QUICK START

ComputeGpu + Claude Code

GPU compute for Claude Code agents. As easy as git. 30 seconds to set up.

Step 1: Install + Signup

npm install -g computegpu
computegpu signup you@company.com my-project

API key saved to ~/.computegpu/config.json. Owner verifies with $1 (credited to wallet).

Already have a key? computegpu login --key YOUR_KEY

Step 2: Add MCP Server

Add to your project's .mcp.json:

{
  "mcpServers": {
    "computegpu": {
      "type": "stdio",
      "command": "computegpu",
      "args": ["mcp-serve"]
    }
  }
}

Claude Code discovers 11 GPU tools automatically: deploy, stop, destroy, types, cheapest, pods, status, resume, topup, and admin tools.

Step 3: Start Using

# Browse GPUs
computegpu types

# Deploy a GPU
computegpu deploy --gpu "RTX 4090" --models llama3.3

# Check status
computegpu pods
computegpu status POD_ID

# Stop when done
computegpu stop POD_ID
computegpu destroy POD_ID

Step 4: Add to CLAUDE.md

## GPU Compute
This project uses ComputeGpu for GPU instances.
- CLI: `computegpu` (installed globally)
- MCP: 11 tools available via .mcp.json
- Deploy: `computegpu deploy --gpu "RTX 4090"`
- Docs: https://computegpu.com/docs

Quick Reference

Command	What it does
computegpu types	47 GPU models with live pricing
computegpu cheapest --vram 24	Find cheapest GPU for VRAM
computegpu deploy --gpu "RTX 4090"	Deploy on-demand instance (~30s)
computegpu deploy --gpu "RTX 4090" --spot	Deploy spot (50% off, preemptible)
computegpu pods	List active instances
computegpu status POD_ID	Instance status + billing
computegpu stop POD_ID	Pause (billing stops)
computegpu resume POD_ID	Resume paused instance
computegpu destroy POD_ID	Permanent destroy
computegpu balance	Wallet balance
computegpu topup 10	Request $10 top-up

Pricing

Tier	Range	Example
Consumer	$0.14–$0.83/hr	RTX 4090: $0.41/hr
Datacenter	$0.42–$8.33/hr	H100 SXM: $3.23/hr
Spot	~50% off	RTX 4090 spot: $0.20/hr

Per-second billing. $1 minimum to start. Stop anytime.