Live public cloud GPU rental prices (USD per GPU-hour) across 20+ providers. Free, no API key. Tools: cheapest GPU, compare providers, price history.
Scored across 3 tools
Each tool targets a distinct granularity: one cheapest offer (cheapest_gpu), a filterable per-provider price table (get_prices), and an all-GPU overview (list_gpus). The descriptions explicitly tell the agent which tool to pick in which situation, eliminating overlap.
get_prices and list_gpus follow a clean verb_noun pattern, while cheapest_gpu is an adjective_noun outlier describing an output rather than an action. The deviation is minor and still self-explanatory, so the set reads as coherent.
Three tools map neatly onto three legitimate access modes (single best price, filtered table, overall summary), so nothing is redundant. It is on the thin side, but for a read-only price lookup service that scope is defensible rather than padded.
The read-only domain is essentially fully covered: overview, aggregate ranges, and raw per-provider rows with variant/term/region fields. Minor gaps remain, such as no dedicated provider listing and no historical price trend access, but core price-discovery workflows have no dead ends.
Cheapest public on-demand rental price for one GPU model (e.g. H100, A100, L40S, RTX 4090), with provider, source URL and comparison page.
Inputs: gpu (required).
Search the raw per-provider GPU rental price table (USD per GPU-hour), sorted cheapest first. Use this when you need more than the single cheapest offer: comparing providers for one GPU, finding spot, preemptible or reserved rates, or checking one provider's prices. For just the lowest on-demand price of one GPU, call cheapest_gpu; for an overview of all GPUs, call list_gpus. Returns up to limit rows, each with provider, gpu, variant, price, price_type, term, region, source_url and date_checked. Rows without a public numeric price (contact sales) are excluded.
Inputs: gpu, limit, provider, price_type. Returns: rows, count
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Add one secure layer between your agents and this server.