A cited GPU quote in one minute

Compare GPU rentals before a training run or estimate a job's compute budget. Python 3.10+ is enough: no package installation, account or API key for these read-only queries.

Download, inspect, run

Read the complete example, then download it and run it locally:

curl --fail --location https://raw.githubusercontent.com/connormatthewdouglas/agent-work-stack-deals/main/examples/first_quote.py --output first_quote.py
python3 first_quote.py --gpu 'H100 SXM' --hours 100 --ref quickstart

The example connects to the public hosted MCP, initializes the session, requests three on-demand H100 SXM rates checked within 24 hours, and estimates a 100-hour, one-GPU job. Output is JSON with vendor source links, check times, assumptions, missing costs and a quote receipt. No purchase or watch is created.

Change --gpu 'A100 PCIe', --hours 10 or --gpus 4 for your workload. Omit --hours to compare hourly prices only. Use --qa for internal tests.

Already using an MCP client?

Connect with Streamable HTTP to:

https://lantern-works-mcp.lantern-works.workers.dev/mcp?ref=quickstart

Ask for compare_gpu_prices(gpu="H100 SXM", on_demand_only=true, limit=3, mode="brief"). Client-specific configuration · Agent discovery JSON

Read the answer safely

• Cheapest means among tracked providers. Match the GPU variant and tier; unknown stock is not a promise of capacity.
• The job result is a compute subtotal. Review assumptions and missing_costs for storage, egress and taxes. Whole-instance billing may allocate more GPUs than requested.
• Check source_url and checked_at before acting. An empty fresh result is not evidence that the model is unavailable everywhere.
• Referral links are labeled and never change ranking. Confirm vendor terms before paying.

Only need JSON?

Start with the small per-model feed index, or the complete GPU feed. Enforce stock expiry when reading a static snapshot. See the agent data contract.

Maintain a GPU tool, dataset or agent guide? Add a short integration to your README.