How it all connects: a visual map of the stack

The short answer

Claude Chat is for one-off questions. Claude Projects are for repeatable work with persistent context. Cowork runs Claude on a schedule. n8n moves data between Claude and your other tools. You’ll use whichever one matches the job.

A picture worth a thousand frameworks

Here’s how the four pieces fit:

                    Your tools
              (Gmail, Notion, Slack,
              Drive, Calendar, etc.)
                       │
                       ▼
                ┌─────────────┐
                │     n8n     │  ← moves data between systems
                └──────┬──────┘
                       │
                       ▼
            ┌──────────────────────┐
            │       CLAUDE         │
            │                      │
            │  Chat   Project      │  ← reactive: you ask, Claude answers
            │                      │
            │  Cowork              │  ← scheduled: Claude runs on its own
            └──────────────────────┘

n8n sits between your everyday tools and Claude. It picks up data, hands it to Claude, and writes Claude’s output back somewhere useful.

Within Claude itself, you have three modes:

• Chat — you type, Claude responds, conversation ends when you close the tab
• Project — same as Chat, but with persistent knowledge files and a system prompt
• Cowork — a Project that runs on a schedule without you opening anything

When to use which

A simple decision tree:

“I have a one-off question.”
→ Claude Chat.

“I’m doing this same kind of task with different inputs.”
→ Claude Project.

“This should happen automatically on a schedule.”
→ Cowork (if Claude is doing the thinking) or n8n (if it’s mostly data plumbing). Often both.

“This needs to read from / write to other apps.”
→ n8n, plus whatever Claude surface is appropriate for the thinking part.

You’ll find that most non-trivial workflows use a mix. The LinkedIn content pipeline (one of our flagship recipes) uses n8n for the RSS-and-database plumbing, Claude API calls for the scoring, and a Claude Project for the drafting — three tools, one workflow.

Where the recipes on this site fit

Each tier of recipes maps to this stack:

• Beginner recipes stay inside Claude Chat or Project. No n8n, no Cowork. You paste in, you get out. Five minutes.
• Intermediate recipes add one external tool, usually through n8n or a built-in connector. Notion, Gmail, Slack — pick one.
• Expert recipes orchestrate the full stack. Multiple tools, scheduled execution, error handling.

The complexity ladder is real. Don’t skip rungs. People who try to build a fully automated multi-system workflow as their first AI project mostly produce a system that breaks in week two and gets abandoned in week three.

Start at Chat. Get comfortable. Move up when you have a real reason.

The thing nobody tells you

Setting up the tools is the easy part. The hard part is identifying what to automate.

Most teams have ten ideas the first day, six of which would deliver zero value, three of which would deliver value but never get used, and one that would actually compound. Picking that one is the skill.

The fastest way to build the skill: try a beginner recipe this week. Try an intermediate recipe next week. After two or three, you’ll start seeing your own work in those shapes. That’s when you know which automation is worth your time.

Last reviewed: 7 May 2026