Live multi-cloud GPU capacity in one source-agnostic catalog. An agent discovers it over MCP, pays in USDC, and gets a real SSH pod — no account, no API key, no human in the loop. Payment is the authorization.
One price, all-in. Launch to any workload.
Serve models with low latency. Spin up a 4090 or L40S in seconds, scale to zero when idle.
Adapt open models on 24–80GB cards. LoRA to full-parameter, your framework, your data.
Multi-GPU A100/H100 clusters for from-scratch runs. Aggregated capacity means you actually find them.
list_gpu_offers over MCP (or GET /api/pods). One normalized schema, one honest USD/GPU-hour price.
x402 on Base — the 402 challenge is the guaranteed price — or a plain USDC transfer on Base or Solana.
The pod provisions on settlement and terminates itself when the paid hours end. Status via get_rental_status.
List them on V-Pod, set your price, and earn while others train and infer. You stay in control of availability.
Seven tools over Streamable HTTP: discover, quote, pay, provision, monitor, close. Open tools/list, typed schemas, exact USDC pricing where the 402 challenge is the guaranteed price. Idempotent paid calls — a replayed payment returns the same rental, never a double charge.
rent_gpu_hours — x402 on Base, provisions a real SSH podredeem_rental_transfer — plain USDC transfer, Base or Solana# connect any MCP client
claude mcp add v-pod --transport http https://v-pod.com/mcp
# or raw JSON-RPC — discovery needs no auth
curl -X POST https://v-pod.com/mcp \
-H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{"name":"list_gpu_offers",
"arguments":{"gpu":"H100","limit":3}}}'
# rent 2 hours: unpaid call returns the exact 402 terms
# → sign EIP-3009 for that amount → retry with X-PAYMENT
# → { rental_id, status:"active", connect:{ssh…} }