Set up MCP server and Agent Skills

Objective: Create a GPU Pod on YottaLabs using RTX 5090, PyTorch, and JupyterLab through natural language with Cursor's MCP integration.

---

🚀 Quick Start Guide

Follow these 5 steps to create and access your GPU Pod:

Step 1: Prerequisites Check ✅

Before you begin, ensure you have the following:

✅ Node.js >= 18

• Check your version: Run node --version in terminal

• Not installed? Download from nodejs.org

✅ Yotta API Key

• Navigate to Yotta Console

• Go to Settings → Access Keys

• Copy our apu keys securely

✅ Cursor IDE

• Download and install the latest version of Cursor

Step 2: Configure MCP Server ⚙️

Configure the Yotta MCP server in Cursor to enable natural language GPU management.

📁 Locate Configuration File

Find or create the configuration file based on your operating system:

Operating System	Configuration File Path
Windows	C:\Users\<YourUsername>\.cursor\mcp.json
macOS	~/.cursor/mcp.json
Linux	~/.cursor/mcp.json

⚙️ Add Configuration

Create or edit the mcp.json file with the following content:

{
  "mcpServers": {
    "yotta": {
      "command": "npx",
      "args": ["-y", "@yottascale/agent-native-infra"],
      "env": {
        "YOTTA_API_KEY": "your-yotta-api-key-here"
      }
    }
  }
}

⚠️ Important: Replace your-yotta-api-key-here with your actual Yotta API Key from Step 1.

🔄 Restart Cursor

After saving the configuration:

1. Completely close Cursor (all windows)

2. Restart Cursor

3. The Yotta MCP server will now be available

Step 3: Create Your GPU Pod 🎯

Now you can create a GPU Pod using natural language through Cursor Composer.

🎯 Method: Natural Language (Recommended)

1. Open Cursor Composer by pressing:

   • Windows/Linux: Ctrl+I

   • macOS: Cmd+I

2. Type your request in natural language:

Create a Yotta Pod named pytorch-rtx5090-jupyter using RTX 5090 GPU, 
PyTorch image, enable JupyterLab, and set password to yotta2025

1. Let AI handle it - The AI will call the pod_create tool with appropriate parameters

📝 Key Parameters Reference ( just an example )

Parameter	Description	Example Value
name	Pod name	pytorch-rtx5090-jupyter
image	Docker image	yottalabsai/pytorch:2.8.0-py3.11-cuda12.8.1-cudnn-devel-ubuntu22.04-2025081902
gpuType	GPU type	NVIDIA_RTX_5090_32G
gpuCount	Number of GPUs	1
environmentVars	Environment variables	JUPYTER_PASSWORD (required)
expose	Ports to expose	22 (SSH), 8888 (JupyterLab)
regions	Preferred regions	["us-east-1", "us-east-2"]

Step 4: Wait for Pod Creation 🕐

After submitting your request, you'll receive a response confirming the pod creation.

✅ Success Response Example

{
  "id": "420522713875330018",
  "name": "pytorch-rtx5090-jupyter",
  "gpuType": "NVIDIA_RTX_5090_32G",
  "gpuDisplayName": "RTX 5090",
  "gpuCount": 1,
  "singleCardVramInGb": 32,
  "singleCardPrice": "0.65",
  "status": "INITIALIZE",
  "environmentVars": [
    {"key": "JUPYTER_PASSWORD", "value": "yotta2025"}
  ],
  "expose": [
    {"port": 22, "protocol": "SSH"},
    {"port": 8888, "protocol": "HTTP"}
  ]
}

📊 Status Progression

Your pod will go through these states:

1. INITIALIZE → Pod is being created

2. RUNNING → Pod is ready to use (usually takes 1-2 minutes)

Step 5: Access Your Pod 🌐

Once the pod status changes to RUNNING, you can access it.

🌐 Via Yotta Console

1. Open Yotta Console

2. Navigate to Compute → Pods

3. Click on your pod: pytorch-rtx5090-jupyter

4. View the access URLs in the details panel

Get the SSH address from the pod details in the console.

---

Pod Management Commands

Once your pod is running, you can manage it using natural language commands in Cursor Composer:

📋 List All Pods

List all my Yotta pods

🔍 Get Pod Details

Get details for pod pytorch-rtx5090-jupyter

⏸️ Pause Pod (Stop Billing)

Pause pod pytorch-rtx5090-jupyter

💡 Tip: Pause pods when not in use to save costs!

🗑️ Delete Pod

Delete pod pytorch-rtx5090-jupyter

⚠️ Warning: This action is irreversible!