Home

The Full Pipeline: Claude Code + GSD + n8n MCP

Let me get the full picture before describing this properly.Good — now I have the full picture from primary sources. Here’s the accurate, complete description.

The full pipeline: Claude Code + GSD + n8n MCP

These are three independent tools stacked together. Each solves a distinct problem. None of them is “middleware” — they operate at completely different layers.

The problem each tool solves

Claude Code alone is powerful but unreliable at scale. As a session grows, context rot sets in — quality degrades as the context window fills: at 0–30% context Claude is thorough and comprehensive, but past 50% it starts rushing, cutting corners, and past 70% it hallucinates and forgets requirements. For building something non-trivial like a YouTube automation system, one long Claude Code session will drift and degrade before you’re done.

Claude Code with n8n knowledge alone generates plausible but broken workflow JSON. Without the MCP server, Claude is “basically playing a guessing game — ‘Is it scheduleTrigger or schedule? Does it take interval or rule?’” It doesn’t know the exact node type strings, property schemas, or version numbers your n8n instance expects.

Claude Code with n8n MCP alone, but no structure still degrades on complex multi-phase projects and produces inconsistent results across a long build session.

All three tools together eliminate all three failure modes simultaneously.

Tool 1 — GSD (Get Shit Done): the development workflow orchestrator

GSD is a lightweight, powerful meta-prompting, context engineering, and spec-driven development system for Claude Code. It doesn’t add AI capability — it makes Claude Code reliable by imposing a disciplined structure around how work gets done.

GSD is a spec-driven development workflow and prompt/agent harness that prevents context rot by externalizing state into files, splitting work into small plans, executing each plan in a fresh context, and verifying output against explicit goals.

The core mechanism is a phase loop:

Each phase gets your input (discuss), proper research (plan), clean execution (execute), and human verification (verify). Context stays fresh. Quality stays high.

Concretely, how it works under the hood:

In the context of building an n8n automation project, GSD manages the project build — discussing requirements with you, planning the phases (e.g. “Phase 1: webhook trigger and data fetch; Phase 2: AI summarization node; Phase 3: YouTube upload node”), executing each phase with a fresh subagent, and verifying each phase works before moving to the next.

Tool 2 — n8n MCP server: the node knowledge layer

The n8n MCP server is a Model Context Protocol server that provides AI assistants with comprehensive access to n8n node documentation, properties, and operations. It gives Claude deep knowledge about n8n’s 1,239 workflow automation nodes (809 core + 430 community), with 99% coverage of node properties with detailed schemas, 2,646 pre-extracted configurations from popular templates, and validation tools that give AI real feedback so it catches its own mistakes before you even notice.

This is what makes Claude Code go from guessing to composing. Instead of hallucinating node names and property keys, Claude Code can query the MCP server for the exact schema of any node before writing JSON for it. It can look up real-world configuration examples for the Slack node, the YouTube node, the HTTP Request node — all pulled from actual popular templates, not training data.

AI creates workflows directly in your n8n instance — no copy-paste, no import errors, just working automation. AI reads your existing workflows, understands the context, and makes targeted improvements. Documentation is synced with the latest n8n releases within 48 hours. Every node, every parameter, always accurate.

The MCP server also acts as a deployment bridge — Claude Code doesn’t just generate JSON locally, it uses the MCP server to push the workflow directly into your n8n instance, activate it, and trigger a test execution, all without you touching the n8n UI.

Tool 3 — Claude Code: the execution agent

Claude Code is the AI coding agent that actually writes the workflow JSON, reasons about the structure, and issues the MCP tool calls. It sits between GSD (which tells it what to build and how to manage the build process) and the n8n MCP server (which gives it accurate node knowledge and the ability to deploy).

A companion n8n-skills repository provides 7 complementary Claude Code skills that teach it how to build production-ready n8n workflows using the MCP server — covering correct n8n expression syntax, 5 proven workflow architectural patterns, how to interpret validation errors, and expert guidance on using MCP tools effectively. These skills are loaded into Claude Code’s .claude/skills/ directory and activate automatically when relevant.

How the three tools work together in the YouTube automation example

You start the project with GSD:

/gsd:new-project

GSD interviews you about the automation goal — faceless YouTube channel, automated topic research, script generation, voiceover, video assembly, upload. It spawns parallel research subagents that explore the n8n node landscape (via the MCP server), identify what nodes are needed, and assess dependencies. It produces a phased roadmap.

Then for each phase, the loop runs:

/gsd:discuss-phase 1   → locks in your preferences and requirements
/gsd:plan-phase 1      → creates a PLAN.md (2-3 atomic tasks)
/gsd:execute-phase 1   → spawns fresh subagents per task
/gsd:verify-work 1     → runs UAT, catches issues before shipping
/gsd:ship 1            → commits verified work

During /gsd:execute-phase, each subagent uses the n8n MCP server to look up exact node schemas, validate configurations, pull real-world examples, and push the completed workflow JSON directly into your n8n instance. Because each subagent starts with a full 200K context and a precise PLAN.md, it produces correct, working JSON on the first or second attempt — not after five rounds of manual fixing.

When verification catches an error — say the YouTube upload node has a wrong parameter — GSD doesn’t restart the phase. It creates a targeted fix plan, spawns a fresh subagent with just that context, and patches only the broken node.

The result: complex, multi-node, production-ready n8n workflows, built from a natural language description, without you touching the n8n UI or editing JSON by hand.

Here are complete, accurate installation instructions for both tools, pulled from their official repos.

Prerequisites

Before installing either tool you need these on your machine:

Part 1 — Installing GSD

GSD installs into Claude Code globally (affects all projects) or locally (one project only). For the n8n automation use case, global is recommended.

Install globally:

npx get-shit-done-cc --claude --global

This installs to ~/.claude/.

Or install locally into a specific project:

cd your-project-folder
npx get-shit-done-cc --claude --local

This installs to ./.claude/.

Verify the install:

claude
/gsd:help

You should see the full list of GSD commands. If you do, it’s working.

Keep GSD up to date — GSD evolves fast, so update periodically by re-running the same install command.

To uninstall if needed:

npx get-shit-done-cc --claude --global --uninstall

Part 2 — Installing n8n-mcp

Step 1 — Get your n8n API key

In your n8n instance, go to Settings → API → Create API Key. Copy the key and your instance URL (e.g. http://localhost:5678 or https://your-n8n.cloud).

Step 2 — Add the MCP server to Claude Code

The quickest method for Claude Code is to run the claude mcp add command directly in your terminal:

claude mcp add n8n-mcp --scope user \
  -e N8N_API_URL=http://localhost:5678 \
  -e N8N_API_KEY=your_api_key_here \
  -e MCP_MODE=stdio \
  -e LOG_LEVEL=error \
  -e DISABLE_CONSOLE_OUTPUT=true \
  -- npx -y n8n-mcp

Replace http://localhost:5678 and your_api_key_here with your actual values. Use --scope user to keep credentials private to you, or --scope project to share a team config via .mcp.json.

The n8n API credentials are optional — without them you get documentation and validation tools only. With credentials you get full workflow management: creating, activating, and executing workflows directly from Claude Code.

Step 3 — Verify the connection

Inside a Claude Code session:

/mcp

This shows server status and available tools. You should see n8n-mcp listed as connected. You can also test it directly:

Use the list_nodes tool to show me all n8n trigger nodes

Step 4 (optional but recommended) — Install n8n-skills

The n8n-skills package gives Claude Code 7 specialist skills for building production-ready n8n workflows — expression syntax, validation error handling, 5 proven workflow patterns, and more. Install via the Claude Code plugin system:

/plugin install czlonkowski/n8n-skills

Or manually:

git clone https://github.com/czlonkowski/n8n-skills.git
cp -r n8n-skills/skills/* ~/.claude/skills/

Skills activate automatically when relevant — no manual invocation needed.

Putting it together

Once both are installed, your workflow for any n8n automation project looks like this:

# Create a project folder and initialise git
mkdir my-n8n-project && cd my-n8n-project
git init

# Open Claude Code
claude

# Start GSD
/gsd:discuss-phase 1

From there, GSD guides the discussion, creates a phased plan, spawns fresh subagents for each task, and those subagents use the n8n-mcp server to look up exact node schemas, validate configurations, and deploy working workflows directly into your n8n instance — no copy-paste, no JSON editing by hand.

video source: https://www.youtube.com/watch?v=1S4ZZLZCdno

Tags: N8nN8n-McpClaude-CodeClaude-Code-to-N8n