# RunPod MCP for AI Agents AI Agent Connect

> RunPod MCP lets you manage your cloud GPU infrastructure directly through your AI client. Provision scalable pods, monitor active instances, and audit serverless endpoints without leaving your chat interface. It turns your agent into a cloud infrastructure specialist for machine learning workloads.

## Overview
- **Category:** superpower
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_mqfmppnjNgzd7DnF3DKymtiD5h0tXpOI6WsCj5ji/ai-agent-connect
- **Tags:** gpu-computing, serverless-deployment, cloud-instances, machine-learning-ops, container-orchestration, infrastructure-as-code

## Description

Imagine you are in the middle of a training run and need to spin up a new A100 instance. Instead of hunting through the RunPod dashboard, navigating menus, and copy-pasting Docker images, you just tell your agent what you need. This Connector gives your agent the keys to your RunPod account. You can ask it to find the right hardware for a specific Llama model, create the pod immediately, and then tell it to keep an eye on the costs by stopping any idle machines. It handles the heavy lifting of infrastructure management, letting you focus on the actual model work. Whether you are managing a fleet of inference endpoints or just need a quick burst of compute power, this tool bridges the gap between your conversation and your cloud hardware. It is one of the many ways the Vinkius catalog helps you put your AI to work on actual infrastructure.

## Tools

### create_pod
Create a new GPU pod with a specific name, hardware type, and Docker image. This is useful for starting new training jobs.

### get_pod
Fetch the status and details of a specific GPU pod. Use this to check if your training job is still active.

### list_endpoints
See every serverless endpoint currently routing your inference traffic. This helps you audit your production deployments.

### list_gpu_types
Check the list of available GPU hardware options in your region. Use this to find the best hardware for your model.

### list_pods
Show every pod in your account, including active and paused ones. This is great for a quick inventory of your resources.

### stop_pod
Halt a running GPU pod to stop incurring hourly costs. This is your primary tool for managing your cloud budget.

### list_templates
View your saved pod configurations and deployment templates. Use this to quickly see your pre-configured environments.

## Prompt Examples

**Prompt:** 
```
List my active pods and tell me which ones are for Llama-3.
```

**Response:** 
```
### Active Pods

| Pod Name | Status | GPU Type |
| :--- | :--- | :--- |
| **llama-3-train-01** | Running | A100 80GB |
| **test-inference** | Paused | L4 |

I found **llama-3-train-01** currently running on an A100 80GB instance. Your **test-inference** pod is currently paused.
```

**Prompt:** 
```
I need a new pod for a training run. What GPUs are available?
```

**Response:** 
```
I checked the available hardware for your region. Here are the current options:

* **NVIDIA A100 (80GB)**
* **NVIDIA H100 (80GB)**
* **NVIDIA L4**
* **NVIDIA T4**

Would you like me to provision one of these for you?
```

**Prompt:** 
```
Stop all my pods that have been running for more than 4 hours.
```

**Response:** 
```
I have scanned your active pods. I have successfully stopped the following instances to save on costs:

* **old-experiment-run** (Stopped)
* **temp-data-processing** (Stopped)

Your active training pod remains running.
```

## Capabilities

### Spin up new GPU pods
Create new hardware nodes with specific names and Docker images from your chat.

### Find available hardware
Check the list of GPU types currently available in your region.

### Stop billable instances
Halt running pods immediately to stop incurring hourly costs.

### Audit serverless endpoints
View all registered endpoints routing your containerized inference applications.

### Manage saved templates
Access and list your saved pod configurations for faster deployment.

## Use Cases

### Quickly spinning up training hardware
An AI researcher needs an A100 for a Llama-3 run. They ask their agent to find an available type and create a pod with their custom Docker image.

### Auditing cloud spend at the end of the day
A DevOps engineer asks the agent to list all pods and stop any that have been idle for over four hours to prevent unnecessary billing.

### Checking production inference routes
An MLOps engineer needs to verify that all serverless endpoints are correctly routing traffic. They ask the agent to list all active endpoints.

### Deploying from saved configurations
A developer wants to recreate a specific environment. They ask the agent to list their templates and provision a new pod based on the best match.

## Benefits

- Stop paying for idle time by using stop_pod to halt instances immediately when they are no longer needed.
- Find the right hardware faster by querying list_gpu_types for specific model requirements.
- Speed up your deployments by pulling from your saved configurations with list_templates.
- Keep your account organized by auditing all active resources with list_pods.
- Manage production traffic easily by checking your list_endpoints status in one go.

## How It Works

The bottom line is you get a hands-free way to manage cloud GPU resources through natural language.

1. Connect your RunPod API key to the Vinkius interface.
2. Grant the agent permission to read and write to your account.
3. Ask the agent to manage your pods or templates.

## Frequently Asked Questions

**Can RunPod MCP help me save money on my GPU costs?**
Yes. You can ask your agent to list all your active pods and stop any that are idle. This prevents you from paying for compute cycles you aren't actually using.

**Does RunPod MCP support custom Docker images?**
Yes, it does. When you ask your agent to create a new pod, you can specify the exact Docker image you want to use for your workload.

**Can I use RunPod MCP to manage serverless inference?**
Yes. You can use it to list and audit all your registered serverless endpoints, making it easier to manage production traffic.

**What hardware can I see with RunPod MCP?**
You can see all the GPU types currently available in your RunPod region, including high-performance options like A100s and H100s.

**How do I connect my RunPod account to this Connector?**
You just need to generate an API key from your RunPod settings and paste it into the secure connection module in your AI client.

**Can the agent create pods automatically?**
Yes. Once connected, your agent can provision new pods immediately based on your natural language requests for specific hardware and images.

**Can the AI forcefully terminate or delete critical production endpoint fleets on demand?**
No. This module safely allows the AI to only pause and manage running instances. Destructive deletion actions (like completely erasing a pod) are intentionally prohibited by the tooling design to protect your critical compute resources from unintended loss.

**Can the AI provision large GPU arrays automatically?**
Yes. Using the `create_pod` capability, the AI can query the available hardware models (such as A100 or H100) and immediately launch new Docker clusters based on existing community templates, simplifying complex DevOps scaling actions significantly.

**Will the AI know the billing state or the real-time cost of running each endpoint?**
No. The current RunPod AI module is concentrated on operational control and system orchestration, such as discovering inactive processes and booting new instances. Deep billing analytics or invoice extraction is not natively integrated in the commands exposed to the AI at this time.