Skip to main content
Glama
Verda Cloud MCP Server
by sniper35

Verda Cloud MCP Server

An MCP (Model Context Protocol) server for managing Verda Cloud GPU instances through Claude. Deploy spot B300 GPU instances, attach volumes, apply startup scripts, and get notified when your instance is ready.

Features

  • List instances - View all your running and stopped instances

  • Check spot availability - Find available B300 (or other) GPU spot instances

  • Deploy spot instances - Create new spot GPU instances with one command

  • Attach volumes - Attach your data volumes to instances

  • Apply startup scripts - Automatically run setup scripts on boot

  • Wait for ready - Polls until your instance is running and returns SSH info

Related MCP server: Vast.ai MCP Server

Quick Start

1. Install dependencies

# Using uv (recommended)
uv sync

# Or using pip
pip install -e .

2. Configure credentials

# Copy the example config
cp config.yaml.example config.yaml

# Edit with your credentials
# Get your API credentials from: https://console.verda.com/dashboard/api

Edit config.yaml:

client_id: "your-actual-client-id"
client_secret: "your-actual-client-secret"

defaults:
  project: "vllm-omni"
  gpu_type: "B300"
  volume_id: "your-volume-id"  # Optional
  script_id: "your-script-id"  # Optional

3. Run the server

# Using uv
uv run python -m verda_mcp

# Or directly
python -m verda_mcp

Claude Desktop Configuration

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "verda-cloud": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/verda-cloud-mcp",
        "run",
        "python",
        "-m",
        "verda_mcp"
      ]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "verda-cloud": {
      "command": "verda-mcp"
    }
  }
}

Restart Claude Desktop after updating the configuration.

Available Tools

Tool

Description

list_instances

List all your Verda Cloud instances

check_instance_status

Get detailed status of a specific instance

check_spot_availability

Check available spot GPU instances (default: B300)

list_volumes

List your block storage volumes

list_scripts

List your startup scripts

list_ssh_keys

List your SSH keys

deploy_spot_instance

Deploy a new spot GPU instance

delete_instance

Delete an instance (requires confirmation)

shutdown_instance

Shutdown a running instance

start_instance

Start a stopped instance

attach_volume

Attach a volume to an instance

detach_volume

Detach a volume from an instance

create_startup_script

Create a new startup script

Example Usage

Once configured with Claude Desktop, you can ask Claude:

Check availability

"Check if there are any B300 spot instances available"

Deploy an instance

"Deploy a B300 spot instance with my data volume and startup script"

Quick workflow

"Check for B300 availability, and if available, deploy one with volume abc123 and script xyz789"

Monitor instances

"List all my instances and their status"

Get SSH info

"Check the status of instance inst-12345 and give me the SSH command"

Configuration Reference

config.yaml

# Required: Verda API credentials
client_id: "your-client-id"
client_secret: "your-client-secret"

# Optional: Default values for deployments
defaults:
  project: "your-project-name"
  gpu_type: "B300"
  volume_id: ""  # Pre-configured volume to attach
  script_id: ""  # Pre-configured startup script
  image: "ubuntu-24.04-cuda-12.8-open-docker"
  hostname_prefix: "spot-gpu"

# Optional: Deployment behavior
deployment:
  ready_timeout: 600  # Max seconds to wait for instance
  poll_interval: 10   # Seconds between status checks
  use_spot: true      # Default to spot instances

Default Project

The defaults.project setting determines how your instances are named and organized:

  • project - A project identifier used for organizing instances (e.g., "vllm-omni", "sglang-slime")

  • hostname_prefix - Automatically set to {project}-{gpu_type} (e.g., "vllm-omni-B300")

When you deploy a new instance, it will be named with the hostname prefix followed by a timestamp.

Switching Projects

To switch to a different project during a session, use the /verda-project command:

/verda-project

This will:

  1. Show your current project and hostname prefix

  2. Ask which project you want to switch to

  3. Update the config.yaml with the new project name and hostname prefix

You can also provide the project name directly:

/verda-project my-new-project

The hostname prefix will automatically be updated to {project}-B300.

Environment Variables

You can also set the config path via environment variable:

export VERDA_MCP_CONFIG=/path/to/config.yaml

Getting Your API Credentials

  1. Log in to Verda Console

  2. Go to Dashboard → API

  3. Create new API credentials

  4. Copy the Client ID and Client Secret

Project Structure

verda-cloud-mcp/
├── pyproject.toml           # Project configuration
├── config.yaml.example      # Configuration template
├── config.yaml              # Your credentials (gitignored)
├── README.md
└── src/
    └── verda_mcp/
        ├── __init__.py
        ├── __main__.py      # Entry point
        ├── server.py        # MCP server with tools
        ├── client.py        # Verda API client
        └── config.py        # Configuration loader

Development

# Install dev dependencies
uv sync --dev

# Run tests
uv run pytest

# Run server in development mode
uv run python -m verda_mcp

Troubleshooting

"Config file not found"

Copy config.yaml.example to config.yaml and fill in your credentials.

"No SSH keys found"

Add an SSH key in the Verda console before deploying instances.

Instance never becomes ready

Check the Verda console for any deployment errors. The instance may have failed to start.

Connection refused

Ensure your firewall allows SSH (port 22) from your IP.

License

MIT

verda-cloud-mcp

Available Tools

20 tools
attach_volumeA

Attach a volume to an instance.

Note: The instance must be shut down first.

Args: volume_id: The ID of the volume to attach. instance_id: The ID of the instance to attach to.

Returns: Confirmation of attachment.

Inputs: volume_id (required), instance_id (required). Returns: result

check_instance_statusB

Check the status of a specific instance.

Args: instance_id: The ID of the instance to check.

Returns: Instance status details including SSH connection info if running.

Inputs: instance_id (required). Returns: result

check_spot_availabilityA

Check if spot GPU instances are available.

Uses the official Verda SDK is_available() method to check across all locations.

Args: gpu_type: GPU type to check (default from config, e.g., "B300", "B200"). gpu_count: Number of GPUs (default from config, e.g., 1, 2, 4, 8).

Returns: Availability status with location if available.

Inputs: gpu_type, gpu_count. Returns: result

create_and_set_default_scriptA

Create a new startup script and set it as the default for new instances.

This creates a new script and updates config.yaml to use it as the default script_id for future instance deployments.

Args: name: Name for the new script. content: Bash script content.

Returns: Confirmation with script ID and updated config status.

Inputs: name (required), content (required). Returns: result

create_startup_scriptB

Create a new startup script.

Args: name: Name for the script. content: Bash script content.

Returns: Created script ID.

Inputs: name (required), content (required). Returns: result

create_volumeA

Create a new block storage volume.

Args: name: Name for the volume (e.g., "my-data-volume"). size: Volume size in GB (default: 150GB from config). volume_type: Volume type (default: "NVMe").

Returns: Created volume details with ID.

Inputs: name (required), size, volume_type. Returns: result

delete_instanceB

Delete an instance.

Args: instance_id: The ID of the instance to delete. confirm: Must be True to actually delete (safety check).

Returns: Confirmation of deletion.

Inputs: instance_id (required), confirm. Returns: result

deploy_spot_instanceA

Deploy a new spot GPU instance.

Args: gpu_type: GPU type (default from config, e.g., "B300"). gpu_count: Number of GPUs (default from config, e.g., 1, 2, 4, 8). volume_id: Block volume ID to attach (default from config). script_id: Startup script ID (default from config). hostname: Instance hostname (auto-generated if not provided). image: OS image (default from config). wait_for_ready: If True, wait for instance to be ready (default: True).

Returns: Instance details and SSH connection info when ready.

Inputs: gpu_type, gpu_count, volume_id, script_id, +3 more. Returns: result

detach_volumeA

Detach a volume from its current instance.

Note: The instance must be shut down first.

Args: volume_id: The ID of the volume to detach.

Returns: Confirmation of detachment.

Inputs: volume_id (required). Returns: result

get_instance_startup_scriptB

Get the startup script attached to a specific Verda Cloud instance.

Args: instance_id: The ID of the instance.

Returns: The script name, ID, and content, or a message if no script is attached.

Inputs: instance_id (required). Returns: result

list_imagesB

List available OS images.

Returns: A list of available OS images.

Inputs: no named parameters. Returns: result

list_instancesA

List all your Verda Cloud instances with their status.

Returns: A formatted list of all instances with ID, hostname, status, type, and IP.

Inputs: no named parameters. Returns: result

list_scriptsB

List your startup scripts.

Returns: A list of scripts with ID and name.

Inputs: no named parameters. Returns: result

list_ssh_keysB

List your SSH keys.

Returns: A list of SSH keys with ID and name.

Inputs: no named parameters. Returns: result

list_volumesB

List your block storage volumes.

Returns: A list of volumes with ID, name, size, and attachment status.

Inputs: no named parameters. Returns: result

monitor_spot_availabilityA

Monitor for spot GPU availability and optionally auto-deploy when available.

Polls using the official Verda SDK is_available() method until a spot becomes available.

Args: gpu_type: GPU type to monitor (default from config). gpu_count: Number of GPUs (default from config). check_interval: Seconds between checks (default: 30). max_checks: Maximum number of checks before giving up (default: 60 = 30 min). auto_deploy: If True, automatically deploy when available (default: False). volume_id: Volume to attach if auto-deploying (default from config). script_id: Startup script if auto-deploying (default from config).

Returns: Status updates and deployment info if auto_deploy is enabled.

Inputs: gpu_type, gpu_count, check_interval, max_checks, +3 more. Returns: result

set_default_scriptA

Set an existing script as the default for new Verda Cloud instances.

Args: script_id: The ID of an existing script to set as default.

Returns: Confirmation of the updated default.

Inputs: script_id (required). Returns: result

show_configA

Show the current MCP server configuration (without secrets).

Returns: Current configuration settings.

Inputs: no named parameters. Returns: result

shutdown_instanceB

Shutdown a running instance (can be restarted later).

Args: instance_id: The ID of the instance to shutdown.

Returns: Confirmation of shutdown.

Inputs: instance_id (required). Returns: result

start_instanceB

Start a stopped instance.

Args: instance_id: The ID of the instance to start.

Returns: Confirmation that start was initiated.

Inputs: instance_id (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 20 tools

    Disambiguation4/5

    Most tools target distinct resource-action pairs (instances, volumes, scripts, SSH keys, config). The main ambiguity is between check_spot_availability, monitor_spot_availability, and deploy_spot_instance—especially since monitor can auto-deploy, blurring its boundary with deploy_spot_instance. create_startup_script and create_and_set_default_script also overlap slightly.

    Naming Consistency5/5

    Tool names consistently follow verb_noun snake_case, e.g., list_instances, create_volume, attach_volume, delete_instance. Minor compounds like create_and_set_default_script and show_config vs list_* are acceptable deviations and don't create confusion.

    Tool Count3/5

    With 20 tools, the server is on the heavier side for a cloud management MCP. While most tools have a purpose, some areas (script management, volume management, spot monitoring) could be consolidated without losing clarity, making the count feel slightly bloated.

    Completeness3/5

    Core instance lifecycle is covered (deploy, list, check, start, shutdown, delete), but there are notable gaps: volumes can be created/listed/attached/detached but not deleted; scripts can be created and set as default but not edited or deleted; SSH keys can be listed but not added or removed. There is also no on-demand instance creation separate from spot deployment.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers