n8n-nodes-cloudflare-ai-gateway

An n8n community node that exposes Cloudflare AI Gateway as a Language Model, so it can be attached to the AI Agent and other LangChain nodes in n8n that need an LLM.

It connects to Cloudflare's OpenAI-compatible endpoint and gives access to all models served through AI Gateway — Cloudflare Workers AI (@cf/...) and third-party providers (openai/..., anthropic/..., google/..., xai/..., …) — through a single credential.

Installation

In n8n: Settings → Community nodes → Install, then enter n8n-nodes-cloudflare-ai-gateway.

Or via npm:

npm install n8n-nodes-cloudflare-ai-gateway

Credential setup

Create a Cloudflare API token with these permissions (Account-scoped):

| Permission | Scope | Used for | |---|---|---| | AI Gateway - Run | Account | Inference (chat completions) | | AI Gateway - Read | Account | Read gateway config | | Workers AI - Read | Account | @cf/... models + model list |

Then in n8n create a Cloudflare AI Gateway API credential with:

• Account ID — your Cloudflare Account ID (how to find it)
• Gateway ID — your gateway name (default is created automatically)
• API Token — the token above

Usage

1. Add the Cloudflare AI Gateway node as a sub-node of an AI Agent (or any node that takes a Language Model).
2. Model:
   • From List — Workers AI text-generation models, loaded live from the Cloudflare API.
   • ID — type any model the gateway serves, e.g. openai/gpt-4.1, anthropic/claude-sonnet-4, google/gemini-3-flash, or @cf/moonshotai/kimi-k2.6.

Billing

• Workers AI models (@cf/...): billed per Workers AI pricing. Work out of the box.
• Third-party models (openai/..., anthropic/..., …): require Unified Billing credits loaded on your account, or BYOK (stored keys). Without them you will get Insufficient balance (error 2021).

Under the hood

The node wraps LangChain's ChatOpenAI pointed at Cloudflare's OpenAI-compatible endpoint:

POST https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/v1/chat/completions
Authorization: Bearer {CF_TOKEN}
cf-aig-gateway-id: {gateway_id}

This keeps it compatible with tool calling (agents) and standard chat-completion parameters.

Development

This node was developed using OpenCode as an AI coding copilot, powered by the GLM-5.2 language model.

License

MIT