Before building anything, four parts of the environment need to be connected. The diagram below shows how the coding assistant connects to your project and n8n — and what you check before getting started.
Illustrative diagram — conceptual recreation of the environment, not a real screenshot.
🗣️ What is vibe coding
Vibe coding is building software and automations by talking with an AI in natural language: you describe the result you want, and the AI handles the technical implementation. Instead of learning the mechanics of each node or each line of code, you focus on the problem—the AI handles tool selection, configuration, and debugging.
💡 The core idea
Natural-language conversation drives the build and iteration. When something needs to change, you don't reopen a configuration editor: you just ask for the change. The AI does the "heavy lifting"—choosing nodes, configuring parameters, and debugging.
- •Speed: from concept to functional prototype in minutes, not hours.
- •Low technical overhead: without memorizing how each node works.
- •Focus on the problem: the effort goes into the result, not the tool.
✓ Describe the RESULT
- ✓“Check Gmail every 10 minutes; if a new email arrives, send a DM in Slack.”
- ✓Let the AI choose the nodes and structure.
- ✓Answer her clarification questions.
✗ Don’t Dictate the Technical Steps
- ✗“Add a trigger node, connect it to an IF, then to a code node…”
- ✗Micromanage the configuration of each parameter.
- ✗Expect the AI to guess without you answering its questions.
🎯 Practical tip
Start your sentence with the goal ("I want to…"), not the tool. Say what should happen in the end; let the AI suggest how. You still approve everything before anything runs.
Natural language drives the build.
Just ask to make a change.
The AI chooses and configures the nodes.
Fundamentals → build → enhancement.
🧠 The Code Assistant in the Editor
The code assistant runs inside the editor (a third-party editor, such as VS Code, that you install to host it). It can read project files, write and edit code, run terminal commands, and connect external tools via MCP. It's the engine that turns your conversation into real actions in the project.
📊 Subscription vs. API — saving tokens
The AI rereads the entire conversation with each new message—so the longer the session, the more each message costs. Understanding the billing model helps you avoid surprises.
- Subscription: a fixed rate, based on chat/text, with a message limit—recommended for this type of work.
- API: token-based, pay-as-you-go — designed for people building applications.
✓ Best practices
- ✓Keep sessions focused to control costs.
- ✓“One folder per project” — open only the dedicated folder.
- ✓Use the subscription path for this work.
✗ Avoid
- ✗Open a root folder with multiple projects inside.
- ✗Huge sessions that mix unrelated tasks.
- ✗Ignore that every message rereads the entire history.
Sees the entire project.
Creates and changes code.
Run in the terminal.
Connects external tools.
📄 Dedicated Project + CLAUDE.md
Configuring the project means creating a dedicated folder and a file CLAUDE.md, which works as a persistent system prompt: it defines the role, context, and preferences that the AI follows throughout the project. Instead of repeating context in every conversation, you write it once in this file.
🧭 What goes in CLAUDE.md
- •Role: who the AI is in this project (e.g., an n8n workflow assistant).
- •Context: what you’re building (automations) and why.
- •Preferences: communication style, technical constraints, dos and don'ts.
# Role You are an n8n workflow assistant. # Context I'm building automations for an AI agency. Help build, debug, and optimize. # Preferences - Concise, direct explanations. - Always propose a plan before executing.
You don’t need to write this file by hand: ask the AI to generate it (“create a CLAUDE.md for this project; I’m building n8n workflows for an agency—help me build, debug, and optimize them”). Then iterate by asking for changes (“add a section about keeping explanations concise”).
🎯 Practical tip
Start with a folder completely empty and named according to its purpose. CLAUDE.md can be edited at any time—just ask the AI to update it as the project grows.
Applies to the whole project.
Plain text, easy to edit.
The AI writes the first version.
Request changes whenever you want.
🔌 Connecting n8n via MCP server
O n8n is a visual, node-based workflow builder that connects apps, data, AI, and external APIs. A MCP server connects n8n to the coding assistant so the AI can read, create, modify, test, debug, and even self-correct its own workflows. Building everything by hand is time-consuming, error-prone, and difficult to debug—the MCP removes that friction.
Installation walkthrough
Clone the MCP repository
Ask the AI to clone the repository and configure it. If it offers installation via NPX, decline — clone it manually.
Project-level config
Keep the MCP configuration inside the project folder. Global config triggers a security warning on every restart.
Create the API key + copy the instance URL
In n8n's settings, create an API key (label + expiration) and copy it right away. Also copy the instance URL (cloud or localhost/custom).
Paste the credentials into .env and run the health check
Paste the API key and URL into the file .env of the project. Ask the AI to start the MCP and run a health check.
⚠️ Attention: credential security
- ✗Never paste credentials into the chat—only in the
.env. - ✗Don’t let the AI run the installation via NPX—it causes silent failures that are hard to diagnose.
- ✗The API key appears only once during creation—copy it and store it securely.
🎯 Practical tip
The health check is your first sign of life: it confirms the connection status, the instance URL, the MCP version, and the API response time. Run it whenever something seems off.
Visual workflow builder.
AI ↔ n8n bridge.
The sole home for credentials.
Confirms the connection.
📚 n8n Skills
The skills are documentation that teaches the AI how to use n8n—its node types, parameters, best practices, and common patterns. With them, the AI understands how each node works and how to structure a workflow, so you you don’t need to know that. Installation is identical to that of the MCP server; only the repository URL changes.
Installation and verification
Clone the skills repository
Ask the AI to clone and configure it (same process as the MCP, different link). Approve the permission prompts during installation.
Update the CLAUDE.md and restart
Ask it to update CLAUDE.md to note that the AI now has access to the n8n documentation. Restart the assistant to register the new components.
Check by asking about a node
Ask "tell me about the Gmail trigger node" and confirm the answer includes accurate details (polling interval, parameters, filters).
🎯 Practical tip
If the MCP server doesn't appear after installation, restart the assistant before any troubleshooting — most of the time, that solves it.
Types, parameters, defaults.
Same as the MCP, another repo.
Records the components.
Ask about a real node.
✅ Verification of the 4 connections
Before building, run a verification sequence to confirm that the assistant has access to the project files, the MCP connection, the ability to read workflows, and the node documentation. Four simple tests — each a question in natural language.
The 4 tests
Project directory
“Show me what’s in my project directory.”
Confirm that CLAUDE.md, .env, MCP files, and the skills folder exist (no workflows yet).
MCP connection
“Can you confirm that you’re connected to my n8n instance?”
Confirm a healthy connection: connected instance, configured API, available tools.
Read workflows
“What workflows do I currently have?”
Confirm that the AI lists the workflows (or reports "none" in a new account), with node count and status.
Node documentation
“Tell me about the Gmail trigger node.”
Confirm node-level details: polling interval, parameters, filters.
✓ Environment ready
- ✓All 4 tests pass without errors.
- ✓The AI can see files, the instance, workflows, and nodes.
- ✓You can start building safely.
✗ Common Mistakes (and the Fix)
- ✗Invalid/partial API key → check that it wasn’t truncated.
- ✗MCP doesn't connect → restart the server.
- ✗n8n authorization fails → recheck the credentials.
🎯 Practical tip
If any test shows an error, ask the AI to diagnose it—it usually identifies the cause and suggests a fix. You don't need to decipher the message yourself.
Self-healing (optional): where should the credentials (API key + instance URL) be stored?
📌 Module Summary
Next Module:
1.2 — The Complete Cycle: Plan → Build → Troubleshoot → Optimize