← All articles
n8nCost TrackingNo-CodeTutorial

How to Track AI Agent Costs in n8n

2025-04-05·5 min read·Nova — @NovaShips

The Problem With n8n AI Workflows

n8n makes it easy to connect AI models to your workflows. OpenAI node here, Anthropic node there, and suddenly you have a powerful automation. What you don't have is any visibility into what each run costs.

The provider dashboard shows your total API spend. It doesn't show which workflow spent what, which run spiked, or which automation is running 10x more than expected.

What You Need

Before you start, you need:

  • An n8n instance (cloud or self-hosted)
  • An AI node in your workflow (OpenAI, Anthropic, or any LLM)
  • An AgentShield account — free at agentshield.one
  • Step 1: Get Your AgentShield API Key

    Sign up at agentshield.one → Settings → API Keys → Create key.

    Your key starts with `ags_live_`.

    Step 2: Find Where Your AI Node Outputs Token Usage

    Most LLM nodes in n8n return token usage in the response. For OpenAI:

    $json.usage.prompt_tokens
    $json.usage.completion_tokens

    For Anthropic:

    $json.usage.input_tokens
    $json.usage.output_tokens

    Step 3: Add an HTTP Request Node

    After your AI node, add a new node: HTTP Request.

    Configure it exactly like this:

    Method: POST
    URL: https://api.agentshield.one/v1/track
    
    Headers:
      Authorization: Bearer ags_live_xxxxx
      Content-Type: application/json
    
    Body (JSON):
    {
      "agent": "my-n8n-workflow",
      "model": "gpt-4o",
      "input_tokens": {{ $json.usage.prompt_tokens }},
      "output_tokens": {{ $json.usage.completion_tokens }}
    }

    Replace `my-n8n-workflow` with a name that identifies this workflow in your dashboard.

    Step 4: Run Your Workflow

    Execute the workflow. In AgentShield's dashboard, you'll see:

  • Cost per run, in real time
  • Token breakdown (input vs output)
  • Cost trend over time for this workflow
  • Anomaly detection — if a run costs 3x the baseline, you get an alert
  • Setting Budget Caps

    Once you're tracking costs, you can set a budget cap per workflow. If your research workflow hits $5 in a day, AgentShield can send you a Slack alert or freeze the workflow automatically.

    This is the per-agent, per-session visibility that your provider dashboard doesn't give you.

    Conclusion

    Five minutes of setup. An HTTP Request node. That's all it takes to go from zero visibility to complete cost attribution for your n8n AI workflows.

    Track costs, catch spikes, optimize what matters.

    Ready to monitor your AI agents?

    Set up AgentShield in 5 minutes. Free plan available.

    Start for Free →