Antwork connects over MCP, so drafting, scheduling and publishing happen in the assistant you already work in. One consent screen, no API key, and the same eight platforms whichever one you pick.
Antwork does not add a chat box, a calendar view or a second inbox to your day. It adds tools your assistant can call, so the work happens in the conversation you were already having. These are the asks that carry most of it.
“Write me a LinkedIn post about the migration we shipped this week.”
Before drafting, the agent pulls the workspace context: your name and site, which account it is writing for, and the recent posts already published on that account. The draft comes back in the voice of the account rather than the voice of a language model, and inside that network’s character limit.
“Same idea for X, Threads and Instagram.”
One post per account, grouped under a shared campaign so they stay findable together. Each is written against its own limits rather than one text truncated four times — 3,000 characters on LinkedIn and 280 on X are not the same post.
“Queue it for Tuesday at 09:00.”
Scheduling hands the post to a timed job on Antwork’s side. Your laptop can be shut and the assistant closed; the post still goes out, and comes back with a published state you can ask about later.
“What is in the calendar for next week?”
Drafts, scheduled posts and published ones are all readable from the same tools. In clients that render panels you get the calendar as an interactive card; in a terminal you get the same data as text.
“How did last month do?”
Published posts carry their metrics back, per account and per post. That matters less as a dashboard than as an input: the next draft can be written against what actually performed rather than what you remember performing.
Every client above reaches the same eight networks. What changes between them is not the connection but the rules: how long the text can be, how many files fit, whether an image and a video can share a post. Antwork checks a draft against these before publishing, so a rejection happens in the chat rather than on the network.
One command, one consent screen, and every connected account is something you can just ask for.
