Skip to content
MCP server

supercomputer

Run GPU jobs on L4, A100 or H100 machines from a chat, billed per second

Install
Pricing
Paid
Platforms
API, Web
Runs
Remotely
Version
0.4.4
Listing updated
1 day ago

About supercomputer

supercomputer is a paid MCP server for cloud and infrastructure. It runs as a hosted remote server, so there is nothing to install locally.

Supercomputer lets an assistant pick a GPU, start a command, read its logs and stop it, with spend charged to your prepaid account by the second. Each run is one command on a fresh machine with a standard image containing CUDA, PyTorch, the Hugging Face stack and vLLM, and a plain Python image is available for CPU work. You ask for a GPU by name such as h100, 8x h100 or l4.

Spending guards are built in. Every run stops itself after max_minutes (default 60), so its maximum cost is that time at the hourly price, and run states the figure when it starts. Starting a job and cancelling one are the actions the assistant asks you about first. Identical name and command return the existing job instead of a duplicate.

Named persistent volumes can hold model downloads and results across jobs, for example mounted at the Hugging Face cache path. Inline scripts are limited to 64 KB, so larger work should clone a repo. Sign-in uses Google or email through OAuth, and credits start at $10. One MCP key per account can be revoked to disconnect every assistant.

Install

Run in a terminal

claude mcp add --transport http supercomputer https://supercomputer.sh/mcp

Add to Settings → Connectors → Add custom connector

https://supercomputer.sh/mcp
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "supercomputer": {
      "url": "https://supercomputer.sh/mcp"
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "supercomputer": {
      "type": "http",
      "url": "https://supercomputer.sh/mcp"
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "supercomputer": {
      "serverUrl": "https://supercomputer.sh/mcp"
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "supercomputer": {
      "type": "http",
      "url": "https://supercomputer.sh/mcp"
    }
  }
}

Tools this server provides (9)

Taken from the project documentation; the running server may differ.

list_gpus
GPU types, hourly prices and image contents.
get_balance
Signed-in account and available credit.
run
Start a command or script on a GPU.
get_job
Job status, run time and cost so far.
job_logs
Last lines of a job's output.
list_jobs
Recent jobs.
list_volumes
Persistent volumes on the account.
cancel_job
Stop a running or queued job.
buy_credits
Checkout link to add prepaid credit.

What it does

Per-second billing
Jobs are charged by the second against prepaid credit.
Automatic stop
Each run ends after max_minutes, default 60, capping cost.
Persistent volumes
Keep model downloads and results across jobs.
Job dedupe
Repeating the same name and command returns the existing job.
Live logs
Check job status and fetch recent log lines from the chat.
One-click revoke
Revoking the MCP key disconnects all assistants.

Good for

  • Run nvidia-smi on an L4 to check a machine
  • Fine-tune or serve a model with vLLM on an H100
  • Execute a repo's training script and read its logs
  • Cache Hugging Face models between jobs

Pricing

Paid

Pay per second from prepaid credit; top-ups from $10

Prices change. Check supercomputer.sh before you buy.

supercomputer FAQ

How is it billed?

By the second from prepaid credit, with top-ups from $10 and a saved card for auto-refill.

Can a job run up a huge bill?

Each run stops after max_minutes, default 60, so the cost is capped at that time at the GPU's hourly price.

Which GPUs are available?

The docs name L4, A100 and H100, including 8x configurations; list_gpus returns current types and prices.

How do I disconnect?

Revoke the account's mcp key in the dashboard.

Is supercomputer free?

No. supercomputer is a paid product.

How do I add supercomputer to Claude Code?

Run this command in your terminal: claude mcp add --transport http supercomputer https://supercomputer.sh/mcp

What are the best alternatives to supercomputer?

The closest alternatives to supercomputer are Claude Code Ultimate Guide, Agent Yurt, Proofline, signafAIde and Toolkien.

Alternatives to supercomputer

  1. MCP server for searching the Claude Code Ultimate Guide from any MCP client

    Runs locally
    • Open source
    • 6.5K GitHub stars
    • AI models and LLM tooling
  2. Agent Yurt

    MCP server

    Back up agent memory, files and dashboards, then recover from a reviewed snapshot

    Not checked yet
    • Remote
    • Free
    • Knowledge and memory
  3. Proofline

    MCP server

    Production-aware review of changes against your AWS or Google Cloud setup, via MCP

    • Remote
    • Freemium
    • Developer tools
  4. signafAIde

    MCP server

    Local MCP server to set up, verify and check coverage of signed agent action records

    Runs locally
    • 5 tools
    • Freemium
    • Security
  5. Toolkien

    MCP server

    One MCP gateway for all your MCP servers, with code mode, permissions and audit

    Not checked yet
    • 3 tools
    • Remote
    • Open source
    • AI models and LLM tooling
  6. filex

    MCP server

    Self-hosted file manager in one Go binary with a built-in MCP server for agents

    Not checked yet
    • Remote
    • Open source
    • Files and documents