cube-mcp 0.3.0

MCP server for EdgescaleAI Cube operations

Cube MCP Server

MCP server for EdgescaleAI Cube operations via Teleport and Apollo Container Registry (ACR) deployments.

Prerequisites

• Teleport client (tsh) installed
• kubectl installed
• helm installed
• docker with buildx (for image builds)
• aws CLI (for Apollo publishing)
• Access to EdgescaleAI Teleport (edgescaleai.teleport.sh)

Apollo Container Registry (ACR) Setup

To use the deployment tools, add these to your ~/.bashrc or ~/.zshrc:

export APOLLO_CLIENT="<your-client-id>"
export APOLLO_SECRET="<your-client-secret>"

Then restart your terminal or run source ~/.bashrc.

Installation

cd cube-mcp
pip install -e .

Usage with Claude Code

Add to your Claude Code settings (.claude/settings.json or global settings):

{
  "mcpServers": {
    "cube": {
      "command": "cube-mcp"
    }
  }
}

Or run directly with uvx:

{
  "mcpServers": {
    "cube": {
      "command": "uvx",
      "args": ["--from", "/path/to/cube-mcp", "cube-mcp"]
    }
  }
}

Available Tools

cube_login

Login to a Cube cluster via Teleport. Opens browser for SSO authentication.

Parameters:

• cluster (required): Cube cluster name (e.g., staging-int, staging-pre-prod)
• proxy (optional): Teleport proxy URL (defaults to edgescaleai.teleport.sh)

Example:

"Connect me to the staging-int Cube"

cube_list

List all available Cube clusters accessible via Teleport.

Example:

"What Cubes are available?"

cube_status

Get the status of the connected Cube node using kubectl describe node.

Example:

"What's the status of the Cube?"

---

ACR / Apollo Deployment Tools

helm_deploy

The main deployment tool. Builds, packages, pushes, and publishes a Helm chart to Apollo.

Handles the full pipeline:

1. Login to ACR (Docker + Helm)
2. Build and push Docker image (if Dockerfile found)
3. Bump Chart.yaml version
4. Package Helm chart
5. Push to ACR
6. Publish to Apollo via Lambda

Parameters:

• chart_path (optional): Path to chart directory. Auto-detected if not specified.
• version (optional): Exact version to use (e.g., "1.0.0"). Auto-bumps if not specified.
• bump (optional): "patch" (default), "minor", or "major"
• app_version (optional): Docker image tag version
• skip_docker (optional): Skip Docker build even if Dockerfile exists
• dry_run (optional): Show what would happen without doing it

Examples:

"Deploy my chart"                    → auto-detect everything, bump patch
"Deploy my chart as version 1.0.0"   → use exact version
"Deploy my chart, bump minor"        → bump minor version
"Deploy the chart at ./charts/foo"   → specify path
"Deploy my chart, dry run"           → preview without changes

acr_login

Manually login to Apollo Container Registry (Docker + Helm).

Example:

"Login to ACR"

acr_get_token

Get an OAuth2 token from ACR (for debugging).

Example:

"Get ACR token"

Quick Start

Cube Operations

1. Ask Claude to list available Cubes: "What Cubes can I connect to?"
2. Connect to a Cube: "Connect me to staging-int"
3. Check status: "Show me the Cube status"

Helm Deployments

1. Navigate to your chart directory
2. Ask Claude: "Deploy my chart"
3. Or with options: "Deploy my chart, bump minor version"

Manual Setup (if needed)

# Login to Teleport
tsh login --proxy=edgescaleai.teleport.sh

# List available Cubes
tsh kube ls

# Connect to a Cube
tsh kube login staging-int

# Verify
kubectl get nodes

Development

# Install in development mode
pip install -e .

# Run directly
python -m cube_mcp.server

Releasing Updates to Developers

1. Bump the version

Edit pyproject.toml and update the version number:

[project]
version = "0.1.1"  # Bump this

Version format: MAJOR.MINOR.PATCH

• PATCH (0.1.0 → 0.1.1): Bug fixes
• MINOR (0.1.0 → 0.2.0): New features
• MAJOR (0.1.0 → 1.0.0): Breaking changes

2. Build the wheel

pip3 install build  # Only needed once
python3 -m build

This creates dist/cube_mcp-X.X.X-py3-none-any.whl

3. Distribute to developers

Send them:

• The new .whl file
• DEVELOPER_SETUP.md (for new developers)

4. Developer installation

New version (recommended):

pip install cube_mcp-0.1.1-py3-none-any.whl

Same version (force reinstall):

pip install cube_mcp-0.1.0-py3-none-any.whl --force-reinstall

Note: If the version number doesn't change, pip will say "Requirement already satisfied" and won't update. Either bump the version or have developers use --force-reinstall.