Manage Virtual Machines via MCP Server

Gcore MCP Server provides compute management tools that AI clients can invoke through natural language requests. The server returns structured data that the AI client interprets and presents in a readable format.

Before starting, install and configure Gcore MCP Server with GCORE_TOOLS="instances" to load Virtual Machine tools, or GCORE_TOOLS="instances,baremetal,gpu_baremetal,gpu_virtual" for full compute capabilities.

How it works

When a request is made in natural language, the AI client:

1. Resolves the project and region (using lookup tools or configured defaults)
2. Identifies the appropriate MCP tool
3. Extracts parameters from the request
4. Calls the tool and receives JSON data
5. Formats the response for readability

Example request:

List my virtual machines in Luxembourg

The MCP server returns structured data:

{
  "id": "a1b2c3d4-5678-90ab-cdef-1234567890ab",
  "name": "web-server-01",
  "status": "ACTIVE",
  "flavor": {
    "vcpus": 4,
    "ram": 8192,
    "flavor_name": "g1-standard-4-8"
  },
  "addresses": {
    "private": "10.0.0.15",
    "public": "92.38.xxx.xxx"
  },
  "image": "Ubuntu 22.04 LTS"
}

The AI client presents this as readable text, summarizing instance name, status, resources, and network configuration.

View and manage instances

Listing instances returns all Virtual Machines in the specified project and region. The response includes instance names, statuses, flavors, IP addresses, and attached resources.

List my virtual machines

For detailed information about a specific instance, reference it by name or ID:

Show details for instance web-server-01

The detailed view includes additional information such as attached volumes, security groups, network interfaces, and creation timestamp.

To check available configurations before creating an instance:

What VM flavors are available in Luxembourg?

The response lists flavor names with vCPU count, RAM, and any special characteristics like GPU availability or high-frequency CPUs.

Create an instance

Creating a Virtual Machine requires specifying the configuration. The AI client resolves flavor names, images, and network settings from natural language descriptions.

Create a virtual machine named api-server with 4 vCPUs, 8 GB RAM, 
Ubuntu 22.04, and 80 GB SSD in Luxembourg

The AI client maps this to specific API parameters:

• Flavor: g1-standard-4-8 (4 vCPUs, 8 GB RAM)
• Image: Ubuntu 22.04 LTS from available images
• Volume: 80 GB High IOPS SSD
• Region: Luxembourg (region_id resolved from name)

For simpler requests, the AI uses reasonable defaults:

Create a small Linux server for testing

This provisions a standard flavor with Ubuntu and default storage in the configured default region.

Instance creation typically completes within two minutes. The AI client waits for the operation to finish and confirms with the new instance details including assigned IP addresses.

Network configuration

By default, instances receive a private IP address. To attach a public IP:

Create a VM with Ubuntu 22.04 and assign a floating IP

For specific network requirements:

Create a VM attached to network my-private-network with IP 10.0.1.50

Security groups

Security groups control network access to instances. Assign them during creation:

Create a VM with Ubuntu and attach security group allow-ssh-http

Or modify after creation:

Add security group allow-https to instance web-server-01

Instance lifecycle operations

Running instances can be controlled through power state operations.

Stop and start

Stopping an instance powers it down while preserving configuration and attached storage. Billing for compute resources pauses, but storage charges continue.

Stop instance dev-server

Starting resumes the instance:

Start instance dev-server

Reboot

Rebooting restarts the operating system without changing the power state:

Reboot instance web-server-01

Suspend and resume

Suspending preserves the instance memory state to disk, allowing faster resume than a full start:

Suspend instance batch-processor

Resume instance batch-processor

Resize an instance

Resizing changes the instance flavor, adjusting vCPU count, RAM, or both.

Resize instance api-server to 8 vCPUs and 16 GB RAM

Resizing requires stopping the instance. The AI client handles the stop-resize-start sequence automatically. Plan for brief downtime during this operation.

The AI client confirms the new configuration after resize completes:

Instance api-server resized to g1-standard-8-16 (8 vCPUs, 16 GB RAM). 
Instance is now running.

Access instance console

When SSH access is unavailable due to network issues or misconfiguration, console access provides direct connection through the browser.

Get console access for instance db-server

The AI client returns a URL that opens a VNC console session in the browser. Console access works regardless of network configuration and is useful for troubleshooting boot issues or firewall misconfigurations.

Work with instance images

Images serve as templates for creating instances. Custom images capture an instance state for replication or backup.

List available images

What Ubuntu images are available?

The response shows image names, versions, architecture (x86-64 or ARM), and availability by region.

Create an image from an instance

Capturing an image preserves the current state including installed software and configuration:

Create an image from instance web-server-01 named web-template-v2

The resulting image appears in the image list and can be used for creating new instances.

Upload a custom image

External images in standard formats (QCOW2, RAW) can be uploaded:

Upload image from URL https://example.com/custom-image.qcow2 named my-custom-os

Bare Metal servers

Bare Metal provides dedicated physical hardware without virtualization overhead. Use Bare Metal when workloads require direct hardware access, consistent performance without noisy neighbors, or specific hardware configurations.

List Bare Metal options

What Bare Metal configurations are available in Frankfurt?

The response includes server specifications: CPU model, core count, RAM, storage type and capacity, and network bandwidth.

Create a Bare Metal server

Create a Bare Metal server with Ubuntu 22.04 in Frankfurt

Bare Metal provisioning takes longer than VM creation (typically 10-15 minutes) as it involves physical server allocation and operating system installation.

Rebuild a Bare Metal server

Rebuilding reinstalls the operating system while preserving the hardware allocation:

Rebuild server db-bare-metal with CentOS 9

This is useful for repurposing servers or recovering from OS issues without relinquishing the hardware.

GPU clusters

GPU clusters provide multi-node GPU infrastructure for AI/ML training, inference, and high-performance computing workloads. Available in Bare Metal and Virtual configurations.

GPU Bare Metal clusters

Bare Metal GPU clusters provide dedicated physical servers with direct GPU access, interconnected via high-speed networking (InfiniBand where available).

List my GPU Bare Metal clusters

Creating a cluster provisions multiple GPU-equipped servers:

Create a GPU Bare Metal cluster named ml-training with 4 H100 nodes

GPU cluster creation involves significant resource allocation and may take 15-30 minutes. The AI client monitors progress and confirms when all nodes are ready.

Managing cluster nodes:

Get console access for node 1 in cluster ml-training

Reboot all nodes in cluster ml-training

GPU Virtual clusters

Virtual GPU clusters offer flexibility with faster provisioning and easier scaling compared to Bare Metal:

Create a GPU Virtual cluster with 2 A100 nodes for inference

Scaling a cluster:

Add 2 more nodes to cluster inference-dev

Checking attached storage:

Show volumes attached to cluster inference-dev

Configure default project and region

Cloud tools require project and region context. Configure defaults to simplify requests:

GCORE_CLOUD_PROJECT_ID="780356"
GCORE_CLOUD_REGION_ID="76"

With defaults configured, requests like “Create a VM with 2 vCPUs” use the specified project and region automatically without requiring explicit mention.

To find project and region IDs:

List my projects

List available regions

Troubleshooting

If a request fails, the AI client displays the error message from the MCP server. Common issues and resolutions:

Project or region not found

Cloud tools require valid project and region context. If operations fail with “not found” errors, verify the project exists and list available regions:

List my projects

What regions are available for project production?

Flavor not available in region

Not all flavors exist in all regions. GPU and high-frequency flavors have limited regional availability:

What GPU flavors are available in Luxembourg?

The AI client suggests alternative regions when a requested configuration is unavailable locally.

Quota exceeded

Resource limits apply per project. Check current usage:

Show quota usage for project production

If quota is exhausted, delete unused resources or request a quota increase through the Customer Portal.

Instance creation fails

Common causes include:

• Insufficient quota: Check quota limits
• Invalid network: Verify the specified network exists in the region
• Image incompatible with flavor: Some images only work with specific architectures (x86-64 vs ARM)

For detailed VM configuration options, see Types of Virtual Machines. Step-by-step creation through the Customer Portal is documented in Create a Virtual Machine. GPU cluster documentation is available in GPU Cloud.