Weekly Newsletter Drafts With an AI Agent, n8n and Kit

The short answer: if your team publishes two to five posts a week and someone spends Monday morning turning them into the newsletter, give the first draft to an agent and leave the send button with an editor. A weekly n8n run is 4 or 5 executions a month, inside the 2,500 that come with Starter at $24 a month billed monthly, and the self-hosted Community Edition costs nothing (n8n pricing, checked on 26 October 2026). For example, if building an issue by hand takes 90 minutes and reviewing a draft takes 25, the editor gets 65 minutes back a week: 65 x 52 = 3,380 minutes, or 56 hours 20 minutes a year.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

A digest of your own posts is the most repetitive writing a content team ships: same structure every week, new links. The risk is that one wrong link or invented claim reaches the whole list at once. So the build below gives the agent no way to send. It writes a draft, and a person schedules it.

Before you build:

1. The bill, worked out. A Schedule Trigger counts one execution each time it fires (executions docs). Fifty-two Mondays are 52 executions a year, so the allowance is not the constraint; the chat model's tokens, billed by your provider, grow with the length of your posts.
2. Your feed and your sending account. The RSS feed of your blog and an account on Kit, MailerLite or beehiiv. For Kit, create a V4 API key in the Developer tab of your account settings; Kit shows it once (Kit authentication).
3. An editor with the last word. One named person who reads every draft and is the only one who schedules it, plus a one-page house style: tone, length, words you never use, claims you never make.
4. Two numbers from the old way. Time the next two hand-built issues, first click to scheduled, and note the click rate of your last eight issues. That is the baseline for the last section.

Where the draft lands, and what stops it going out

The orchestration is the same everywhere. The way in is not.

The workflow in n8n, with Kit

1. Trigger on Monday, in your own time zone. A Schedule Trigger, Weeks interval, Monday, 07:30. It follows the workflow's time zone, then the instance's, which defaults to America/New York when self-hosted and GMT on n8n Cloud, so set yours in the workflow settings. Save and publish, or it never fires (Schedule Trigger).
2. Read the week's posts. An RSS Read node takes your feed URL. A Filter node after it keeps the items whose date is after seven days ago, using the Date & Time comparison "is after"; everything older is dropped from the output.
3. The agent reads, decides and writes. Connect an AI Agent node and a chat model. The node needs at least one tool, and here the tool has a real job: an HTTP Request tool that opens each post's URL, with the tool-only option that trims the response switched on, so the model reads the article and not the page around it. The system message holds your house style and four rules: one short paragraph per post, links only to URLs from the feed, a figure only if the post contains it, and anything uncertain goes into a note for the editor. The agent decides the order and the line that sells each post.
4. Shape the answer. Attach a Structured Output Parser and generate its schema from a JSON example with four fields: subject, preview_text, html and editor_notes. Fields generated this way are all mandatory, so a draft without a subject fails here, not in the inbox.
5. Put the draft in Kit. An HTTP Request node sends POST to https://api.kit.com/v4/broadcasts with a Header Auth credential carrying X-Kit-Api-Key. Map subject, preview text and content from the parser, write "Agent draft, week" and the week number into description, set public to false, choose the segment or tag in subscriber_filter, and leave send_at as null (Create a broadcast).
6. The editor approves in Kit. A last Gmail or Slack node tells the editor the draft is waiting and pastes in editor_notes (Gmail node). The editor clicks every link, checks every figure against its post, reads the subject against the body, sends a test to their own inbox and schedules. That click is the approval and the only route to subscribers, because no node sets a send time.

The per-tool setup behind each step is in how to use n8n and how to use Kit, and the products in the n8n and Kit entries.

On MailerLite or beehiiv instead

MailerLite. Swap step 5 for POST to https://connect.mailerlite.com/api/campaigns with type regular, your verified sender and the group (campaigns API). The catch is the body: the API docs tie HTML content to the Advanced plan, a name the current pricing page no longer lists beside Free, Comfort and Power, so ask MailerLite which plan that is before you build. Otherwise run the agent step each Monday from Claude or ChatGPT through the MCP ("draft this week's issue from these four URLs"), and the campaign stays a draft until you confirm. Setup is in how to use MailerLite.

beehiiv. On Enterprise, the create-post call works from step 5 only if the body says "status": "draft"; a draft cannot be scheduled through the API, so the editor schedules it in the app (Create post). On Lite or Pro, use the MCP from a chat client: it creates and edits drafts, and publishing stays in beehiiv (MCP article). More in how to use beehiiv; still choosing, see Kit vs beehiiv.

The free version, which is enough on its own

Every Monday, copy the titles, URLs and opening paragraphs of the week's posts into a free chat assistant with your house-style prompt saved at the top, paste the result into a new draft, send yourself a test and schedule it. MailerLite's Free plan allows 250 subscribers and 2,500 sends monthly (pricing), so five issues to a full list are 250 x 5 = 1,250 emails, half the allowance, and beehiiv's Free plan takes up to 2,500 subscribers (pricing). To automate the schedule without a subscription, n8n's Community Edition is the self-hosted version on GitHub, and someone keeps that server running.

What the paid route adds is narrow and real: the run happens when nobody remembers, the agent reads whole posts, and every draft arrives in the same shape with its notes attached.

What it costs a month

Prices checked on 26 October 2026 on the n8n, MailerLite and beehiiv pricing pages. Kit's pricing page answered automated reading with a bot check that day, so read it in a browser.

How to know it worked

• Editor minutes per issue, against the two hand-built issues you timed. If review takes as long as writing did, the system message is missing rules.
• Edits per draft. The same edit twice is a sentence for the house style, added on a fixed day so you can trace which change helped.
• Clicks, not opens. Compare the click rate of four agent-drafted issues with the median of your last eight. Apple's Mail Privacy Protection downloads remote content in the background when a message arrives (Apple), so opens say less than they did.
• No subscriber holdout. Readers expect the issue every week, so withholding it from a slice tests nothing. Compare four agent-drafted issues with the four hand-built ones before them, same day and send time.

The default to check before the first run

A publishing platform gains when posts go out, and its API default leans that way: beehiiv's create-post endpoint publishes a post unless the request says draft, so on that route the whole approval step hangs on one field, and it belongs in your test run before anything else. Kit's draft rule works the other way, with null as the safe value, which is one reason this page builds on it.

If you would rather have the workflow built, tested and handed over with the prompt and the approval rule written down, that is the agentic marketing work I do.

FAQ

Can the agent send the newsletter by itself?

Not in this build. No node sets send_at in Kit or calls MailerLite's schedule endpoint, Kit's and MailerLite's MCP servers ask for explicit confirmation before a send, and beehiiv's MCP cannot send at all.

Does the agent see my subscribers?

In the n8n build, no: it reads your posts, and the draft targets a segment or tag by ID. Connecting MailerLite's MCP is different: it grants the AI tool access to manage your MailerLite data (MCP guide). Article 5(1)(c) of the GDPR limits data to what the purpose needs, an argument for the build that never touches the list.

Do I need a developer?

For the n8n build, someone who can read an API reference and a JSON response. For the MCP route, no: a weekly prompt in a chat client does the drafting, and the editor still schedules.