Claude Code + n8n: 7 Workflows I Ship Every Week

Once Claude can see your n8n instance over MCP, the day-to-day changes: you stop hand-wiring nodes and start describing outcomes. But “AI builds your automations” is easy to oversell, so here is the honest version, the seven workflows I actually ship on a weekly rhythm with Claude Code doing the heavy lifting on design, and the pattern that keeps it from going wrong.

If you have not connected the two yet, start with the n8n MCP server setup; everything below assumes Claude can read your nodes.

How I actually use Claude Code with n8n

The mental model that works: Claude is the fast, well-read junior who drafts; I am the reviewer who ships. I describe the outcome and the constraints, Claude drafts the workflow against the real node schema over MCP, I read it, and only then does it run. That loop is quick enough to do many times a day and safe because nothing acts until I approve it.

What Claude is genuinely good at here is the tedious, error-prone part: remembering exact parameter names, structuring branches, wiring error handling, and writing the expressions. What it is not good at is knowing your business rules. That stays with you.

The seven workflows

1. Lead triage and scoring

New form submission comes in, an LLM step scores it against a rubric (fit, intent, budget signals), and the workflow routes it: hot leads to a Slack channel and a fast-reply sequence, everything else to a nurture list. Claude drafts the scoring prompt and the routing branches; I tune the rubric. This one alone repays the whole setup.

2. Keyword to content brief

Give the workflow a keyword and it assembles a brief: search intent, suggested headings, questions to answer, and internal-link targets pulled from a sheet of existing posts. It is the front of the content pipeline, and it is the kind of structured, repetitive generation LLMs are built for.

3. Scheduled reporting with a written summary

Instead of dumping numbers into a channel, the workflow pulls the week’s metrics and hands them to an LLM to write a short, plain-language summary: what moved, what did not, what to look at. Delivered to Slack every Monday. Claude Code is good at drafting both the data-shaping steps and the summary prompt.

4. Inbox to task routing

Incoming email in a shared inbox gets classified (support, sales, billing, noise) and routed to the right place, with a draft reply attached for the human to approve. The classification is cheap, the time saved on triage is not.

5. Content publish and distribution

When a post is marked ready, the workflow handles the mechanical fan-out: update the sitemap trigger, post to the channels, notify the list. This is the same idea behind the pipeline that publishes this very blog, described in the self-building workflows guide.

6. Error alerting that is actually useful

An error-handler workflow catches failures across the other workflows, groups them, and sends one clear alert with the workflow name, the node, and the error, not a wall of raw JSON. Boring, essential, and exactly the kind of resilience covered in the reliability guide.

7. Data enrichment on demand

A lead or company comes in thin; the workflow enriches it from your allowed sources, normalizes the fields, and upserts it into your CRM or database. Claude is good at the field-mapping drudgery that makes enrichment brittle when done by hand.

The one rule that makes this work: human-in-the-loop

Every workflow above that acts on something important has a review point. Claude proposes, a person or an explicit rule approves, and only then does it send the email, message the lead, or write to the database. This is not a limitation to engineer away; it is the design. Full autonomy is where automation projects quietly go wrong, a topic worth its own read in when plain automation beats AI.

What I would tell my past self

Start smaller than feels worthwhile. The lead-triage and reporting workflows are unglamorous and they are the ones that still run months later. Novelty fades; reliability compounds. Let Claude remove the tedium of building, keep the judgment for yourself, and ship the boring workflows first.

Key takeaways

Claude Code plus n8n is a design-and-review loop, not a replace-the-human button. Describe outcomes, let Claude draft against your real schema over MCP, review, then run. The workflows that pay off weekly are the plain ones, triage, briefs, reporting, alerting, and the rule that keeps it safe is human-in-the-loop approval before anything acts.

Want a stack like this built with you? Let’s talk about your automation roadmap.