CLI

Usage: npx @reduck-ai/cli@latest [options] [command]

Run Reduck scripts against your browser, via the Reduck MCP.

Options:
  -V, --version               output the version number
  -h, --help                  display help for command

Commands:
  login                       Authenticate against the Reduck MCP (OAuth2 PKCE).
  logout                      Revoke and remove saved credentials for the active
                              MCP target.
  list|ls [options] [handle]  List scripts. Without a handle, searches across
                              your own, your projects, and the official
                              catalogue. With a handle, scopes to that owner.
  read [options] <script>     Read a saved script's contract (loggedIn +
                              input/output JSON Schemas) via the MCP's
                              read_script — the arg names to pass to `run`.
  skill [name]                Print Reduck skill docs to stdout. Public — no
                              login required.
  whoami                      Show which Reduck account the CLI is authenticated
                              as.
  run [options]               Run one or more saved scripts via the MCP's
                              run_script — on your browser, via the paired
                              extension. Repeat --script to run several at once,
                              each on its own browser.
  help [command]              display help for command

Some commands are experimental and hidden from this list. Set REDUCK_EXPERIMENTAL=1 to show them.

Usage: npx @reduck-ai/cli@latest list|ls [options] [handle]

List scripts. Without a handle, searches across your own, your projects, and the
official catalogue. With a handle, scopes to that owner.

Arguments:
  handle                     Optional owner scope: "@user" (with @) for a user,
                             or a project handle (without @) — e.g. "reduck" for
                             the official catalogue. Omit for a cross-owner
                             search.

Options:
  -q, --query <q>            Search across name, slug, host, description and the
                             input/output schemas. Ranks by relevance.
  --host <host>              Filter by host (e.g. linkedin.com).
  --visibility <visibility>  Visibility filter: "public" or "private". Omit to
                             show both. (choices: "public", "private")
  --archived <filter>        Archive filter: "exclude" (default) hides archived,
                             "only" shows archived exclusively, "include" shows
                             both. (choices: "exclude", "include", "only",
                             default: "exclude")
  -h, --help                 display help for command

Examples:
  $ npx @reduck-ai/cli@latest list
  $ npx @reduck-ai/cli@latest list -q linkedin
  $ npx @reduck-ai/cli@latest list official
  $ npx @reduck-ai/cli@latest list @myhandle -q scrape
  $ npx @reduck-ai/cli@latest list myproject --host linkedin.com

Usage: npx @reduck-ai/cli@latest read [options] <script>

Read a saved script's contract (loggedIn + input/output JSON Schemas) via the
MCP's read_script — the arg names to pass to `run`.

Arguments:
  script             Script address as printed by `list`:
                     [<handle>/]<host>/<slug>. Bare host/slug = your own; a
                     handle prefix (@user or a project like reduck for the
                     official catalogue) = that owner's public script.

Options:
  --code             Also print the raw script body (the contract — description,
                     loggedIn, input/output schemas — is always shown).
  --version-id <id>  Read a specific version instead of the current one (from
                     list_script_versions).
  -h, --help         display help for command

Examples:
  $ npx @reduck-ai/cli@latest read cfspro.impots.gouv.fr/list_avis_cfe
  $ npx @reduck-ai/cli@latest read reduck/linkedin.com/get_post --code

Usage: npx @reduck-ai/cli@latest run [options]

Run one or more saved scripts via the MCP's run_script — on your browser, via
the paired extension. Repeat --script to run several at once, each on its own
browser.

Options:
  --script <addr...>      A script to run: its address
                          ([<handle>/]<host>/<slug>) followed by its key=value
                          args, e.g. `--script linkedin.com/get_post url=…
                          max=5`. Bare addr = your own; a handle prefix (@user,
                          or a project like reduck for the official catalogue) =
                          that owner's public script. Arg values are coerced to
                          the script's type — plain `siret=913…` stays a string,
                          pass JSON for arrays/objects (`ids=[1,2]`). Bind a
                          file input with `file:<input>=<path>` (e.g.
                          `file:attachment=./banner.png`) — the file is uploaded
                          with the run, and the page gets it under the input's
                          name, so pass the script's own name argument if it has
                          one (send_email: attachmentName). Repeat --script to
                          run several at once, each on its own browser (see
                          --sequential to chain them instead). (default: [])
  --sequential            Chain the --script list through ONE browser, in order,
                          instead of running them at once: each script inherits
                          the previous one's cookies, page and navigation (not
                          its result). Stops at the first failure. Use it when a
                          script needs the state another leaves behind (log in,
                          then act).
  --file <input=path...>  File input (repeatable): bind a declared file input (a
                          property in the script's input schema marked { format:
                          "file" }) to a local file, e.g. `--file
                          attachment=./banner.png`. The file is uploaded with
                          the run, up to 50 MB, so it works on your paired
                          browser and on a cloud browser alike. The page gets it
                          under the input's name, not the file's.
  --device <id>           Target a specific registered device (id from
                          `list_devices`) instead of auto-picking the paired
                          extension.
  --out <dir>             Directory to deliver downloaded files into, under
                          their real names (default: ~/Downloads).
  --version-id <id>       Run a specific version instead of the current one
                          (from list_script_versions). Only for single-script
                          runs.
  -h, --help              display help for command

Runs on your browser via the paired extension — the MCP auto-picks it,
or pass --device <id> (see list_devices) to choose. Prints the script's result as
JSON on stdout (pipe it to jq); the run's step trace is read separately, through the
MCP's read_run_trace({runId}).
Repeat --script to run several at once, each on its own browser — stdout is then a
JSON array, one result per script, in the order given, and one failing does not stop
the others. Add --sequential to chain them through one browser instead (in order,
stopping at the first failure — the rest then report that they never ran, so the array
still has one entry per script). Exit code 0 iff every script succeeded. Set
REDUCK_MCP_URL to target another MCP (staging, a local dev server) —
`npx @reduck-ai/cli@latest login` once per target.

Examples:
  $ npx @reduck-ai/cli@latest run --script linkedin.com/get_profile_info username=dhuynh95
  $ npx @reduck-ai/cli@latest run --script reduck/linkedin.com/get_post url=…   # official catalogue
  $ npx @reduck-ai/cli@latest run --script news.ycombinator.com/top --device 4f3c…   # a specific device
  $ npx @reduck-ai/cli@latest run --script reduck/mail.google.com/send_email [email protected] subject=Hi attachmentName=banner.png file:attachment=./banner.png   # file input
  $ npx @reduck-ai/cli@latest run --script site.com/a --script site.com/b   # both at once, a browser each
  $ npx @reduck-ai/cli@latest run --sequential --script site.com/login --script site.com/scrape max=5   # chain, one browser
  $ REDUCK_MCP_URL=http://localhost:7575 npx @reduck-ai/cli@latest run --script example.com/title