Recipes / Turn a recipe into a scheduled n8n workflow

Turn a recipe into a scheduled n8n workflow

n8n

Who it's for

RevOps people who have a chat recipe that works by hand, like the weekly pipeline digest, and want it to run on a schedule without anyone opening a chat. n8n can be swapped for Zapier or Pipedream, though the build steps differ.

What you get

Workflow plan, validation result and a pin-data test, before publishing. Sample output: every company and person in it is fictional.

Workflow (created, not published): Weekly pipeline digest

1. Schedule trigger: Mondays 7:00 am
2. CRM node: read open deals (stage, amount, close date, owner, last activity)
3. Code node: group by stage, flag no activity in 14+ days and close dates in the next 30 days
4. Slack node: post the summary to #pipeline-test

Credentials: uses 'CRM - RevOps read only' and 'Slack - digest bot', both set up in n8n by you. No keys were shared in this chat.

Validate: passed, 0 errors, 1 warning (Slack channel is a test channel).

Test run with pinned sample data (6 fictional deals):
Pipeline digest, week of Oct 13 (baseline week: no comparison yet)
Open: 6 deals, $142K
- Mossgate Labs (fictional), $31K, no activity in 18 days (owner: Priya Raman)
- Fernhill Software (fictional), $22K, closes Oct 30 (owner: Alex Rivera)

Next: switch the channel to #pipeline and publish. Reply 'approve publish' when ready.

Time and cost

Estimate About 30–60 minutes for the first build, most of it testing.

Assumptions: One build session for a two-system weekly digest, with two test runs. Weekly runs after that happen inside n8n, outside the chat.

• n8n: Several reads (SDK reference, node searches, credential listing), one validate call per draft, one workflow created, one or two test runs with pin data, then one publish after you approve it. Credit use not confirmed by n8n.

Exact credit costs depend on your plan with each vendor. Check the tool pages before a large run.

MCP servers

Every server is confirmed in: ClaudeChatGPTClaude CodeCursor

• n8n Not probed
  ClaudeChatGPTClaude CodeCursorRead + writeCredits not confirmedNeeds admin or IT

Set up in a chat app

Every server in this recipe is confirmed for Claude and ChatGPT. Connect each one, then paste the prompts below into one conversation.

• Connect n8n without code (Claude, ChatGPT)

Developer setup

{
  "mcpServers": {
    "n8n": {
      "url": "None"
      // headless alternative: "headers": { "Authorization": "Bearer YOUR_N8N_API_KEY" }
    }
  }
}

Step-by-step prompts

Copy all prompts

1. Read n8n's workflow SDK reference, then search n8n's nodes for a schedule trigger, my CRM ({crm}) and Slack. Turn this recipe into a workflow plan, one line per node: {paste the recipe steps}. Run it {schedule, e.g. Mondays at 7am}. Show me the plan and don't create anything yet.
2. List my n8n credentials and tell me which ones match {crm} and Slack by name. If one is missing, stop and tell me to set it up in n8n. Don't ask me for any keys.
3. Write the workflow code for the plan with the Slack node posting to {test channel}, and validate it. Show me any errors and fix them. Create the workflow in n8n only after I approve the validated code.
4. After I approve it, run a test with pinned sample data: about six fictional deals with the same field names my CRM uses. Show me the exact Slack message it produced.
5. After I approve, update the Slack node to {real channel}, show me the diff against the tested version, and publish the workflow only after I approve it.

Run it on a schedule (optional)

• This recipe builds the schedule rather than running on one. The schedule trigger inside the workflow sets when it runs.
• Before you publish, decide whether the Slack post may go out with no human review. If not, have the workflow post to a private channel or to you, and forward it yourself.
• After the first scheduled time, ask the assistant to search n8n executions for this workflow and show you the result, so you know the run worked.

Where this goes wrong

• A test that posts to the real channel. A test run can still reach Slack. Point the Slack node at a private test channel during testing, and only switch to the real channel in the last edit before publishing.
• Pinned data that doesn't look like real data. If the sample deals have different field names than your CRM returns, the workflow passes the test and breaks on the first real run. Copy the field names from one real CRM record (without customer details) into the pin data.
• Credentials pasted into the chat. The model may ask for an API key to fill in a node. Never paste one. Set up the CRM and Slack credentials in n8n yourself, then ask the assistant to list credentials and pick yours by name.
• Publishing the wrong version. If you edit after testing, the published version may not be the one you tested. Ask for the workflow's version history and a diff against the tested version before you approve the publish.

Gotchas

• There's no shared n8n endpoint. Each instance has its own MCP URL ending in /mcp-server/http (on Cloud, https://<your-instance>.app.n8n.cloud/mcp-server/http).
• An n8n owner or admin must turn on the MCP server under Settings > Instance-level MCP before anyone can connect.
• Each workflow has to be marked Available in MCP on its own. If the assistant can't find or test a workflow, check that setting in n8n.
• n8n 2.33.0 and later has a Connect a client dialog with one-click setup for Claude.ai. Cloud-based clients like Claude.ai and ChatGPT need your instance to be publicly reachable, which matters if you self-host.
• Community n8n MCP servers are not n8n's official server. Use the built-in instance-level one.

Download

Get the prompts, guardrails and expected output as one Markdown file. Paste it into a Claude Project or ChatGPT project's instructions, or save it as CLAUDE.md for Claude Code.

Download recipe-to-n8n-workflow.md