AI scheduling: let an agent plan and queue your week's content

AI scheduling means splitting the job in two: an agent plans the week — what to post, where, when — and a self-hosted n8n workflow executes the plan on schedule. You review the queue in between. With ReDock, both halves run free on your Mac: n8n as a one-click stack, and an agent connected through MCP or built into the app itself.

Planning is the part worth automating

The execution half of scheduling is a solved problem — a cron trigger and a loop, as laid out in free scheduling with n8n. The half that still eats your evening is planning: deciding what goes out Tuesday, rewording the same announcement three ways for three platforms, picking times, filling the queue row by row.

That’s language work, and language work is what agents are good at. Give an agent your inputs — this week’s blog post, a changelog, a rough list of things worth saying — and ask for a queue: seven days, platform-appropriate variants, spread-out publish times, written into the same sheet or table your n8n workflow reads.

The output isn’t posted. It’s queued. You get a review step for free, because the plan lands in a table you can edit before the cron trigger ever sees it.

The setup, concretely

You need three pieces, all of which ReDock — a free (alpha) native macOS app — provides or hosts:

An n8n instance. One click, warm-started, with a real HTTPS .test domain. The full picture is in n8n on your Mac, one click, and the stacks docs list everything else ReDock runs.
A queue the agent can write to. Simplest version: an n8n workflow with a webhook node that accepts a JSON array of posts and stores them. Now “fill the queue” is one HTTP call.
An agent with hands. ReDock exposes the whole app as MCP tools on localhost:47200, so Claude Desktop, Claude Code, Cursor, or any MCP client can see your sites and stacks. There’s also a built-in agent and a local-LLM catalog, so this works without any API key — see the built-in agent docs.

The working loop looks like this: you paste your raw material into a conversation, the agent drafts the week’s queue as structured JSON, you skim and edit, the agent POSTs it to the n8n webhook, and the schedule trigger takes it from there.

Keep the agent out of the posting path

A design principle worth stating plainly: the agent plans, n8n posts. Don’t wire a model directly to your social accounts.

Determinism where it matters. The posting flow is dumb on purpose — read row, post row, mark row. It behaves the same at 6 a.m. as in your demo.
A human gate by default. Queued-but-not-posted is the natural review checkpoint. You approve a table, not each individual send.
Cheap failure. If the agent drafts something off-tone, you delete a row. Nothing shipped.

This is the same draft-then-gate shape that works for WordPress pipelines — automate WordPress publishing uses it with a review status instead of a queue table.

What this costs

Nothing, at the software layer. n8n’s fair-code license makes self-hosting free. ReDock is free. And because ReDock ships a local-LLM catalog, even the model can run on your machine — no per-token API bill for routine drafting. A local model drafting platform variants of an announcement is a well-scoped, realistic job for it.

The honest constraints from the execution side still apply: your Mac must be awake for triggers to fire, and each platform’s API rules are yours to respect.

FAQ

Does the AI post directly to my accounts?

Not in this design. The agent writes to a queue; a deterministic n8n workflow does the posting on schedule. You review the queue in between, so nothing goes live without a human having had the chance to look.

Do I need an API key for the AI part?

No. ReDock includes a built-in agent and a local-LLM catalog, so planning can run entirely on your Mac. If you prefer Claude or another hosted model, any MCP client connects to ReDock’s tools on localhost:47200.

What does the agent actually see through MCP?

ReDock’s MCP surface covers the app: your sites, stacks, and their state. For the scheduling loop, the agent mostly needs to reach your n8n webhook and your content sources — background on the protocol is in what is MCP.

Plan with a model, execute with a cron job, review in between. Download ReDock and let an agent fill next week’s queue.