Lium GPU Cloud — Claude Code MCP server: rent GPU pods
Lets your agent rent a GPU pod on Lium, run a command on it and tear it down. Billing runs per second.
claude mcp add lium -e LIUM_API_KEY=<LIUM_API_KEY> -- uvx lium.ioWhat Lium GPU Cloud does
The server wires an agent into the Lium platform, where NVIDIA GPU machines are rented out. Through MCP tools the agent picks a node, executes commands on it and removes the pod once the job is done. Billing is per second and keeps running until removal, so a pod left behind keeps costing money.
Under the hood sits the lium.io Python package, launched with uvx. The same package ships a CLI and an SDK. The CLI covers lium ls for listing nodes, lium quote for pricing a job, lium up to rent, lium scp to copy files over and lium ssh to get a shell. On the SDK side there is the @lium.machine decorator, which pushes a function's execution onto a GPU pod, and the Lium() client for longer orchestration code.
A LIUM_API_KEY is required — it goes into the environment when you register the server. You create the key in your Lium account. Without it the server cannot authenticate.
What it helps with
- rent a GPU pod for a specific job
- run a command on a remote GPU machine
- list available nodes and their hourly price
- copy a script onto a pod and execute it
- remove a pod to stop per-second billing
Example prompts for the agent
- Rent an A100 pod and tell me the hourly price
- Run
nvidia-smion my pod and show me the output - Remove pod
swift-fox-c8, I'm done with it
How to install Lium GPU Cloud
Platform CLI
claude mcp add lium -e LIUM_API_KEY=<LIUM_API_KEY> -- uvx lium.ioReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from lium.io <command>
Platform CLI
codex mcp add lium-gpu-cloud --env LIUM_API_KEY=<LIUM_API_KEY> -- uvx lium.ioReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from lium.io <command>
Config file
{
"mcpServers": {
"lium-gpu-cloud": {
"command": "uvx",
"args": [
"lium.io"
],
"env": {
"LIUM_API_KEY": "<LIUM_API_KEY>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from lium.io <command>
Config file
{
"servers": {
"lium-gpu-cloud": {
"type": "stdio",
"command": "uvx",
"args": [
"lium.io"
],
"env": {
"LIUM_API_KEY": "<LIUM_API_KEY>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from lium.io <command>
Platform CLI
gemini mcp add -e LIUM_API_KEY=<LIUM_API_KEY> lium-gpu-cloud uvx lium.ioReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from lium.io <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Lium GPU Cloud
How do I install Lium GPU Cloud in Claude Code?
Run claude mcp add lium -e LIUM_API_KEY=<LIUM_API_KEY> -- uvx lium.io. Alternatively use claude mcp add --env LIUM_API_KEY=<LIUM_API_KEY> --transport stdio lium-gpu-cloud -- uvx lium.io.
What does Lium GPU Cloud do?
It lets an agent rent GPU pods on the Lium platform, run commands on them and remove them. Billing is per second for as long as the pod exists.
What do I need to run the server?
A LIUM_API_KEY, passed as an environment variable when adding the server. You generate the key in your Lium account.
