gpu.nz GPU cloud

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Rent cloud GPUs from AI agents: live prices, VRAM estimates, launch, logs and stop.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "gpu.nz GPU cloud": {
      "url": "https://gpu.nz/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (16)

• list_gpu_models read-only

  GPU models available on gpu.nz right now with VRAM, class and the cheapest hourly price per GPU. Use it to pick a gpu_slug for search_offers.

  Parameters:

  • query — string — Filter by name, e.g. "4090" or "H100"
  • min_vram_gb — number — Only models with at least this much VRAM per GPU

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• search_offers read-only

  Search rentable GPU servers (offers) with live prices. Returns the cheapest matches first unless `order` says otherwise. Each result has an offer id for create_instance.

  Parameters:

  • gpu — string — GPU model slug(s), comma-separated, e.g. "rtx-4090-24gb" or "h100-sxm-80gb,h100-pcie-80gb" (see list_gpu_models)
  • type — string — on_demand (default) or interruptible (cheaper, can be preempted)
  • limit — integer — How many offers to return (default 10)
  • order — string — Sort: price (default), dlperf_per_price for value, reliability, vram, inet_down
  • country — string — ISO-2 country codes, comma-separated, e.g. "DE,NL,FI"
  • num_gpus — string — GPUs per server, comma-separated, e.g. "1" or "2,4,8"
  • verified — boolean — Only verified datacenters
  • min_disk_gb — number — Minimum disk the server can give, GB
  • min_vram_gb — number — Minimum VRAM per GPU, GB
  • max_price_rub_h — number — Maximum price for the whole server, rubles per hour
  • min_reliability — number — Minimum host reliability, 0–1 (0.98 is a good bar)

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• get_offer read-only

  Details and the current price of one offer by id (from search_offers).

  Parameters:

  • offer_id (required) — string — Offer id

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• list_templates read-only

  Ready environments a server can start with: ComfyUI, Stable Diffusion Forge, Ollama + Open WebUI, vLLM, Jupyter/PyTorch, Kohya, FluxGym, Unsloth, chats with preloaded models, plain Ubuntu + CUDA, or a custom Docker image. Pass `slug` as `template` to create_instance.

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• estimate_vram read-only

  How much GPU memory an AI model needs (LLMs such as Qwen, DeepSeek, Llama, Gemma, GPT-OSS; image and video models such as FLUX, SDXL, Wan) at a given precision and context length, and which GPUs on gpu.nz fit it with their current price. Call without `model` to list the known models.

  Parameters:

  • model — string — Model name or id, e.g. "qwen3-32b", "Llama 3.3 70B", "flux.1-dev"
  • precision — string — Weights precision: "fp16", "fp8", "q8", "q6", "q5", "q4" (LLMs; default q4 for LLMs, fp16 for image models)
  • context_tokens — integer — LLM context length in tokens (default 8192)
  • concurrent_requests — integer — Parallel sequences sharing the GPU (default 1)

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world no

• get_account read-only

  The signed-in account and its balance, hourly burn and runway. Needs a key with user:read and billing:read.

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• list_instances read-only

  Your rented servers with status, GPU, price and SSH command. Needs instances:read.

  Parameters:

  • status — string — active (default): servers that are not destroyed; all: include destroyed ones

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• get_instance read-only

  Status, SSH command, app URLs (ComfyUI, Jupyter, chat…), price and cost so far of one server. The root password is left out unless include_password is true (your SSH keys are installed on every server).

  Parameters:

  • instance_id (required) — string — Instance id
  • include_password — boolean — Include the root SSH password

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• create_instance

  Rents the server of an offer and starts it with a template. Billing starts right away and runs every minute until you stop or destroy it. Confirm the price with the user first. Needs instances:write. The response has the instance id; poll get_instance until status is running (usually 1–5 minutes).

  Parameters:

  • offer_id (required) — string — Offer id from search_offers
  • template (required) — string — Template slug from list_templates, e.g. "comfyui", "ollama", "pytorch", "custom"
  • disk_gb (required) — integer — Disk size, GB (use the template's recommended_disk_gb)
  • name — string — A name for the server
  • type — string — on_demand (default) or interruptible
  • image — string — Docker image, only with the custom template
  • ssh_key_ids — array — SSH key ids to install (default: all of yours)
  • idempotency_key — string — Reuse the same value when retrying the same request, so a retry never rents twice

  Hints: read-only no · destructive no · idempotent no · open-world yes

• stop_instance

  Stops a running server: the GPU is released and billing drops to the disk only. Data on the disk is kept and the server can be started again (the GPU may then be taken by someone else; gpu.nz finds a replacement when it is). Needs instances:write.

  Parameters:

  • instance_id (required) — string — Instance id

  Hints: read-only no · destructive no · idempotent yes · open-world yes

• start_instance

  Starts a stopped server again; GPU billing resumes. Confirm with the user. Needs instances:write.

  Parameters:

  • instance_id (required) — string — Instance id

  Hints: read-only no · destructive no · idempotent yes · open-world yes

• destroy_instance destructive

  Destroys a server and deletes its disk for good; billing stops completely. Irreversible: confirm with the user. Needs instances:write.

  Parameters:

  • instance_id (required) — string — Instance id

  Hints: read-only no · destructive yes · idempotent yes · open-world yes

• get_instance_logs read-only

  The last lines of a server's container log: boot progress, app start-up, errors. Needs instances:read.

  Parameters:

  • instance_id (required) — string — Instance id
  • tail — integer — How many last lines (default 200)

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• list_ssh_keys read-only

  SSH public keys on the account; they are installed on new servers. Needs user:read.

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world yes

• add_ssh_key

  Adds an SSH public key (a line from a .pub file) to the account. Needs user:write.

  Parameters:

  • public_key (required) — string — The public key line, e.g. ssh-ed25519 AAAA… you@laptop
  • name — string — A name for the key

  Hints: read-only no · destructive no · idempotent no · open-world yes

• topup_link read-only

  A link to the balance top-up page (SBP, crypto) for the user to open in a browser; payments never happen through the API. Optionally pre-fills the amount.

  Parameters:

  • amount — integer — Amount in the account currency (rubles by default)
  • language — string — Page language (default ru)

  Hints: read-only yes · destructive not declared · idempotent not declared · open-world no

Publisher website Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.