Provision Stateless GPU Compute with Claude Code's Remote Control

Name: Provision Stateless GPU Compute with Claude Code's Remote Control
Availability: InStock
Author: Facingsouth

by Facingsouth·Feb 25, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidBig BrainNiche Gem

Claude talks to RunPod/Lambda/Lambda/Vast — but needs working provider integrations to matter.

Strengths

•MCP abstraction layer cleanly separates Claude from cloud APIs; credentials stay local and pluggable.
•Multi-cloud price comparison in one query solves genuine friction for researchers comparing spot rates.
•Kubernetes + InferX deployment pipeline addresses real cold-start latency pain in ML inference.

Weaknesses

•README shows only RunPod setup working; unclear which of 11 providers are actually implemented vs. planned.
•No evidence of tested dry-run execution, cost tracking accuracy, or multi-cloud failover in production.
•Competes with Runwayml, Modal, and cloud-native IDEs that already bake GPU provisioning into their workflows.

Post Description

claude mcp add terradev --command terradev-mcp

Ask Claude Code to find the cheapest spot A100 from your own directory of APIs for providers (keys kept local), dry-run multi-cloud provisioning, compress and cache datasets for egress optimization, spin up NUMA-aware Kubernetes clusters, and deploy a GPU snapshot to InferX for fast cold starts, all with conversational language, all running locally with your own API keys.