# rent.dev > rent.dev rents Linux sandboxes to AI agents: one call for an account, USDC or a pay link to fund it, per-second billing, over REST, MCP, the E2B SDK or the rent CLI. ## What do you want? | Want | On rent.dev | Start with | | --- | --- | --- | | A VPS, server, VM, virtual machine, dev box, cloud machine or container to use | A sandbox: a Linux machine billed by the second that ends at its timeout (timeout_s, default 300 seconds, at most 86400). | REST: `POST https://api.rent.dev/v1/accounts`, then `POST /v1/sandboxes {"timeout_s": 3600}`, then `POST /v1/sandboxes/{id}/exec {"cmd": "…"}`. MCP: `rent_create_sandbox {"timeout_s": 3600}`, then `rent_exec`. CLI: `rent init`, then `rent sbx create --timeout 3600`, then `rent sbx exec -- `. | | A GPU | A GPU pod on your own RunPod key: RunPod bills that key and no rent.dev balance is used. | REST: `POST /v1/byok/credentials {"vendor": "runpod", …}`, then `POST /v1/gpu-pods`. MCP: `rent_set_byok_credential`, then `rent_create_gpu_pod`. CLI: `rent byok add runpod`, then `rent gpu create --gpu-type --image `. | | Prices | The price of one task on rent.dev and at each vendor with published rates, and GPU prices per hour, with no key. | REST: `GET https://api.rent.dev/v1/quotes`. MCP: `rent_quote`. CLI: `rent quote`. | | Your own VPS, server or VM on rent.dev, to earn | A machine you run, your own or one rented elsewhere, hosts rent.dev sandboxes as a node, for a share of what they bill. | REST: `GET https://api.rent.dev/v1/supply/guide`. MCP: `rent_supply_guide`. | A machine to use is a sandbox; a machine you run that should earn is a node. If the request does not say which, ask the person; if you cannot ask, take the sandbox, which costs cents and ends at its timeout, then say that you did and name the other reading. Over REST, send the key as `Authorization: Bearer rent_sk_…` and create with `Prefer: wait=30`; without it, or past the wait, the create answers 202 with an Operation to poll at `GET /v1/operations/{id}`, where {id} is the `op_…` that ends its `name`. A new account has a zero balance unless a promotion gives it promotional credit, and a create the balance cannot cover answers 402 with `topup.x402_url`, which an agent pays in USDC, and `topup.checkout_url`, a pay link for a person. MCP is at `https://mcp.rent.dev/mcp`, which connects without a key. Its account tools take a key in the client's config or the client's own sign-in, which it starts when one answers 401; with neither, `rent_signup` answers a key for REST. The CLI is the npm package `rentdev`, binary `rent`. On rent.dev's own machines a sandbox is a Firecracker microVM; a partner's machine declares its own isolation, which may be a container. By default (`active`) a sandbox may run on a verified partner's machine, billed from the same balance; `routing.partners` set to `none` keeps it on rent.dev's own machines and your own vendor keys. The operator of a partner machine can read what runs on it. Signup is one call: `curl -sX POST https://api.rent.dev/v1/accounts` takes no credentials and no body, and answers an API key that is shown once. A sandbox ends at its timeout, which timeout_s sets at create. POST https://api.rent.dev/v1/sandboxes/{id}:extend sets a new timeout from now, up to the maximum timeout_s. At the timeout it is stopped, or paused if it was created with auto_pause. POST /v1/sandboxes/{id}:pause keeps its memory and disk, and rent.dev bills nothing while it is paused. POST /v1/sandboxes/{id}:resume starts it again with a new timeout, 300 seconds unless you send timeout_s. No call opens an SSH session. POST /v1/sandboxes/{id}/exec runs a command and waits for it, up to its timeout_s (default 50 seconds). Work that outlasts one call runs as a process: POST /v1/sandboxes/{id}/processes. A port is reached at the HTTPS URL that GET /v1/sandboxes/{id}/ports/{port}/url answers, while the sandbox runs and something listens on the port. Tell the person the sandbox id (lse_…), how to run a command in it (POST /v1/sandboxes/{id}/exec, or rent sbx exec -- ), when it ends (timeout_at), what it has cost and holds (captured_micros, held_micros), and where its API key is kept. ## Docs - [Agent quickstart](https://rent.dev/docs/agents.md): Create an account, fund it and run a sandbox over REST, with no person in the loop. - [MCP server docs](https://rent.dev/docs/mcp.md): rent.dev's MCP server: two endpoints, their tools with scopes and hints, and how payments work inside a tool call. - [REST API](https://rent.dev/docs/rest.md): Every route of the rent.dev API with its parameters, body fields and status codes. - [E2B compatibility](https://rent.dev/docs/e2b.md): Run the unmodified E2B SDK against rent.dev: setup, the compatibility matrix and the known gaps. - [CLI](https://rent.dev/docs/cli.md): The rent CLI: running it, signing in, every command and its flags, exit codes and environment variables. - [Payments](https://rent.dev/docs/payments.md): How a rent.dev balance is funded: Checkout and pay links, auto-recharge, x402 USDC, wallet-first signup and the fees. - [Bring your own key](https://rent.dev/docs/byok.md): What runs on rent.dev's balance and what needs your own vendor key: GPU pods, and sandboxes at E2B Cloud or Vercel. - [Provide compute](https://rent.dev/docs/supply.md): How to sell sandbox time on your own machines through rent.dev: the two ways to bring a node, the commands, the checks, and how rates, the take and the holdback work. - [Pricing](https://rent.dev/docs/pricing.md): How a rent.dev sandbox is priced: provisioned memory plus active CPU, metered at no more than E2B's list price, a top-up fee of 5.5% (at least $0.80) by card or 5.5% over x402, and worked examples. - [Limits](https://rent.dev/docs/limits.md): The limits that apply to rent.dev sandboxes and GPU pods today: the current node's cap on running sandboxes and the vendor limits passed through. - [Errors](https://rent.dev/docs/errors.md): Every Problem code the rent.dev API answers, its HTTP status and what to do about it. ## Prices and status - [Price board](https://rent.dev/prices.md): What one task costs on rent.dev's own sandboxes and on every vendor with published rates, plus per-GPU hourly rates. - [Price board as JSON](https://api.rent.dev/v1/quotes): The same board. It needs no key. - [Catalog](https://rent.dev/catalog): Search every provider price the rent.dev catalog may compare: GPUs, CPUs, hosted models and other rates, each with its source and date. - [Status](https://rent.dev/status.md): Live health of rent.dev: its endpoints, their free slots and the open alerts. - [Status as JSON](https://api.rent.dev/v1/status): The same answer. It needs no key. - [Uptime and incidents](https://status.rent.dev/status.md): One go word, each component's state and the open incidents, as Markdown. Read it before starting work. The JSON is `https://status.rent.dev/api/v2/summary.json` (Statuspage v2). The page itself, `https://status.rent.dev`, shows uptime over the past 90 days. ## API and MCP - [OpenAPI document](https://api.rent.dev/openapi.json): OpenAPI 3.1: every REST route with its schemas. - [MCP server](https://mcp.rent.dev/mcp): 35 tools for the provider catalog, funds, sandboxes, GPU pods, the account and partner supply. The catalog gives each provider's capabilities, agentic status and prices, and a listed provider may be unpriced. It connects without credentials and runs the tools that need no account; the others take an API key or an OAuth access token, and answer 401 with a sign-in challenge without one. - [Public MCP server](https://mcp.rent.dev/mcp/public): No credentials: `rent_quote`, `rent_list_providers`, `rent_search_docs`, `rent_signup`, `rent_supply_guide`. ## For agents - [Setup skill](https://rent.dev/SKILL.md): Setup and the first verified run, written for an agent. - [Authentication](https://rent.dev/auth.md): API keys, agent keys, wallets and OAuth. - [Every docs page in one file](https://rent.dev/llms-full.txt): One heading per section, each with its own address.