Shunwang Tech MCP. Manage GPU Clusters & Edge Resources via Chat.

Listen up. This server gives your agent total command over massive edge computing setups—the kind of stuff running in PC Bangs or distributed GPU clusters. You're managing infrastructure, not just sending emails. It lets your AI client monitor complex hardware networks and schedule tasks across nodes without you ever having to touch a dashboard interface.

Your agent handles the grunt work, letting you focus on what matters.

When you need to know exactly what compute resources are available, your agent can first run list_clusters to get a rundown of every computing cluster in your infrastructure. From there, it uses list_nodes to pull a comprehensive list of every connected node, so you'll always know the scope of the network.

If y'all need to know what kind of OS or environment is ready for deployment, list_images shows exactly which system images are available for your tasks. For hardware specifics—like figuring out if a job needs 24GB VRAM or something else—you gotta use get_gpu_specs. This returns the detailed specs on all the GPUs connected to your network so you can match jobs to proper hardware.

To see what's running and how much juice it’s using, your agent first checks in with list_tasks for a history of everything that’s happened across the whole cluster. For current health checks, get_resource_usage spits out a summary report detailing the CPU, GPU, and RAM utilization rates across all defined resources.

If you need to dig into one specific machine, get_node lets you retrieve the detailed status of a single computing node, whether you know its ID or its name. That's how you keep tabs on things.

When it comes time for action—and there’s always time for action—your agent can schedule compute tasks by calling create_task. This deploys a new job onto the cluster and kicks off execution immediately. If that task goes sideways or you just need to test something else, you'll use stop_task to terminate a running process using its unique ID.

You control the entire lifecycle from deployment right through to termination.

Beyond scheduling, your agent manages the physical state of the machines. To give a node a direct command—like initiating a reboot or forcing a shutdown—it uses send_node_command. This sends that specific management instruction straight to an identified compute node. And remember, if you need to know why something went down and want to check the status again later, running get_node gives you all the current details on that machine's operational state.

Basically, your agent handles inventory checks—listing clusters and nodes; it monitors resources by getting utilization reports and node specs; and it manages the whole job lifecycle from deployment via create_task to termination with stop_task. It’s built for infrastructure engineers running massive, distributed networks. You use this so you don't gotta stare at dashboards all day long.