Onboard Team Members with AI Agents

Use MCP tools to show available templates, explain environments, create a dev environment, and deploy a first instance — helping new team members get started.

When a new team member joins, your AI agent can walk them through the workspace: show available templates, explain the environment structure, create a dev environment, and deploy their first instance.

---

”I just joined the team. What resources and templates are available?”

The agent gives an overview:

1. list_templates — shows available instance templates
2. list_environments — shows the environment structure
3. list_gpu_types — shows GPU options with pricing
4. billing_balance — checks workspace credits

Welcome to the workspace. Here’s what’s available:

Instance templates:

Template	Description
ubuntu-devbox	Ubuntu 22.04 with Python 3.11, CUDA 12.4, dev tools
ubuntu-train	Ubuntu with PyTorch, DeepSpeed, NCCL pre-installed
ubuntu-inference	Ubuntu with vLLM, TGI, optimized for serving

Environments: main (default), staging, production

GPU options:

GPU	VRAM	Price/hr
RTX 4090	24 GB	$0.35/hr
A100 80 GB	80 GB	$1.60/hr
H100 80 GB	80 GB	$2.50/hr

Workspace balance: $1,247.80

For development, I’d recommend starting with an RTX 4090 ($0.35/hr) using the ubuntu-devbox template.

---

”Create a dev environment for me and deploy my first instance.”

The agent sets up the new team member:

1. create_environment — creates a dev-alex environment
2. create_instance — deploys alex-devbox with RTX 4090 using the devbox template
3. instance_status — waits until running
4. get_instance — retrieves connection details

• Created environment dev-alex
• Deployed alex-devbox (RTX 4090, 24 GB VRAM, $0.35/hr)

Connect: ssh root@203.0.113.55

Pre-installed: Python 3.11, CUDA 12.4, pip, git, htop, tmux. Your instance runs in the dev-alex environment, separate from staging/production. All environments share the workspace credit balance. Terminate when done to stop billing.

---

”Show me what the team is running so I understand the setup.”

The agent maps out the workspace:

1. list_environments — lists all environments
2. list_instances — lists all instances across environments
3. list_volumes — lists all storage
4. list_ssh_keys — lists team SSH keys

Team infrastructure overview:

production (do not modify without approval)

• prod-inference (H100) — Llama 70B vLLM server, running 12 days
• model-weights volume (500 GB) — production model checkpoints

staging

• staging-test (A100) — testing new model versions before promotion

main (shared development)

• dev-box (RTX 4090) — Sarah’s dev instance

dev-alex (your environment)

• alex-devbox (RTX 4090) — your dev instance

Team SSH keys: 4 keys registered (macbook-pro, ci-deploy, workstation, alex-laptop)

Ground rules: Production instances are managed by the ML ops lead. Staging is for pre-production testing. Use your own environment (dev-alex) for experiments. Terminate instances when not in use.

---

Tools used in this workflow

Tool	Purpose
list_templates	Show available instance templates
list_environments / create_environment	View and create environments
list_gpu_types	Show GPU options and pricing
billing_balance	Check workspace budget
create_instance / instance_status / get_instance	Deploy and connect
list_instances / list_volumes / list_ssh_keys	Map team infrastructure