CLI reference (ait)

ait is a single-file, zero-dependency CLI (Node 18+ or Bun) that speaks Automate It’s MCP server — everything a remote MCP client can do, headlessly. Every command goes through MCP; there is no separate REST path.

Install

npm install -g automate-it   # provides the global `ait` command

Runtimes that read AgentSkills folders (OpenClaw, Hermes) can instead install the whole skill folder — SKILL.md plus the same CLI — see the connect guides.

Configure

ait needs a scoped API key. Mint one in the Automate It web app: click your avatar at the bottom of the sidebar to reach Profile (https://automate.it.com/app/<workspace-slug>/settings/profile), then Developer API Keys → Create Key. The ak_… secret is shown once — see API keys & scopes for which scopes to tick. ait reads its configuration from the environment. Put the key in a mode-600 env file and load that — don’t export it in an interactive shell, where it lands in your history:

# ~/.automate-it.env   (chmod 600)
AUTOMATE_IT_API_KEY=ak_...                        # required
AUTOMATE_IT_WORKSPACE=<workspace-id>              # optional; auto-resolved if the key sees one workspace
AUTOMATE_IT_API_URL=https://api.automate.it.com   # optional; this is the default

set -a; . ~/.automate-it.env; set +a   # or your process manager's EnvironmentFile
ait workspaces

If an agent runtime drives the CLI, use that runtime’s credential store instead of a file you maintain by hand — on OpenClaw the key goes in the shared secret store behind a secret reference. Never paste the key into a chat message: transcripts get persisted to memory files and swept into Git backups. If it has ever appeared in a chat, ticket, log, or commit, rotate it under Profile → Developer API Keys. Every workspace command also accepts --workspace <id>. Errors print {"error": "..."} to stderr and exit 1.

Commands

Tasks — the review-gate loop

Command	Does
`task create --title <t> [--instructions <text>] [--output-types x,linkedin,…] [--no-review] [--claim] [--skills <names or ids>] [--assign <userId\|me>]`	Create a task. `--claim` = claimed and assigned to you; `--assign` hands it to a member
`task submit --title <t> --body <text> [--type <contentType>] [--media '<json>']`	One-shot: create with finished content, straight into review
`task list [--status <s>] [--mine]`	Statuses: `todo working review approved publishing published deleted failed`; `--mine` = assigned to you
`task get <taskId>`	Full detail, including reviewer `comments`
`task update <taskId> [--title] [--instructions] [--output-types x,…] [--publish-mode …] [--publish-at] [--clear-publish-at] [--review] [--no-review]`	Patch task fields (admin keys). Only provided flags change; `--review` turns the gate back on; `--clear-publish-at` drops a scheduled time
`task links <taskId>`	Published post URLs
`task claim-next`	Atomically claim the oldest unassigned `todo` task
`task claim <taskId>`	Claim a specific task (assigns it to you)
`task add-content <taskId> --items '<json array>'`	Append several content items in one call (each entry takes `body`, `contentType`, `title`, `media`, `sortOrder`)
`task add-content <taskId> --type <ct> [--body <text>] [--media '<json>'] [--sort-order <n>]`	Append one content item
`task update-content <taskId> <contentItemId> [--body] [--title]`	Revise in place
`task delete-content <taskId> <contentItemId> --yes` / `task clear-content <taskId> --yes`	Remove one / all content items
`task complete <taskId>`	Finish work (→ review, or auto-approve)
`task comment <taskId> --comment <text>`	Leave a note for the reviewer
`task next-review`, `task approve <id>`, `task reject <id> [--comment]`, `task publish <id>`, `task delete <id> --yes`	Reviewer/admin operations

Skills, files, automations

Command	Does
`skills list` / `skills get <skillId>`	Workspace voice & formatting rules + bundled file ids
`skills create --name <n> [--description] [--instructions] [--file-ids a,b] [--folder-ids a,b]`	Create a skill (instructions capped at 24,000 characters; name unique per workspace)
`skills update <skillId> [--name] [--description] [--instructions] [--file-ids] [--folder-ids]`	Change only the fields you pass
`skills delete <skillId> --yes`	Delete a skill and its memory — refused while an automation or active task still references it
`files list [--search] [--mime-type]`	Browse workspace files
`files download-url <fileId>`	Presigned URL — fetch the bytes yourself
`upload-url --filename <f> --mime-type <m>`	Presigned upload URL for media
`files move <fileId> [--folder <id>]` / `files copy <fileId> [--folder <id>]`	Relocate or duplicate a file; omit `--folder` for the workspace root
`folders list [--parent <id>] [--root]`	Browse the folder tree
`folders create --name <n> [--parent <id>]` / `folders rename <id> --name <n>`	Organize files
`folders delete <folderId> --yes`	Delete a folder and everything inside it
`automation create/list/get/update`	Standing briefs that spawn tasks on a schedule
`automation delete <id> --yes`	Delete an automation and every task it spawned, published ones included (Admin only)
`automation run <automationId>`	Trigger now; returns the spawned task (`alreadyRunning: true` if one is in flight)

Composing posts

Command	Does
`integrations`	Which platforms are connected for publishing
`x-limit [--text <draft>]`	X’s limit for the connected account (280, or 25,000 with Premium). With `--text`, returns the draft’s weighted length — a URL counts 23, most emoji 2
`shorten <url> [--title <t>]`	Shorten a link for a post
`workspaces` / `whoami`	Workspaces the key can reach; the user it acts as

No passthrough

Every tool the MCP server exposes has exactly one named command above — there is no ait call <tool> escape hatch, so the CLI’s capabilities are exactly what ait help prints. That’s deliberate: agents read this surface to decide what they’re allowed to do, and a generic gateway makes that unreadable. Adding a server tool means adding its command here, in the same change.

Destructive commands

task delete, task delete-content, task clear-content, automation delete, folders delete, and skills delete permanently remove data and refuse to run without --yes:

ait automation delete <automationId>
# {"error": "Refusing to run without --yes: this would delete the automation and every
#            task it ever spawned, published ones included. It cannot be undone — …"}

Each refusal names the real blast radius, which is not always obvious: deleting an automation takes its whole task history with it, and deleting a folder takes every file and nested folder inside it. The check runs locally, before any request goes out. There’s no trash can — nothing deleted comes back. Agents driving this CLI should pass --yes only when a human named that deletion.

Get started

Connect your agent

Guides

Reference

Install

Configure

Commands

Tasks — the review-gate loop

Skills, files, automations

Composing posts

No passthrough

Destructive commands

​Install

​Configure

​Commands

​Tasks — the review-gate loop

​Skills, files, automations

​Composing posts

​No passthrough

​Destructive commands

Install

Configure

Commands

Tasks — the review-gate loop

Skills, files, automations

Composing posts

No passthrough

Destructive commands