How to Connect Claude to n8n via OpenRouter (2026 Guide)

Why OpenRouter Instead of Direct Anthropic API

You could connect directly to Anthropic's API, but OpenRouter is better for n8n workflows for three reasons:

OpenRouter is trending at 70 on Google Trends (January 2026) — it's becoming the standard way to access AI models in automation tools.

Setup: OpenRouter + n8n in 5 Minutes

Step 1: Create an OpenRouter Account

1. Go to openrouter.ai
2. Sign up (free — you get $1 in credits to start)
3. Navigate to Keys
4. Click Create Key, name it "n8n-production"
5. Copy the key (starts with sk-or-...)

Step 2: Store the Key in n8n

In n8n, go to Credentials → Add Credential → Header Auth:

Name: OpenRouter API
Header Name: Authorization
Header Value: Bearer sk-or-YOUR_KEY_HERE

Now you can reuse this credential across all your AI workflows without exposing the key in each node.

Your First Claude Workflow

Let's build a simple content generator — the "Hello World" of n8n + AI.

The flow: Manual Trigger → Set Topic → Call Claude → Extract Result

The HTTP Request Node

Method: POST
URL: https://openrouter.ai/api/v1/chat/completions

Authentication: Use your Header Auth credential

Body (JSON):
{
  "model": "anthropic/claude-sonnet-4-20250514",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "{{ $json.topic }}"
    }
  ],
  "max_tokens": 1000
}

Prompting Patterns for Automation

Prompts in automation workflows are different from chat. You don't get to iterate — the prompt needs to work right the first time, every time.

Pattern 1: Role + Task + Format

System: "You are a [ROLE]. Your task is to [TASK].
Always respond in this exact format: [FORMAT]."

User: "[DYNAMIC INPUT FROM WORKFLOW]"

Pattern 2: Few-Shot Examples

System: "Convert customer feedback into action items.

Example input: 'The dashboard is too slow'
Example output: {"category": "performance", "action": "Optimize dashboard load time", "priority": "high"}

Example input: 'Love the new feature!'
Example output: {"category": "positive", "action": "No action needed", "priority": "low"}"

User: "{{ $json.feedback }}"

Pattern 3: Chain of Thought

System: "Analyze this data step by step:
1. First, identify the key metrics
2. Then, compare to benchmarks
3. Finally, provide 3 actionable recommendations

Think through each step before responding."

Getting Structured Output (JSON)

For automation, you almost always want JSON output so the next node can use specific fields.

// Add to your request body:
"response_format": { "type": "json_object" }

// And tell Claude in the system prompt:
"Always respond with valid JSON. No markdown, no explanation, just JSON."

Then in your next Set node, parse it:

{{ JSON.parse($json.choices[0].message.content) }}

Error Handling

AI API calls fail. Rate limits, timeouts, malformed responses. Here's how to handle it in n8n:

• Retry on failure: In the HTTP Request node settings, enable "Retry On Fail" with 3 attempts and 5-second wait between.
• Validate JSON output: Add an IF node after parsing to check if required fields exist before passing to the next step.
• Fallback model: Add a second HTTP Request node on the error output that calls a different model (e.g., GPT-4 as backup for Claude).
• Timeout: Set the HTTP Request timeout to 60 seconds. Claude can take 10-30 seconds for complex prompts.

Download the Free Workflow

Import this starter workflow into your n8n instance to get going immediately:

Want more? The TurtleTools template library has 8 workflows covering content, leads, MCP integration, and more.

What's Next

• What is MCP and Why It Matters for n8n — the next evolution of AI + workflows
• Build a Vibe Marketing Stack — full content automation pipeline
• Getting Started with n8n + Claude — the 10-minute quickstart