Lium Docs
Measured conclusion: Alive, not deep-tested. L2 outcome: not_attempted.
Search and read the Lium GPU rental docs: pods, CLI, SDK, REST API and agent guides.
| Author | io.github.Datura-ai |
|---|---|
| Access | remote |
| Publisher key required | No |
| Measured state | Alive, not deep-tested |
| L1 liveness | alive |
| L1 latency | 2558 ms |
| Last measured | 2026-10-26 |
Add to your agent
Copy this into your MCP client config:
{
"mcpServers": {
"Lium Docs": {
"url": "https://docs.lium.io/mcp",
"transport": "streamable-http"
}
}
}
Measured tools (11)
searchread-onlySearch the Lium documentation for pages matching the query. Use for how-to questions (renting, SSH, templates, providing GPUs); for live GPU availability or prices use search_gpus or get_gpu_prices instead. Optionally filter by audience (providers, validators, pod-users, developers); without one, pages for the other side (renter or provider, guessed from the query) rank lower.
Parameters:
query(required) — string — Search querylimit— integer — Maximum number of results to return (default 10, max 50)audience— string — Filter results to a specific audience section
Hints: read-only yes · destructive no · idempotent not declared · open-world no
read_pageread-onlyReturn the raw Markdown source for a documentation page. Accepts a full URL, absolute slug (e.g. /providers/quickstart), or basename (e.g. quickstart). On tie among basenames, prefers pages in audience order: Providers > Validators > Renters > Developers.
Parameters:
url_or_slug(required) — string — Full URL (https://docs.lium.io/providers/quickstart), absolute slug (/providers/quickstart), or basename (quickstart)
Hints: read-only yes · destructive no · idempotent not declared · open-world no
search_gpusread-onlyList GPU nodes that can be rented on Lium right now, cheapest per GPU-hour first, with live USD prices and a rent link for each. Use when the user wants a specific machine (GPU model, count, country, price cap); use get_gpu_prices for a price overview. Each node has tier (secure or spot), reliability_score and min_gpus_per_rental (some nodes rent only whole). Read-only: next_step says how to rent one.
Parameters:
gpu— string — GPU model, e.g. H100, H200, B200, RTX 4090 (case and spacing ignored)count— integer — GPUs needed on one nodelimit— integer — Max nodes to return (default 10, max 50)country— string — Two-letter country code, e.g. US, DEmax_price_per_gpu_hour— number — Price cap in USD per GPU-hour
Hints: read-only yes · destructive no · idempotent not declared · open-world no
get_gpu_pricesread-onlyLive price range (min, median, max USD per GPU-hour) and free GPU count for each GPU model rentable on Lium right now. Use for "what does an H100 cost" or comparing models; use search_gpus to pick a node to rent. Pass count for multi-GPU quotes: a 1-GPU price can come from a node with one free GPU.
Parameters:
gpu— string — GPU model, e.g. H100, H200, B200, RTX 4090 (case and spacing ignored)count— integer — GPUs needed on one node: prices only nodes that can rent that manycountry— string — Two-letter country code, e.g. US, DE
Hints: read-only yes · destructive no · idempotent not declared · open-world no
estimate_provider_earningsread-onlyEstimate what a provider earns per month on Lium by listing GPUs of a model, as the rewards calculator does: rental income (95% of the renter price) and rented extra at the last 30 days' utilisation, plus idle pay at the validator's rate for every unrented hour; with what those GPUs actually earned and the middle-half range across providers. Use when someone asks what their GPUs could earn on Lium; for renting GPUs use search_gpus. An estimate, not a guarantee.
Parameters:
gpu(required) — string — GPU model, e.g. H100, B200, RTX 4090 (case and spacing ignored)count— integer — How many GPUs of that model (default 1)
Hints: read-only yes · destructive no · idempotent not declared · open-world no
rent_podRent a GPU pod on Lium with your API key and wait until it runs; returns the pod id and the ssh command. Spends money from your Lium balance: the pod is billed per second until remove_pod, also when idle. Give gpu (and count, optional max_price_per_gpu_hour and country) to let Lium pick the cheapest matching node, or node_id from search_gpus alone for that node. idempotency_key is required: if a call fails or times out, retry with the same key within 24 hours so it cannot rent twice.
Parameters:
idempotency_key(required) — string — A new unique string per rental (e.g. a UUID). Send the same key again within 24 hours to retry the same rental: it never rents twicegpu— string — GPU model, e.g. H100, RTX 4090; Lium picks the cheapest node that fitsname— string — Pod namecount— integer — GPUs on one node (default: 1 with gpu, every free GPU with node_id)country— string — Two-letter country code (with gpu, not node_id)node_id— string — A node id from search_gpus, to rent that nodetemplate_id— string — From list_templates; default: Lium's PyTorch image for the nodewait_seconds— integer — How long to wait for the pod to run (default 25, max 50)ssh_public_key— string — OpenSSH public key to log in with (e.g. ~/.ssh/id_ed25519.pub); default: the keys saved on your Lium accountmax_price_per_gpu_hour— number — Price cap in USD per GPU-hour (with gpu, not node_id)
Hints: read-only no · destructive no · idempotent no · open-world yes
list_podsread-onlyYour Lium pods with status, ssh command and price per hour.
Hints: read-only yes · destructive no · idempotent not declared · open-world no
get_podread-onlyOne pod: status, ssh command (once RUNNING), price per hour and spend so far.
Parameters:
pod_id(required) — string — Pod id from rent_pod or list_pods
Hints: read-only yes · destructive no · idempotent not declared · open-world no
remove_poddestructiveRequest pod deletion. Billing stops when accepted; deletion may finish asynchronously (get_pod shows DELETING). Data on the pod is lost.
Parameters:
pod_id(required) — string — Pod id from rent_pod or list_pods
Hints: read-only no · destructive yes · idempotent yes · open-world yes
get_balanceread-onlyYour Lium balance in USD.
Hints: read-only yes · destructive no · idempotent not declared · open-world no
list_templatesread-onlyDocker templates a pod can start from (official ones, your own and other users' public ones; official and yours mark which), for rent_pod template_id.
Parameters:
search— string — Filter by name or image, e.g. pytorch, vllm
Hints: read-only yes · destructive no · idempotent not declared · open-world no
Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.