How to Build AI Agents with Claude Code and N8N

What each tool is actually good at

Before combining them, it's worth being clear about where the strengths lie.

Claude Code is Anthropic's agentic coding environment. It can read files, execute commands, reason over large codebases, and write and iterate on code autonomously. For building the logic of an agent — the reasoning, the decision trees, the API integrations — it's extremely capable. It's also a developer tool: you need to be comfortable in a terminal to get the most out of it.

N8N is a visual workflow automation platform. It's built for running processes reliably: triggering workflows on schedules or events, connecting to hundreds of business apps, handling errors, retrying failed steps, and giving non-technical team members visibility into what's running. It's where your business logic lives once it's been built.

| | Claude Code | N8N | |---|---|---| | Best for | Building agents, writing logic, reasoning over complex tasks | Running workflows reliably, visual debugging, business app integrations | | Requires | Terminal comfort, developer mindset | Low-code familiarity | | Output | Code, agents, workflow definitions | Running automations, triggered processes | | Weaknesses | Not designed for long-running scheduled execution | Can't reason; follows rules, doesn't think |

The architecture that works

The most effective pattern treats Claude Code as the builder and N8N as the runtime.

Claude Code
  → Builds and tests the agent logic
  → Generates N8N workflow JSON from natural language descriptions
  → Iterates on the workflow based on test results

N8N
  → Receives the workflow definition
  → Runs it on a schedule or trigger
  → Handles retries, error routing, and logging
  → Exposes a webhook that client-facing tools can call

This means you don't manually wire together 40 N8N nodes. You describe the workflow to Claude Code, it builds the node structure, and you deploy it. Workflow build time drops from hours to minutes.

Setting up the integration

To connect Claude Code to your N8N instance, you use the N8N MCP server. This gives Claude Code direct access to your N8N API — it can create, update, and activate workflows without you touching the UI.

Step 1: Install the N8N MCP server

The community-built N8N MCP server is available on GitHub. Install it globally and point it at your N8N instance URL and API key:

npm install -g n8n-mcp

Then add it to your Claude Code MCP configuration:

{
  "mcpServers": {
    "n8n": {
      "command": "n8n-mcp",
      "env": {
        "N8N_API_URL": "https://your-n8n-instance.com/api/v1",
        "N8N_API_KEY": "your-api-key-here"
      }
    }
  }
}

Step 2: Describe the workflow in natural language

Once connected, you can ask Claude Code to build a workflow by describing what it should do:

Build me an N8N workflow that:
- Triggers via webhook when called with a JSON body containing "company_name"
- Searches for the company in our CRM via the HTTP Request node
- If found, pulls their last 5 support tickets
- Summarises them using the Anthropic node
- Posts the summary to Slack channel #account-management
- Returns a 200 response with the summary text

Claude Code will generate the workflow JSON, validate the node configuration, and deploy it to your N8N instance.

Step 3: Test and iterate

Claude Code can run the workflow, read the output, identify failures, and update the configuration — without you manually editing nodes. This loop is where significant time is saved on complex workflows.

Connecting your N8N workflows to a client portal

Once your workflows are running in N8N, the next layer is giving clients a way to trigger and interact with them — without exposing the N8N interface.

This is where a client portal comes in. Each N8N workflow exposes a webhook URL. The portal calls that webhook when a user submits a task, displays the result in the task panel, and logs the interaction in the task history.

Client submits task via portal
  → Portal calls N8N webhook with task parameters
    → N8N runs the Claude-built workflow
      → Result posted back to portal callback
        → Client sees output in their task panel

The client never sees N8N. They see a branded interface with their agents listed by name, their task history, and a conversational interface for triggering new work.

Common mistakes to avoid

Enabling Dangerous Mode without a trusted environment. Claude Code requires elevated permissions to operate efficiently. Make sure it's running in a sandboxed environment — not against a production N8N instance without safeguards.

Treating Claude Code output as production-ready without review. Claude Code is fast and capable, but generated workflows should be reviewed before going live. Spot-check the node configuration, especially for any steps that write to external systems.

Rebuilding the N8N workflow every time you update agent logic. Store your workflow descriptions as version-controlled prompts. When you need to update behaviour, update the description and regenerate — don't manually edit deployed workflows if you can avoid it.

What this stack enables for agencies

For an AI automation agency, the Claude Code + N8N combination significantly changes the economics of building. Workflows that previously took a developer a day can be produced in an hour. That means lower build costs, faster iteration with clients, and the ability to take on more work without proportionally more time.

The pattern also scales cleanly. You can maintain a library of workflow templates — described in plain language — that Claude Code can instantiate and customise for each new client. Your IP is in the descriptions and the logic, not in hand-built node configurations.

---

Once your N8N workflows are built, Agentic Vessel gives your clients a clean portal to use them. Get started free.