How to set a hard OpenAI budget limit in n8n (without writing a single line of code)

If you run AI workflows in n8n, you've probably noticed there's no built-in way to say "stop this workflow if it spends more than $50 this month." n8n's OpenAI and Anthropic nodes don't expose a budget knob. The provider's own dashboard offers alerts, not stops. And if you're running a dozen workflows on a single API key, you can't even tell which one is burning the budget.

Here's the fix, end-to-end, in under five minutes — and it doesn't require touching code.

What you'll end up with

• Per-workflow cost breakdown — every workflow tagged separately on a dashboard, so you know who's spending what.
• Hard budget cap per workflow — when a workflow hits its monthly USD limit, n8n gets a 429 back and the request never reaches OpenAI. You don't pay for blocked calls.
• Full request log — model, tokens in/out, cost, latency, timestamp.
• Zero code changes — same JSON body, same auth header, just a different URL.

Why n8n's native nodes won't work for this

n8n's OpenAI and Anthropic nodes are convenient but they don't let you change the base URL, and they don't expose any kind of cost field. You can read the response, sure, but to know what a call cost you'd have to maintain your own pricing table per model — and even then, you can't block calls based on a running total.

The cleanest path is the one n8n itself uses for every "we don't have a node for that" case: the generic HTTP Request node.

Step-by-step setup

Method:           POST
URL:              https://api.promptcost.io/anthropic/v1/messages
Authentication:   None  # we use header auth instead
Send Headers:     Yes
Send Body:        Yes
Body Content Type: JSON

x-api-key:    sk-ant-••••••••••     # your provider key
cg-key:       sk-pc-••••3f9a         # your PromptCost key
cg-agent:     {{$workflow.name}}      # tag this workflow
Content-Type: application/json

{
  "model": "claude-haiku-4-5",
  "max_tokens": 1000,
  "messages": [
    { "role": "user", "content": "{{$json.prompt}}" }
  ]
}

Bonus: how to handle the 429 in n8n

You'll want your workflow to do something graceful when the budget is exceeded — send a Slack alert, fall back to a cheaper model, or just skip the run. The HTTP Request node returns a 429 with a body like:

{
  "error": "budget_exceeded",
  "message": "Monthly cap of $50 reached for agent 'lead-scorer'",
  "agent": "lead-scorer"
}

In n8n, set the HTTP Request node's "On Error" setting to "Continue (using error output)", then branch on the error output to a Slack node, an email, or a fallback path.

What about streaming responses?

n8n's HTTP Request node doesn't support SSE streaming, so you'll use non-streaming responses. PromptCost handles both, but for n8n specifically, just leave the body as standard non-streaming JSON.

What about Make.com, Python, or anywhere else?

Same pattern. The proxy is a thin layer in front of OpenAI and Anthropic, so anywhere you can make an HTTP request — Make.com's HTTP module, a Python requests.post(), a curl command — works identically. See the Make.com guide for that flow.

Further reading

• How a single retry loop burned $72K overnight
• Track AI costs per workflow in n8n — full guide
• Track AI costs per scenario in Make.com