FastGPU

npx add-mcp 'https://fastgpu.co/api/mcp'

FastGPU API

One normalized, always-fresh feed of live GPU cloud rental prices and availability across the whole market: marketplaces, neoclouds and hyperscalers, ranked cheapest-first. Every value is a live figure with a fetch timestamp. Power procurement tools, FinOps dashboards, or your agents, over plain REST or MCP. Prefer a bulk download? Grab the free open dataset (CSV/JSON, CC BY 4.0). To try every endpoint without writing code, open the public FastGPU API workspace on Postman. To see the matcher size any Hugging Face model and price the GPUs that hold it, open the VRAM & GPU cost calculator on Hugging Face, a browser client of /api/v1/match.

GET

/api/v1/gpus

public

One entry per GPU with the current cheapest live rental price. No key required.

curl https://fastgpu.co/api/v1/gpus

{
  "updated_at": "2026-10-19T22:08:34.0213+00:00",
  "count": 64,
  "gpus": [
    {
      "gpu": "H100 SXM", "slug": "h100-sxm", "vram_gb": 80,
      "vendor": "NVIDIA", "tier": "flagship",
      "cheapest_usd_hr": 1.99, "cheapest_provider": "Voltage Park",
      "cheapest_min_gpu_count": 8, "cheapest_billed_separately": null,
      "provider_count": 29, "url": "/gpus/h100-sxm",
      "page_url": "https://fastgpu.co/gpus/h100-sxm"
    }
  ]
}

GET

/api/v1/prices

API key

Every normalized live offer (provider, GPU, type, region, per-GPU price, availability). Filter with gpu, provider, type, limit.

curl -H "Authorization: Bearer fgpu_live_..." \
  "https://fastgpu.co/api/v1/prices?gpu=h100-sxm&type=on-demand&limit=20"

{
  "updated_at": "2026-10-19T22:08:34.0213+00:00",
  "count": 20,
  "offers": [
    {
      "provider": "voltagepark", "provider_label": "Voltage Park",
      "gpu_model": "H100 SXM", "gpu_slug": "h100-sxm", "vram_gb": 80,
      "offer_type": "on-demand", "region": null,
      "price_usd_hr": 1.99, "min_gpu_count": 8,
      "billed_separately": null,
      "available_count": null, "fetched_at": "2026-10-19T22:08:34.0213+00:00"
    }
  ]
}

Price history API

API key

The recorded history behind the live feed, with every published correction applied. How far back each endpoint reaches depends on your plan; a request outside it answers 403 with the plan that covers it. Where the data comes from is on the data provenance page.

GET/api/v1/history/daily

One row per UTC day, provider, GPU, offer type and region: the day’s low, average and high posted price and the number of observations, the same columns as the open dataset’s daily file. Parameters: from, to (UTC days, up to 31 per request; next_from continues a longer range), gpu, provider, format=csv.

curl -H "Authorization: Bearer fgpu_live_..." \
  "https://fastgpu.co/api/v1/history/daily?gpu=h100-sxm&from=2026-09-01&to=2026-09-30"

GET/api/v1/history/observations

The raw record: one row per offer per collection run, with its timestamp, posted price and stock signal, oldest first. Up to 7 days and 5,000 rows per request; pass next_cursor back as cursor until it is null.

curl -H "Authorization: Bearer fgpu_live_..." \
  "https://fastgpu.co/api/v1/history/observations?gpu=h100-sxm&from=2026-10-01&to=2026-10-02"

GET

Plans and limits

Every keyed request counts toward your account’s daily limit (all keys of an account share it; the count starts again at 00:00 UTC). Each response carries x-ratelimit-limit, x-ratelimit-remaining, x-ratelimit-reset and x-fastgpu-plan; over the limit the answer is 429. Prices are on the pricing page.

Plan	Requests a day	Daily series	Raw observations	Bulk files	Licence
Free	1,000	Last 60 days	Not included	No	Internal use
Pro	10,000	Last 60 days	Last 7 days	No	Internal use

Router API

API keyprovisions real GPUs

Beyond comparing prices, launch and manage a real GPU programmatically. FastGPU routes your job to the cheapest in-stock provider (RunPod, Vast.ai, Hyperstack), provisions it, and bills your prepaid wallet by the second. It is the same routing engine the dashboard uses. Fund your wallet in /app.

POST/api/v1/router/launch

Route + provision. Body: query (describe the job).

curl -X POST -H "Authorization: Bearer fgpu_live_..." -H "Content-Type: application/json" \
  -d '{"query":"fine-tune Llama 3 8B"}' \
  https://fastgpu.co/api/v1/router/launch

{
  "instance": {
    "id": "3f8c…", "status": "launching",
    "provider": "vast", "provider_label": "Vast.ai",
    "gpu_model": "RTX 3090", "gpu_count": 2,
    "price_usd_hr": 0.118, "connection": null
  }
}

GET/api/v1/router/instances/{id}

Poll after launch until status is running, then open the connection URL.

{
  "instance": {
    "id": "3f8c…", "status": "running",
    "gpu_model": "RTX 3090", "price_usd_hr": 0.118,
    "connection": { "jupyterUrl": "https://85.234.22.27:8080/?token=…" }
  }
}

GET/api/v1/router/instances

List every instance + your wallet balance.

MCP server

for AI agentsno install

Connect any MCP client (Claude, Cursor, or an autonomous agent) to the hosted server over Streamable HTTP. Nothing to install. Tools exposed:

• list_gpu_prices List cheapest GPU prices
• match_workload Match a workload to the cheapest GPU

{
  "mcpServers": {
    "fastgpu": { "url": "https://fastgpu.co/api/mcp" }
  }
}

Machine-readable discovery: /openapi.json (OpenAPI 3.1), agent card, llms.txt, and the public Postman workspace.

Agent skill

Agent Skills format

A skill teaches an agent the workflow around the tools above: how to turn a budget into hard limits (budget_usd_hr, max_total_usd with duration_hours, max_quote_age_minutes), check the answer, and hand the launch to the provider's own tooling only after the user agrees. The skill and the tools it uses only compare prices: they never rent or pay for anything.

• gpu-procurement Describe a GPU job, apply a budget (price per hour, total for the job, quote age), choose the cheapest setup that fits, then hand the launch to the provider's own tooling.

Install it into any agent that reads Agent Skills with the open-source skills installer, which finds it through the skills index:

npx skills add https://fastgpu.co

Get an API key

Create and manage keys in your dashboard to access the full offer feed. Keys are shown once at creation. Include yours as an Authorization: Bearer header.

Create an API key →