# Lambda Labs (GPU Cloud) MCP for AI Agents AI Agent Connect

> Lambda Labs (GPU Cloud) MCP lets you manage your AI training infrastructure through your AI client. Launch H100 or A100 instances, check regional GPU availability, and handle SSH key management without leaving your chat window. It's built for engineers who need to scale compute on demand.

## Overview
- **Category:** superpower
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_GornvRH1C8hhZ2KJ4gLtkhB2ck8RuwNlliqUvB8m/ai-agent-connect
- **Tags:** gpu-cloud, machine-learning, infrastructure-as-code, virtual-machines, ai-training, ssh-management

## Description

Connect your Lambda Labs account to your AI agent and take full control of your high-performance GPU orchestration through natural conversation. Instead of jumping between different browser tabs to find available hardware or manually spinning up nodes, you can just tell your agent what you need. If you're looking for a way to keep your infrastructure organized, this Connector is part of the Vinkius catalog, making it easy to plug into your existing workflow. You can check which regions have H100s available, grab your Jupyter Lab access tokens, or kill off old instances to save money. It handles the heavy lifting of cloud management so you can stay focused on training your models. You can see your entire fleet at a glance, manage your SSH keys for zero-trust access, and map out persistent NAS volumes that live in the Lambda ecosystem. It turns a complex cloud dashboard into a simple conversation. You no longer have to hunt for pricing matrices or worry about whether a specific node type is actually in stock in the region you need. Your agent can pull the exact hardware specs and connection strings for any running node, letting you jump straight into your work. Whether you're spinning up a massive cluster for a training run or just need a single box for a quick experiment, you can manage the entire lifecycle of your Lambda Labs hardware without ever leaving your workspace.

## Tools

### terminate_instances
Permanently destroy Lambda GPU instances to stop billing instantly. Use this to clean up your cloud footprint.

### list_instances
List all currently running GPU instances on your Lambda Cloud account. This gives you a quick overview of your active fleet.

### get_instance
Get exact hardware details and the SSH connection string for a specific instance. This makes jumping into your terminal much faster.

### launch_instance
Provision a new Lambda GPU virtual machine with specific hardware and SSH keys. This gets your training environment ready quickly.

### list_instance_types
Discover available Lambda GPU instance specifications and regional pricing. This helps you find the best hardware for your budget.

### list_ssh_keys
Enumerate all globally managed public SSH keys in your Lambda account. This ensures you have secure access to your remote boxes.

### list_filesystems
Map persistent shared NAS volumes within the Lambda ecosystem. Use this to manage shared data across multiple workers.

## Prompt Examples

**Prompt:** 
```
Are there any H100s available in the us-east-1 region right now?
```

**Response:** 
```
I've checked the current inventory for you. Here are the H100 options in **us-east-1**:

| Instance Type | Price per Hour | Status |
| :--- | :--- | :--- |
| gpu_1x_h100 | $2.49 | Available |
| gpu_8x_h100 | $19.92 | Available |

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

**Prompt:** 
```
I need to start a training run. Can you launch an A100 in us-west-1 with my default key?
```

**Response:** 
```
Done! I've triggered the provision of a **1x A100** instance in **us-west-1** using your 'default-key'. 

* **Instance ID:** `inst-44219` 
* **Status:** Provisioning 
* **Estimated Ready Time:** ~2 minutes 

I'll let you know as soon as it's ready for SSH.
```

**Prompt:** 
```
Show me everything I have running right now and tell me the SSH string for the training-node.
```

**Response:** 
```
You have 2 active instances running:

* **training-node-01** (1x H100) - us-east-1
* **inference-test** (1x A100) - us-west-1

Here is the connection string for **training-node-01**:
`ssh -i default-key user@172.16.4.12`
```

## Capabilities

### Launch GPU instances
Spin up H100 or A100 virtual machines with pre-configured SSH keys.

### Shut down old nodes
Permanently delete instances to stop billing immediately.

### Check GPU pricing
See a full list of node types and costs across different regions.

### Manage SSH keys
List your public keys to ensure secure access to your remote boxes.

### Map NAS volumes
Find persistent shared storage that works across your worker nodes.

### Get instance details
Pull specific hardware specs and connection strings for any running node.

## Use Cases

### Quickly spinning up a training box
An engineer needs an H100 for a 4-hour run. They ask their agent to launch a box in us-east-1 using their default key via launch_instance.

### Checking regional availability
A researcher wants to find the cheapest A100s. They ask the agent to list all instance types and prices across all regions.

### Cleaning up idle compute
An admin wants to stop costs. They ask the agent to list_instances, identify old ones, and then use terminate_instances to shut them down.

### Mapping shared storage
A data scientist needs to mount a shared drive. They ask the agent to use list_filesystems to find the correct NAS volume.

## Benefits

- Stop hunting for available hardware by using list_instance_types to see regional GPU stock instantly.
- Save money by using terminate_instances to kill off idle nodes the moment your training job finishes.
- Secure your remote access by managing your public keys with list_ssh_keys directly through your agent.
- Get into your workspace faster by using get_instance to grab Jupyter Lab tokens and SSH strings in one go.
- Simplify multi-node setups by using list_filesystems to map persistent NAS volumes across your fleet.

## How It Works

The bottom line is you get a conversational interface for your entire Lambda Labs GPU fleet.

1. Subscribe to this Connector on Vinkius and provide your Lambda Labs API Key.
2. Connect the Connector to your preferred client like Claude or Cursor.
3. Ask your agent to list available GPUs or launch a new training node.

## Frequently Asked Questions

**Can I use Lambda Labs (GPU Cloud) MCP to launch H100s?**
Yes, it lets you provision high-end hardware like H100s or A100s directly through your agent.

**Does Lambda Labs (GPU Cloud) MCP help me save money?**
It does by making it easy to use the termination tool to stop billing on nodes you aren't using.

**Can I see which regions have GPUs available?**
Yes, you can ask the Connector to list instance types to see regional availability and pricing.

**How do I get my SSH keys with Lambda Labs (GPU Cloud) MCP?**
You can use the key listing tool to see all your public keys managed in your account.

**Can I find shared storage using Lambda Labs (GPU Cloud) MCP?**
Yes, it lets you map out persistent shared NAS volumes in the Lambda ecosystem.

**Can I launch a high-performance H100 instance through my agent?**
Yes. Use the `launch_instance` tool and specify the type (e.g. `gpu_1x_h100`) and region. Your agent will also allow you to attach registered SSH keys so the instance is securely accessible immediately upon boot.

**How do I retrieve the Jupyter Lab access token for a running node?**
Use the `get_instance` tool with your Instance ID. Your agent will fetch the complete telemetry, including the public IP and the Jupyter Lab access token if the environment is configured to provide it.

**Can my agent check for GPU availability across different regions?**
Absolutely. The `list_instance_types` tool queries the cloud boundary for hardware inventory. Your agent will report which GPU node types are currently available and in which physical regions they are hosted.