Claude Code Routines for Marketing: A Weekly Draft-the-Campaign Routine (and Why Routines Make Draft-Only Non-Negotiable)

The best version of Monday morning for a marketing team of one is this: you open the review queue at 9am and the week’s drafts are already there. The two emails due this week, the LinkedIn post, a shell for next month’s campaign, and a short note on what collided with what. You read, edit, approve. Nothing has been sent.

Claude Routines can do the first half of that. They are scheduled, unattended runs that use the connectors you allow, and they are the most useful thing to happen to a small marketing team’s Monday in a while. They are also the sharpest argument for a rule this blog keeps coming back to. A routine runs with nobody watching, and it runs every connector tool it is given, including the ones that write. Which means the only safe connector to put in a routine is one that cannot send.

This post explains what Routines are as of September 2026, quotes the line in the documentation that decides everything, and then builds the Monday 8am routine step by step.

What Routines are

Routines shipped in research preview in April 2026 for paid Claude plans. Per Anthropic’s Routines documentation, you write a prompt, set a schedule, choose which connectors the run may use, and Claude executes the prompt on that schedule without you present. The output lands wherever the prompt says: a message, a document, or, with the right connector, a set of drafts in your marketing system.

Marketing writers noticed quickly. The marketer’s guide to Claude Routines and Brand Brain’s routines for marketing automation both list good use cases: weekly competitive scans, content calendar prep, performance summaries. What neither addresses is the permission model, and that is the part that matters.

The line in the docs that decides everything

The Routines documentation says, as of this writing, that connectors are included by default and that during a run Claude can use every tool from an included connector, including writes, without asking for permission.

Read that twice. In an ordinary Claude session, a write tool triggers a prompt, and you decide. In a routine, there is nobody to prompt, so there is no prompt. Every tool the connector exposes is live for the duration of the run.

That is the correct design for an unattended task; a routine that stopped to ask would not be a routine. It also changes the question you have to ask about every connector before you include it. The question is no longer “will I approve this call” but “what is the worst thing any tool on this connector can do with nobody watching.” A CRM connector that can bulk-update records can bulk-update records at 8am. An email connector with a send tool can send at 8am. Anthropic’s own guidance on using Cowork safely makes the same point from the other side: do not schedule tasks that send messages on your behalf.

The connector that is safe in a routine is the one whose every write is a draft. That is not a preference. It is the only configuration where “runs without asking” and “nobody is watching” can both be true without a customer receiving something you never saw. The full argument, with the incidents that motivate it, is in the draft-only rule.

Which servers a routine can see

One practical caveat before the build. If you connected Marqeable to Claude Code with the terminal command from the Claude Code setup post, that connection lives on your machine and in your sessions. A routine does not see it. Routines use connectors that are added in Claude itself, under Settings > Connectors, which is where you choose what a run may include.

So the first step of the build is to add Marqeable there, even if it is already working in your terminal. The server is https://app.marqeable.com/mcp, a remote MCP server over HTTP with OAuth 2.1. You sign in with your normal Marqeable login in a browser window, and the routine runs with exactly the access your account has. On Team and Enterprise plans an Owner adds the connector under Organization settings and members click Connect.

What the routine gets: ten read tools covering business context (brand, ICP, voice, offers), campaigns with their pieces and send days, semantic search over existing content, the content calendar with scheduled sends and market dates, and every automation with its plan. Six write tools create campaign drafts, content drafts, anchored suggestions on existing copy, brief updates and automation drafts. There is no send, launch, publish or delete tool. The worst thing this connector can do at 8am is fill your review queue.

Building the Monday routine

1. Add the connector in Claude. Settings > Connectors > Add custom connector, name it “Marqeable”, paste the address, Add, Connect, sign in. Run one manual prompt to confirm the line is live: “Run whoami, then get_business_context, then tell me what is on the calendar this month.”
2. Create the routine. Give it a name (“Monday drafts”), a schedule (Monday, 8am, your timezone), and include the Marqeable connector. Do not include any connector that can send; if in doubt, leave it out.
3. Paste the prompt. The one below is a working starting point. Adjust the day counts and the piece types to your cadence.
4. Run it once by hand. Trigger the routine manually and read the output before you trust the schedule.
5. Review on Monday. Open the queue in Marqeable, read each draft, approve or edit. When you approve the copy on a piece, that is when images get generated, so a rejected draft costs nothing beyond the words.

Here is the prompt:

You are preparing this week's marketing drafts. Everything you create must be a draft; do not
attempt to schedule, send or publish anything.

1. Run whoami and get_business_context. Use the current offer, ICP and voice from the account,
   not from memory.
2. Read the content calendar for the next 14 days and list every scheduled send and market date.
3. Read the automation overview. Note any automation whose next step falls this week.
4. For every campaign piece due in the next 7 days that has no draft yet, create a content draft
   attached to its campaign, copy only, on its planned send day. Do not create a piece on a day
   that already has a send.
5. If no campaign is planned to start in the next 21 days, create a campaign draft shell for the
   next one: type nurture, light volume, start date three weeks out, description drawn from the
   current offer.
6. Post a summary: what you drafted (with links), what is already scheduled, any collisions or
   duplicates you avoided, and anything that needs a decision from a person.

Every instruction in that prompt maps to a read or a draft write. There is nothing it could do that reaches a customer, and there is nothing in the connector it could reach for if it tried.

What Monday looks like

The routine ran at 8am. Here is the summary waiting in the thread when you open it:

Monday drafts, 8:02am. Signed in as the marketing lead on the Northwind account. Current offer: 30-day pilot.

Scheduled this fortnight: nurture email Thursday the 4th; webinar reminder Tuesday the 16th; LinkedIn case study the 18th. Market dates: quarter close on the 30th.

Automations: the pilot-offer follow-up sends its day-7 check-in to 14 contacts on Wednesday.

Drafted: the “Pilot for controllers” campaign had two pieces due this week with no draft. I created the day-0 email and the day-2 LinkedIn post, copy only, on the 8th and the 10th. Links: [email draft] [post draft]. I did not draft anything for the 4th because a send already exists that day.

Campaign shell: nothing is planned to start before the 22nd, so I created a draft campaign, “Quarter-close pilot push”, nurture, light, starting the 22nd, described around the pilot offer and the quarter-close date. It has no pieces yet. Link: [campaign draft].

Needs a decision: the day-2 LinkedIn post and the case study on the 18th both use the finance-team story. You may want a different angle on one of them.

Ten minutes of reading, three approvals, one decision. The routine did the reading of the calendar and the automations that you would otherwise do by hand, and every one of its outputs is sitting in the queue, not in an inbox somewhere. This is the weekly rhythm described in the AI marketing agent’s weekly plan, with the preparation moved to a schedule, and it is the kind of leverage the marketing team of one actually needs: not more output, but the review-ready version of the output waiting when the week starts.

Variations that fit the same shape

Once the Monday routine works, the same pattern covers other recurring jobs, all of them reads plus draft writes:

• Friday recap. Read the campaign list and the calendar, summarize what went out, what is still in draft, and what is scheduled for next week. No writes.
• Post-launch follow-up. Three days after a campaign’s start date, read the automation overview and, if no follow-up automation exists for that campaign, draft one with quiet hours and an approval gate on the first step.
• Copy review. Weekly, search existing content marked ready, and add anchored suggestions where the copy uses the old offer or a banned phrase. Suggestions are comments a person accepts or rejects, not edits.
• Stale brief check. Compare each in-progress piece’s brief with the current business context and update the brief where the offer or audience has moved.

None of these need a human present, because none of them can do anything a human would need to stop.

What never goes in a routine

Any connector with a send, launch, publish or delete tool. Not on manual approval, not with a careful prompt, not “just this once.” The permission model of a routine is that there is no permission model during the run. A careful prompt is an instruction the model will usually follow; the incidents in the security post are a list of times “usually” was not enough. The connector’s tool list is the only gate that holds when nobody is watching, and it is the first thing to check about any marketing tool you are thinking of scheduling; there is a checklist for that in the martech evaluation question.

Frequently asked questions

What are Claude Code Routines?

Routines are scheduled, unattended Claude runs, in research preview as of September 2026 on paid plans. You write a prompt, pick a schedule and the connectors the run may use, and Claude executes it without you present. Connector tools, including write tools, run during a routine without a permission prompt.

Why is draft-only non-negotiable for routines?

Because nobody is there to click approve. Anthropic’s documentation states that during a routine Claude can use every tool from an included connector, including writes, without asking. If a connector has a send tool, the routine can send. If every write is a draft, the routine can only fill your review queue.

Which MCP servers can a routine use?

Connectors added in Claude under Settings > Connectors. A server added only on your machine with the claude mcp add command is local to that session and does not appear in a routine. Add Marqeable as a connector in Claude so the routine can include it.

What should a weekly marketing routine produce?

A short review list and a set of drafts. The Monday routine reads the calendar and automations, drafts the pieces due that week copy-first, opens a campaign shell if one is needed, and posts a summary of what is waiting. A person reviews and approves in Marqeable; nothing goes out from the routine itself.

The bottom line

Routines give a small marketing team the Monday morning it wants: the week’s drafts waiting in the queue before anyone logs in. They also run every connector tool without asking, which is exactly right for an unattended job and exactly why the only connector that belongs in one is a connector that cannot send. Build the routine around reads and draft writes, keep the judgment in the review, and the schedule works for you without ever working around you.

See the queue the routine fills: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.

Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access