What you’ll do
Use Claude Code, Cursor, or any MCP-compatible agent to:- Deploy and terminate GPU instances on demand
- Run shell commands on remote machines
- Monitor GPU utilization and training progress
- Track spending and manage storage
- Automate multi-step workflows like “deploy, train, download, teardown”
Setup
One click, one command or one URL; there is no config file to edit and no key to paste. Every client opens your browser to sign in on first use. All clients, including Claude, ChatGPT and Codex, are in Setup.Add to Cursor
One click. Cursor asks to install, then signs you in on first use.
Install in VS Code
One click. VS Code adds the server, then signs you in on first use.
Claude Code
/mcp in Claude Code and pick runcrate.
Every client on this machine at once
Real conversations
”What GPUs do I have running and what are they costing me?”
The agent callslist_instances and summarizes:
You have 3 running instances:Balance: $342.15, about 84 hours at the current burn rate. Each instance’s live hourly rate is in
- training-v2: 1x A100, running 14 hours
- dev-box: 1x RTX 4090, running 3 hours
- inference-prod: 1x H100, running 48 hours
list_instances.
”My training script is throwing CUDA OOM errors. Check GPU memory on training-v2.”
The agent callsssh_execute with nvidia-smi:
training-v2 GPU memory usage:You’re almost maxing out the A100 40GB. Options:
- GPU 0: 38,912 MiB / 40,960 MiB (95% used)
- Reduce batch size
- Enable gradient checkpointing
- Switch to an A100 80GB instance
”Deploy a fresh H100 for a quick benchmark, run my benchmark script, and tear it down.”
The agent orchestrates the full workflow:create_instance: deploys an H100 called “benchmark”instance_status: polls until deployedssh_execute:pip install torch && python -c "import torch; ...benchmark code..."- Returns the benchmark results
delete_instance: tears down immediately
”Kill anything that’s been idle for more than 6 hours.”
The agent callslist_instances, identifies instances by creation time, uses ssh_execute to check GPU utilization on each, and calls delete_instance on idle ones:
Checked 3 instances:Both have stopped billing.
- training-v2: GPU at 98% utilization, keeping
- dev-box: GPU at 0% for 4 hours, terminated
- old-experiment: GPU at 0% for 12 hours, terminated
”I need a persistent volume for my dataset. Create 500GB in us-east, then show me what regions are available.”
The agent callslist_storage_regions, then create_volume:
Created volume imagenet-data (500 GB) in us-east-1. Cost: ~$3.46/week ($15/month). Available regions: us-east-1, us-west-2, eu-central-1, ap-southeast-1.
”Set up a vLLM server on an A100 with Llama 3.1 70B and give me the endpoint URL.”
The agent chains:create_instancewithgpu: "A100"andtemplate: "ubuntu-inference"(vLLM installed)instance_statuspollingssh_execute:systemctl start runcrate-python, which returns once vLLM has finished installingssh_execute: starts vLLM with--model meta-llama/Llama-3.1-70B-Instruct --port 8000 --host 0.0.0.0get_instance: returns the IP
Your vLLM server is live at:Use it as a drop-in OpenAI endpoint.
”What templates and environments do I have?”
The agent callslist_templates and list_environments:
Templates: ubuntu-devbox, ubuntu-train, ubuntu-inference Environment: staging (the one you picked when you connected me)
“Upload my train.py script to the training-run instance at /workspace/train.py”
The agent callsfile_upload with the file content and remote path:
Uploaded train.py (4.2 KB) to training-run at /workspace/train.py.
”Create a new environment called ‘production’ for me”
The agent can’t: it works only in the environment you picked when you signed it in, so it points you to the dashboard:I can only work in staging, the environment this connection is signed in to, so I can’t create or delete environments. A workspace owner or admin can add production in Settings, Environments. To have me work there, disconnect Runcrate in your client and sign in again, picking production.
What it can’t do
- Create, delete or switch environments, or change a budget (use the dashboard; to move the agent, reconnect and pick another environment)
- Port forwarding or SSH tunnels (use native SSH)
- Modify billing settings (use the dashboard)
- Create or delete workspaces (use the dashboard)
Tips
- Be specific with instance names: the agent uses them to target
ssh_executeanddelete_instance - Ask the agent to check
nvidia-smianddf -hbefore debugging: most issues are GPU OOM or disk full - Chain requests: “deploy, install, run, download, teardown” in a single message works
- The agent remembers instance IDs within a conversation, so you can say “check the status of that instance” after deploying one