Transcript
This demo will cover how you can use the natural language capabilities of a Model Context Protocol or MCP server to interact with your storage, automate workflows, and simplify day-to-day operations. The Data Services Cloud Console or DSCC provides an open API for automated storage management and analytics. MCP builds on that by enabling natural language interaction, so you can query, automate, and manage your storage using simple conversational commands instead of complex API calls, while still leveraging the same secure authentication and authorization used by the underlying API. I've set up an MCP server with the commands shown at the top left of the screen. Earlier, I asked it to show me all the storage systems in this DSCC workspace, and it returned a chart with each system and its key information. Now I'll restart the server and start a new session with some more advanced queries. When I ask whether any systems are disconnected from DSCC, the server begins working through the request and asks me to authorize the API calls. It needs to iterate through a couple of approaches, but it quickly converges on the right answer, identifying that four of the systems are offline. Next, let's identify the best system for provisioning a workload. One important factor is available capacity. And when I ask which systems are running low, the good news is that none are critically low. The lowest free capacity is still 38.8%. When I ask it to explore capacity in more detail, it returns a comprehensive table with logical and physical capacity, reducible versus non-reducible data, and the overall data reduction ratios. Now, I'll generate a chart that includes capacity and performance headroom, a unique HPE capability in MCP that goes beyond IOPS and latency. It incorporates workload type and distribution, system hardware configuration, and volume characteristics to determine whether additional workloads can be added without impacting performance. Based on the results, I'll select a system with low headroom utilization and sufficient capacity and ask it to create two volumes for use with VM Essentials. I forgot to specify the size, but fortunately, the model's smart enough to prompt me for that detail. And once I provide it, it's able to complete the task. Now I need to know which hosts can access that array. The model returns a chart with a list of hosts and their personas so I can easily identify the VM Essentials hosts. I'll now ask it to present the new volumes to one of those hosts, and it quickly confirms that the task is complete. Finally, let's replicate this volume set to another array. When I check whether this system has any replication partners configured, we see that there are partnerships with two other systems. So I'll ask it to set up synchronous replication to one of them. Soon enough, however, the model returns to tell me that sync replication can't be added because async replication is already in use on this target. It suggests using async instead, noting that the minimum RPO would be 30 seconds. When I tell it to go ahead, it successfully creates the protection policy and starts replication. One thing to keep in mind is that while I've been using Visual Studio Code to interact with the MCP server, any compatible AI client can do the same. Here's an example of a dashboard a colleague quickly built with a different tool, but using similar natural language commands. In summary, with the combination of DSCC's strong API foundation and the MCP server, tasks that would normally require multiple steps across individual arrays can now be handled with a single natural language query, making it much easier to get the information you need and take action.
