VRAMfarm GPU Cloud is an MCP server published by com.vramfarm, hosted at vramfarm.com. When last checked on 2026-11-22 it was working: it answers and lists its tools. It declares 7 tools, including list_gpu_types, list_offers, launch_server, list_servers, get_server.
Rent EU cloud GPUs per minute: pick a card by VRAM and price, launch, get SSH, stop.
1 connections through 4 shared values. Tap any node.
Inferred from the host name and headers: what a server runs on, not who runs it.
One row per GPU model and card count: the cheapest reliable machine, its price in EUR per hour, VRAM per card and how many are available. Start here to choose a GPU.
No parameters.
{
"name": "list_gpu_types",
"title": "GPU types and prices",
"description": "One row per GPU model and card count: the cheapest reliable machine, its price in EUR per hour, VRAM per card and how many are available. Start here to choose a GPU.",
"inputSchema": {
"type": "object",
"properties": {},
"additionalProperties": false
},
"annotations": {
"readOnlyHint": true
}
}Concrete machines of one GPU model, cheapest first (up to 10), with country, CUDA, disk, network and price. Use the id as offer_id in launch_server to pick an exact machine.
None yet. Changes appear after a server's second daily check.
Squares are this server's values; dots are other servers sharing them. A shared address or publisher is a lead, not proof of a common operator.
| Parameter | Type | Description |
|---|---|---|
| gpu * | string | GPU model code, e.g. rtx-4090, a100, h100 (see list_gpu_types) |
| count | integer | Cards per machine |
{
"name": "list_offers",
"title": "Available machines",
"description": "Concrete machines of one GPU model, cheapest first (up to 10), with country, CUDA, disk, network and price. Use the id as offer_id in launch_server to pick an exact machine.",
"inputSchema": {
"type": "object",
"required": [
"gpu"
],
"properties": {
"gpu": {
"type": "string",
"description": "GPU model code, e.g. rtx-4090, a100, h100 (see list_gpu_types)"
},
"count": {
"enum": [
1,
2,
4,
8
],
"type": "integer",
"description": "Cards per machine"
}
},
"additionalProperties": false
},
"annotations": {
"readOnlyHint": true
}
}Rents a GPU machine and starts a template or a Docker image on it. Costs money from the prepaid balance, billed per minute while running (minimum 15 minutes). Returns at once with the server id; the server is ready in 2–5 minutes (8–15 with a custom image). Refuses machines priced above max_eur_per_hour.
| Parameter | Type | Description |
|---|---|---|
| cmd | string | Start command for template custom |
| gpu * | string | GPU model code, e.g. rtx-4090 (see list_gpu_types) |
| disk | integer | Disk in GB (raised to the template minimum) |
| count | integer | Number of cards, default 1 |
| image | string | Docker image for template custom, as in docker pull |
| offer_id | integer | Exact machine id from list_offers (optional) |
| template | string | What to run; default pytorch |
| max_eur_per_hour * | number | Your budget: the most this server may cost per hour, in EUR, for all cards together |
| restore_workspace | boolean | Restore the saved /workspace folder, default true |
{
"name": "launch_server",
"title": "Launch a GPU server",
"description": "Rents a GPU machine and starts a template or a Docker image on it. Costs money from the prepaid balance, billed per minute while running (minimum 15 minutes). Returns at once with the server id; the server is ready in 2–5 minutes (8–15 with a custom image). Refuses machines priced above max_eur_per_hour.",
"inputSchema": {
"type": "object",
"required": [
"gpu",
"max_eur_per_hour"
],
"properties": {
"cmd": {
"type": "string",
"description": "Start command for template custom"
},
"gpu": {
"type": "string",
"description": "GPU model code, e.g. rtx-4090 (see list_gpu_types)"
},
"disk": {
"type": "integer",
"description": "Disk in GB (raised to the template minimum)"
},
"count": {
"type": "integer",
"maximum": 8,
"minimum": 1,
"description": "Number of cards, default 1"
},
"image": {
"type": "string",
"description": "Docker image for template custom, as in docker pull"
},
"offer_id": {
"type": "integer",
"description": "Exact machine id from list_offers (optional)"
},
"template": {
"enum": [
"pytorch",
"jupyter",
"comfyui",
"comfy-flux",
"comfy-wan",
"comfy-sdxl",
"comfy-wan14",
"comfy-qwen",
"comfy-qedit",
"comfy-hidream",
"chat",
"chat-pro",
"vllm",
"custom",
"tts-qwen3",
"tts-voxcpm",
"tts-chatterbox",
"chat-mimo"
],
"type": "string",
"description": "What to run; default pytorch"
},
"max_eur_per_hour": {
"type": "number",
"description": "Your budget: the most this server may cost per hour, in EUR, for all cards together"
},
"restore_workspace": {
"type": "boolean",
"description": "Restore the saved /workspace folder, default true"
}
},
"additionalProperties": false
},
"annotations": {
"openWorldHint": true,
"idempotentHint": false,
"destructiveHint": false
}
}Your servers, newest first: running, starting and stopped in the last 7 days, with status, price and access.
| Parameter | Type | Description |
|---|---|---|
| status | string | Only servers in this state: provisioning (starting), running or stopped (last 7 days); omit for all |
{
"name": "list_servers",
"title": "My servers",
"description": "Your servers, newest first: running, starting and stopped in the last 7 days, with status, price and access.",
"inputSchema": {
"type": "object",
"properties": {
"status": {
"enum": [
"provisioning",
"running",
"stopped"
],
"type": "string",
"description": "Only servers in this state: provisioning (starting), running or stopped (last 7 days); omit for all"
}
},
"additionalProperties": false
},
"annotations": {
"readOnlyHint": true
}
}One server: status (provisioning, running, stopped), what it is doing now, cost so far, and once running the access: SSH command, Jupyter link with token, app URL.
| Parameter | Type | Description |
|---|---|---|
| id * | integer | Server id, as returned by launch_server or list_servers |
{
"name": "get_server",
"title": "Server status and access",
"description": "One server: status (provisioning, running, stopped), what it is doing now, cost so far, and once running the access: SSH command, Jupyter link with token, app URL.",
"inputSchema": {
"type": "object",
"required": [
"id"
],
"properties": {
"id": {
"type": "integer",
"description": "Server id, as returned by launch_server or list_servers"
}
},
"additionalProperties": false
},
"annotations": {
"readOnlyHint": true
}
}Stops a server and its billing. The disk is wiped; with save_workspace true the /workspace folder is saved first (storage is billed per GB-month) and comes back on the next launch.
| Parameter | Type | Description |
|---|---|---|
| id * | integer | Server id to stop, as returned by launch_server or list_servers |
| save_workspace | boolean | Save /workspace before stopping, default false |
{
"name": "stop_server",
"title": "Stop a server",
"description": "Stops a server and its billing. The disk is wiped; with save_workspace true the /workspace folder is saved first (storage is billed per GB-month) and comes back on the next launch.",
"inputSchema": {
"type": "object",
"required": [
"id"
],
"properties": {
"id": {
"type": "integer",
"description": "Server id to stop, as returned by launch_server or list_servers"
},
"save_workspace": {
"type": "boolean",
"description": "Save /workspace before stopping, default false"
}
},
"additionalProperties": false
},
"annotations": {
"idempotentHint": true,
"destructiveHint": true
}
}Balance in EUR, how long it lasts at the current spend, whether an SSH key is set, and limits.
No parameters.
{
"name": "get_account",
"title": "Balance and account",
"description": "Balance in EUR, how long it lasts at the current spend, whether an SSH key is set, and limits.",
"inputSchema": {
"type": "object",
"properties": {},
"additionalProperties": false
},
"annotations": {
"readOnlyHint": true
}
}VRAMfarm rents NVIDIA GPUs (RTX 3060 to B200) billed per minute in EUR from a prepaid balance. Typical flow: list_gpu_types to pick a card by VRAM and price, launch_server with a template (pytorch for code, comfy-* for images and video, chat* for a private chatbot, vllm for an OpenAI-compatible API, custom for any Docker image), poll get_server every 10–20 s until status is running, then use access (SSH command, Jupyter link, app URL). Stop with stop_server when the job is done: billing runs until then. Add an SSH public key in the dashboard settings before launching if you need SSH.
Names, descriptions and schemas were written by the server and are shown as data.