Skip to content
shipstaticPublic

Repository files navigation

@shipstatic/ship

CLI and SDK for ShipStatic. Deploy static websites, landing pages, and prototypes instantly from the terminal or code. Free, no account needed.

Try it in seconds, no account needed

npx @shipstatic/ship ./dist

That's all. There's nothing to install and nothing to sign up for: if you have Node.js, you're ready. Point it at the folder your site is built into (./dist, ./build, ./out), or at a single file.

You get back a link you can share right away, and a claim link. The site stays up for 3 days, and opening the claim link keeps it for good.

Want it private? Add ShipStatic's built-in password protection, and visitors have to type the password before they see anything:

npx @shipstatic/ship ./dist --password "pick-a-password"

From code, it's one call:

import Ship from '@shipstatic/ship';

const result = await new Ship().deploy('./dist');
// result.deployment: where the site is live (happy-cat-abc1234.shipstatic.com)
// result.claim: open it to keep the site for good

Keep your sites with a free account

Everything above works with no account. A free account keeps your sites for good, lets you use your own domain, and shows you everything you've published in one place.

Get your free API key at my.shipstatic.com/api-key, then save it once:

npx @shipstatic/ship config    # paste your API key when asked
const ship = new Ship({ token: 'ship-your-api-key' });

Every command below works from then on. To drop the npx @shipstatic/ship prefix, install it once:

npm install -g @shipstatic/ship   # then use the `ship` command

As a project dependency: npm install @shipstatic/ship.

Every example below uses the bare ship command. Without the global install, start each one with npx @shipstatic/ship instead (or npx -y @shipstatic/ship where nothing can answer a prompt).

Deployments

ship ./dist                                        # Deploy (shortcut)
ship ./dist --domain www.example.com               # Deploy and serve it there
ship ./dist --ttl 1h                               # Expires in an hour
ship ./dist --label env:production --label v1.0.0  # Deploy with labels
ship deployments list
ship deployments list --limit 20                   # Page size; a hint shows the next cursor
ship deployments list --cursor <cursor>            # Continue from a previous page
ship deployments get <deployment>
ship deployments set <deployment> --label production
ship deployments delete <deployment>
ship.deploy(input, options?)               // Shortcut for deployments.upload()
ship.deployments.upload(input, options?)
ship.deployments.list(options?)            // { limit?, cursor? }; the response carries the next cursor
ship.deployments.get(deployment)
ship.deployments.set(deployment, { labels })
ship.deployments.delete(deployment)

Domains

ship domains set www.example.com                   # Reserve domain (no deployment yet)
ship domains set www.example.com <deployment>      # Link domain to deployment
ship domains set www.example.com --label prod      # Update labels only
ship domains get www.example.com
ship domains list                                  # --limit / --cursor paginate here too
ship domains validate www.example.com
ship domains verify www.example.com
ship domains records www.example.com
ship domains dns www.example.com
ship domains share www.example.com
ship domains delete www.example.com
ship.domains.set(name, { deployment?, labels? })   // Upsert: create, repoint, or label
ship.domains.get(name)
ship.domains.list(options?)                 // { limit?, cursor? }
ship.domains.validate(name)
ship.domains.verify(name)
ship.domains.records(name)
ship.domains.dns(name)
ship.domains.share(name)
ship.domains.delete(name)

domains.set() is a merge-upsert: omitted fields are kept on update, and defaulted on create. Once linked, a domain cannot be unlinked ({ deployment: null } → 400). Switch deployments or delete the domain instead.

Domain names are normalized by the API, so any case and Unicode are accepted:

ship.domains.set('WWW.Example.COM');   // → www.example.com
ship.domains.set('www.münchen.de');    // → Unicode supported

Tokens

ship tokens create --ttl 30d --label project:web  # Or 3600, 90s, 1h: one grammar
ship tokens list
ship tokens get <token>
ship tokens delete <token>
ship.tokens.create({ ttl?, labels? })
ship.tokens.list()
ship.tokens.get(token)
ship.tokens.delete(token)

Account

ship whoami
ship account get
ship config
ship ping
ship.account.get()            // → whoami
ship.ping()                   // → { timestamp } (server clock; reachability is the absence of a throw)
ship.getLimits()              // → the account's platform limits, read now

CLI reference

Composability

The -q flag outputs only the resource identifier, for piping and scripting:

ship tokens create -q is the one exception: it prints the token secret, which is shown once and never again.

# Deploy and link domain in one pipe
ship ./dist -q | ship domains set www.example.com

# Deploy and open in browser
open https://$(ship ./dist -q)

# Batch delete all deployments
ship deployments list -q | xargs -I{} ship deployments delete {} -q

Ephemeral deployments

ship ./dist --ttl 1h          # gone in an hour
ship ./dist --ttl 7d          # a week-long preview
ship ./dist --ttl 3600        # bare seconds work too

The platform reclaims the deployment when the time is up. Seconds, or a <n><unit> duration (s/m/h/d), the same grammar ship tokens create --ttl uses. Bounded at one year.

Two rules, both refused before anything uploads. It needs a token: an anonymous deployment already expires on the platform's own schedule, so there is no deployer to choose a different one. And it cannot be combined with --domain: a domain is a commitment and a deadline is its opposite, so the API refuses to point a domain at a deployment that expires.

To keep something longer, deploy it again. There is no way to extend a deployment's life, and no way to shorten it after the fact.

--domain is the same two calls as one command:

ship ./dist --domain www.example.com

Both spellings are supported and neither replaces the other. The pipe composes interactively: any two commands, wherever -q gives you the value the next one wants. --domain is one process, one exit code, and one --json, which is what CI needs: a run: block is bash -e without pipefail, so a pipeline reports only the last command's status and a failed deploy is masked. It answers as the domain, exactly as ship domains set does (DNS records and setup link included on a new external domain), and it needs a token, which it checks before uploading anything.

Shell completion

ship completion install
ship completion uninstall

Global flags

Available on every command:

Flag Description
--token <token> Any ship token: API key (ship-…) or deploy token (deploy-…)
--api-url <url> API URL override (for development)
--config <file> Custom config file path
--json Output results in JSON format
-q, --quiet Output only the resource identifier
--no-color Disable colored output
-h, --help Display help for command
-V, --version Show version information

Deploy flags

Available on ship <path> and ship deployments upload:

Flag Description
--domain <domain> Serve this deployment at that domain: creates or repoints it. Needs a token
--label <label> Add label, a word or key:value (repeatable)
--password <password> Password-protect this deployment (6–128 chars)
--ttl <duration> Expire this deployment after that long: 3600, 90s, 30m, 1h, 7d. Needs a token; cannot be combined with --domain
--no-path-detect Disable automatic path optimization
--no-spa-detect Disable automatic SPA detection

CLI environment variables

Var Purpose
SHIP_TOKEN Default for --token
SHIP_API_URL Default for --api-url
SHIP_PASSWORD Default for --password (empty string normalized to absence)

SDK reference

Authentication

// No token: deploy only, with no account; the site comes with a claim link and stays up for 3 days
const ship = new Ship();

// API key: durable, full account
const ship = new Ship({ token: 'ship-your-api-key' });

// Deploy token: scoped to deploys, optional TTL, revocable
const ship = new Ship({ token: 'deploy-your-token' });

// OAuth access token: delegated, short-lived, sent verbatim
const ship = new Ship({ token: accessToken });

// Token provider: invoked per request; refresh lives with you
const ship = new Ship({ token: () => mintToken() });

// Cookie session: the platform's own apps, where every account request names its account
const ship = new Ship({ session: true });
ship.setHeaders({ 'X-Account': account });

// Set or rotate the token after construction
ship.setToken('ship-your-api-key');

Retries

Failed requests are retried automatically: transport failures (including a timeout) and 500/502/503/504, twice by default, with full-jitter exponential backoff. maxRetries is the knob; 0 disables it.

const ship = new Ship({ token: 'ship-your-api-key', maxRetries: 5 });

Deliberately never retried: a maintenance 503 (its message says when to come back), 429 (the rate limiter has answered), PUT/DELETE (a repeat can misreport a lost success as a failure), anything stopped by a signal you supplied, and any other request that is not safe to send twice. A deploy is safe to send twice when it carries an idempotencyKey: the API answers with the deployment that key already made, within your account, instead of making a second one.

timeout is the ceiling on one ATTEMPT. For a hard overall deadline pass your own signal (AbortSignal.timeout(ms)), which is never retried past.

Deploy options

ship.deploy(input, {
  labels?: string[],
  password?: string,          // Password-protect the deployment (6–128 chars)
  ttl?: number,               // Seconds until it expires (needs a token; max 1 year)
  signal?: AbortSignal,       // Abort to cancel the deploy
  pathDetect?: boolean,       // Auto-optimize paths (default: true)
  spaDetect?: boolean,        // Auto-detect SPA (default: true)
  via?: string,               // Client identifier
});

Large deployments

There's nothing to set up. A site too big for one request goes in several: the SDK sends a list of every file first, then the files themselves in batches, and the site goes live when the last one arrives. You get the same answer as for a small site. Every batch is safe to retry, so a dropped connection costs you only that batch.

Expiring deployments

Pass ttl in seconds and the platform reclaims the deployment when the time is up, from 1 second to one year. The wire carries the duration and the API stamps expires against its own clock, so the answer says when:

const result = await ship.deploy('./dist', { ttl: 3600 });
// result.expires → unix seconds, one hour after result.created

Needs a credential: a deployment with no account already expires on the platform's own schedule. And a deployment carrying a ttl cannot be linked to a domain: the API refuses, which is what stops a domain pointing at something that is about to be reclaimed. There is no way to extend or shorten a deployment after the fact; redeploy instead.

Password protection

Pass password (6–128 characters) to gate the deployment behind a prompt. Visitors are asked for the password before they can view the site, including on any custom domains pointing at it. To remove protection, redeploy without a password.

ship --password 'your-passphrase' ./dist
await ship.deploy('./dist', { password: 'your-passphrase' });

The CLI also reads SHIP_PASSWORD from the environment when --password is not given.

Browser usage

import Ship from '@shipstatic/ship';

const ship = new Ship({ token: 'ship-your-api-key' });

// From file input
const deployment = await ship.deploy(fileInput.files);

// From StaticFile array
const deployment = await ship.deploy([
  { path: 'index.html', content: new Blob(['<html>…</html>']) }
]);

Events

ship.on('request', (url, init) => {});          // once per attempt
ship.on('retry', (error, url, attempt) => {});  // an attempt failed, another is coming
ship.on('response', (response, url) => {});     // the call succeeded
ship.on('error', (error, url) => {});           // the call failed, terminally
ship.off('request', handler);

One call emits retry* (error | response): every failure is announced, and the event name says whether it ended the call. attempt counts from 1, so it names both the attempt that failed and which retry is happening.

Custom fetch

Pass fetch to override the transport function used for every API call. Defaults to globalThis.fetch. Useful for wrapping requests with tracing, retries, or request signing, and for injecting a Cloudflare service-binding Fetcher from a Worker so calls reach a sibling Worker in-process instead of through the public hostname.

This is also the seam for corporate proxies: Node's built-in fetch ignores HTTP(S)_PROXY environment variables, so behind a proxy inject a proxy-aware transport (e.g. undici's EnvHttpProxyAgent as the dispatcher, or Node 24+'s NODE_USE_ENV_PROXY=1).

import type { Fetch } from '@shipstatic/ship';

const traced: Fetch = (input, init) =>
  globalThis.fetch(input, { ...init, headers: { ...init?.headers, 'X-Trace-Id': 'abc-123' } });

const ship = new Ship({ fetch: traced });
// Cloudflare Worker with a service binding to the API.
// Any parseable apiUrl works: service bindings dispatch by binding identity, not hostname.
const ship = new Ship({
  apiUrl: 'https://api',
  fetch: env.API.fetch.bind(env.API),
});

Error handling

import { isShipError, ErrorType } from '@shipstatic/types';

try {
  await ship.deploy('./dist');
} catch (error) {
  if (isShipError(error)) {
    error.isAuthError();        // semantic category
    error.isNetworkError();     // semantic category: nothing was exchanged
    error.isClientError();      // semantic category (Business | Config | File | Validation)
    error.type === ErrorType.Validation;  // specific-type check
    error.type === ErrorType.Timeout;     // a deadline expired, inside isNetworkError()
    error.status === 429;       // status check
  }
}

Configuration

The CLI (ship) resolves its token in this order:

  1. CLI flag: --token
  2. Environment variable: SHIP_TOKEN
  3. Config file: ~/.shiprc (run ship config to create one)

--config <file> reads any path you name instead of ~/.shiprc, which is how per-environment configs work (ship --config dev.shiprc ...). The file is strict JSON; an empty one means "no config".

No repository file is ever read. A .shiprc or package.json "ship" key in your working directory is ignored: cloning a repo can never change which account you deploy to, or which host your token is sent to.

The SDK (new Ship(...)) resolves its token in this order:

  1. Constructor option: new Ship({ token })
  2. Environment variable: SHIP_TOKEN

--api-url / SHIP_API_URL / apiUrl resolve the same way for the API endpoint.

The SDK never reads .shiprc or package.json: file resolution is a CLI feature, not an SDK feature. This keeps new Ship({}) safe to use from embedded contexts (MCP, n8n, library wrappers) without inheriting the host developer's personal credentials.

SHIP_TOKEN=ship-your-api-key ship deployments list

TypeScript

import type { ShipClientOptions, DeploymentOptions, ShipEvents } from '@shipstatic/ship';
import type { Deployment, Domain, Account, StaticFile } from '@shipstatic/types';

AI agents

ShipStatic's skill for agents, publish-online, teaches an agent to publish a folder with this CLI, with no account needed. Coding agents get it with the ShipStatic plugin; any skills-aware tool can take it as one file from www.shipstatic.com/SKILL.md.

Also available

Surface Reach it
MCP Drop https://mcp.shipstatic.com into any MCP client
VS Code Search "ShipStatic" in the Marketplace
Plugin Install it in Cursor, Claude Code, Codex or Gemini CLI
n8n Search "ShipStatic" in n8n's node panel
GitHub Action shipstatic/action@v3
Agent Skill npx skills add shipstatic/plugin

License

MIT

Releases

Packages

Used by

Contributors

Languages