RunPod MCP. Manage your entire GPU compute lifecycle via chat.

This server connects your AI client straight into RunPod’s cloud infrastructure. You can manage GPU pods—creating new instances, checking existing ones, and shutting down compute cycles—all through natural language commands.

create_pod lets you spin up a brand-new GPU pod by specifying the name, the GPU type you need, and the Docker image it'll run off of. list_pods gives you an inventory list showing every single GPU pod currently running in your account. If you need to check specific details about one pod, get_pod fetches all that info using its unique ID.

To plan out what you gotta build, list_gpu_types retrieves a clean rundown of every available GPU hardware type you can actually deploy. For templates, list_templates shows you the saved and pre-configured setups you can use when starting new pods. You'll also see all running endpoints by calling list_endpoints, which lists every serverless endpoint set up to handle containerized inference applications.

When you’re done with a pod, stop_pod shuts it down instantly, halting billing on those compute cycles and saving your cash. The whole system lets your agent act like a full-on MLOps engineer: managing complex computational workloads, spinning up new hardware instances, and auditing serverless endpoints—all through chat.

Here’s the rundown of what you can do with these tools:

Provisioning Compute: You gotta build something? Use create_pod to spin up a dedicated computing instance (a pod). Just tell it the name, the GPU type required, and which Docker image you want. If you're planning ahead, checking available hardware types with list_gpu_types or reviewing saved configurations via list_templates helps keep your build process smooth.

Checking Status & Inventory: You can get a complete picture of what’s going on by running list_pods, which shows every active or paused pod registered to the account. Need details on just one? Use get_pod with its unique ID to pull all specific info. You can also check out all configured endpoints using list_endpoints, seeing exactly where your containerized inference applications are running.

Cost Management: When the job is done, you gotta kill it fast. Running stop_pod immediately halts a running GPU pod, which means billing for compute cycles stops right there. This keeps your costs under control.

The Lifecycle: The tools manage the whole lifecycle of high-power computational resources. You use create_pod to start it up, check its status with get_pod, and when you’re done, you hit stop_pod. If you want to see what hardware is available for future builds, run list_gpu_types. To keep tabs on everything, list_pods gives the full inventory.

You can also review pre-set options using list_templates, or check out all your serverless endpoints with list_endpoints.

You're not just reading data here; you're executing changes on a live cloud account. The agent uses create_pod to build nodes, using specified GPU types and Docker images. It checks the status of existing instances with get_pod via a unique ID. To see all resources, it runs list_pods. When you're done with compute cycles, stop_pod shuts everything down to save money.

Planning requires checking available hardware types using list_gpu_types or reviewing saved setups through list_templates. The server also audits your infrastructure by listing all configured endpoints via list_endpoints.