Thank you everyone.47:50
About this Video
Author : Cisco Customer Success
Uploaded : September 30, 2026
Explore Cisco Network Services Orchestrator (NSO) 6.7 to see what its Long Lived Release delivers: the Cisco NSO Adaptive MCP Server that exposes schema-driven tools, resources and prompts to AI hosts over the Model Context Protocol, Optimistic Upgrade to reduce downtime during package reloads, default CDB On-demand V1 for lower memory use, XML Compliance Templates for lightweight audits, a refreshed Web UI, and the Java 21 prerequisite. The recording demonstrates installing NSO 6.7, enabling and securing the MCP Server, validating tools, and operating with an MCP client to fetch live device state, plus deploying and verifying BGP using the BGP MGR package. It also explains the two-layer MCP architecture, access control via NSCM, logging, and how NSO preserves transactions, commit queues and rollbacks while AI-assisted workflows query and act on real network data.
Internal error
No related resources available.