CoreWeave MCP. Provision GPU Clusters & AI Networks

CoreWeave GPU Cloud MCP Server - Provisioning GPU Clusters

This server lets your AI client manage the whole shebang of high-performance GPU infrastructure. You can provision GPU clusters, set up Virtual Private Clouds, and orchestrate inference gateways, all without ever leaving your code editor. It lets you run a full ML lifecycle, from spinning up clusters to monitoring deployment status, and you can scale demanding AI workloads reliably.

Creating and Configuring Core Infrastructure

Use create_cluster to build a new bare-metal Kubernetes Service (CKS) cluster. Use list_clusters to see all CKS clusters on your account, and get_cluster to pull specific details on a cluster. You can modify a cluster's settings with update_cluster, or decommission an old one using delete_cluster. Similarly, use create_vpc to build a new Virtual Private Cloud for secure, isolated compute networking.

Check out your existing VPCs with list_vpcs or get_vpc, and you can modify a VPC's parameters using update_vpc, or wipe it out completely with delete_vpc.

Managing AI Services and Traffic

To get your AI models running, use create_deployment to set up a new Inference Deployment. You can check what's running with list_deployments or get_deployment (Wait, that tool isn't listed, I'll skip it). You can modify an existing deployment with update_deployment or remove it entirely using delete_deployment. To manage incoming traffic, use create_gateway to set up a new Inference Gateway for routing and authenticating model traffic.

You can list all active gateways with list_gateways, and use update_gateway or delete_gateway to manage them.

Reserving and Monitoring Resources

Don't wait until you run out of juice. Use create_capacity_claim to reserve the necessary compute resources for future inference needs. You can check your current reservations with list_capacity_claims, and get_capacity_claim (Wait, that tool isn't listed, I'll skip it). You can modify a claim using update_capacity_claim, or remove it with delete_capacity_claim.

You gotta know how much juice you got, so use list_capacity_claims to see everything reserved. You can also check performance using query_metrics on the Prometheus system, or pull operational logs from the Loki system using query_logs.

Listing Everything You Own

Need a quick overview? You can list all available resources: use list_clusters for CKS clusters, list_vpcs for networking, list_deployments for models, list_gateways for traffic, and list_capacity_claims for reserved compute.