your first n8n workflow, built by claude.

Goal: Find which Zaps eat your tasks, and which ones nobody needs. (10 min)

1. Open your Zapier task usage and list Zaps by tasks used.
2. Remember how it counts: triggers, filters, paths and formatter steps are free, but every successful action step is a task.
3. A Zap with 1 trigger and 3 actions uses 3 tasks every run. n8n counts the whole run as one execution.
4. Mark Zaps that are dead, duplicated or that nobody owns.

Here's our Zap list with steps and monthly task usage:

[paste]

Rank them by tasks. Flag likely dead or duplicate Zaps. For the top 3, estimate monthly n8n executions instead (one per run).

Done when: You know your top 3 task-eating Zaps and which ones to just delete.

Goal: Have an n8n instance you can log into, reachable by Claude. (20 min)

1. Easiest: start an n8n Cloud trial. You get a public URL and managed updates.
2. Self-hosting: run n8n with Docker and give it a public HTTPS URL, because Claude on the web must be able to reach it.
3. Use version 2.33 or later for the guided Claude connection, and 2.13 or later for Claude to create and edit workflows.
4. Licence: one instance for your own company is fine. Agencies hosting many clients in one shared instance need an Enterprise licence.

Done when: You can log in and create an empty workflow.

Goal: Let Claude see, run and build workflows, but only the ones you allow. (15 min)

1. In n8n, go to Settings → Instance-level MCP and choose Enable MCP access.
2. Under Connection details, choose Connect, then Claude, with OAuth (recommended). Use one-click setup, or copy the server URL (it ends in /mcp-server/http) into Claude via Customize → Connectors → Add custom connector.
3. Approve access when Claude redirects you to n8n.
4. Turn on "Available in MCP" only for the workflows Claude should run: workflow menu (…) → Settings.

Which n8n workflows can you see and run? List them with what each one does.

Done when: Claude lists exactly the workflows you enabled, and nothing else.

Goal: Rebuild your biggest Zap in n8n without dragging a single node yourself. (20 min)

1. Describe the Zap in plain English: trigger, each step and the result. Or paste its steps.
2. Ask Claude to create it in n8n, unpublished, with a note on each node.
3. Open it in n8n and add credentials (Slack, Sheets and so on) yourself in n8n. Never paste passwords or keys into chat.
4. Run a manual test with sample data.

Create an unpublished n8n workflow: [describe trigger → steps → result]. Name it "[name] (migrated from Zapier)". Add a sticky note explaining each step. Don't add credentials; list which ones I need to connect.

Done when: A manual test run produces the same result the Zap does.

Goal: Switch over with zero missed events. (15 min)

1. Publish the n8n workflow and keep the Zap running too.
2. Each day, compare outputs, or ask Claude to compare n8n's execution log with Zapier's history.
3. Fix mismatches. Expired credentials are the usual culprit.
4. When a full week matches, switch the Zap off. Downgrade Zapier at renewal.

Compare today's n8n executions for "[workflow]" with this Zapier history:

[paste]

List any event that one handled and the other didn't, and any differences in output.

Done when: A full week with matching outputs, and the Zap switched off.