AI agent for n8n workflows

The AI that actually knows n8n

Build, validate, and deploy n8n workflows by describing them. n8nAI works from real node schemas and checks every workflow before it reaches your instance — no hallucinated parameters, no copy-pasting JSON.

Start for freeSelf-host on GitHub

Works with n8n Cloud and self-hosted n8n — only the URL differs.

You:“Poll this RSS feed every hour and post new items to Slack”
RSS Feed
Filter
Slack
Validated — no missing fieldsDeployed to your n8n

Works with your AI client

Claude CodeClaude DesktopCursorCodex CLIChatGPTMIT open sourcen8n CloudSelf-hosted n8n

Everything an AI needs to get n8n right

Not a chatbot with opinions about n8n — an agent wired into it.

Real node schemas

The agent reads the actual parameters of n8n's 400+ nodes before it writes a single field — instead of guessing names that don't exist.

RSS Feed
Slack

Slack · Send Message

Channel#news
Message{{ $json.title }}
CredentialSlack account

Validation before deploy

Every workflow is structurally validated — connections, required fields, expressions — before it ever touches your instance.

Webhook
Filter
Sheets
Workflow valid — deployed

Your instance, directly — with self-healing

Connects to your n8n over its API: lists workflows, reads executions, deploys changes. When a run fails, the agent reads the execution log and patches the workflow itself.

Executions · rss-to-slack

Error in node “Filter”#483
Agent read the execution logget_execution
“Filter” patchedupdate_workflow
#484 — Succeeded · self-healed

Chat or your own AI client

Use the built-in chat, or plug the MCP server into Claude Code, Cursor, or any MCP client and keep the tools you already use.

Poll this RSS feed hourly, post new items to Slack…Sync new Stripe customers to Google Sheets…Alert me in Telegram when a workflow fails…
Agent

From sentence to running workflow

01

Connect your n8n

Paste your instance URL and an API key from Settings → API. Cloud or self-hosted, both work.

02

Describe the workflow

“Poll this RSS feed every hour and post new items to Slack.” The agent picks the right nodes and fills in real parameters.

03

Built, validated, deployed

The workflow is checked for structural errors, then created in your instance — ready to run and easy to iterate on.

Life before and after

Generic AI chat

  • The model guesses node parameters from old forum posts
  • Copy-paste JSON between the chat and the n8n editor
  • Import, hit an error, fix, re-import — repeat
  • Debugging means pasting screenshots into a chat
  • One generic AI chat for everything

n8nAI

  • Reads the real schemas of 400+ nodes, matched to your n8n
  • Workflows are created directly in your instance
  • Structural validation catches errors before deploy
  • The agent reads execution logs and self-heals the workflow
  • Built-in chat or the MCP client you already use

MIT licensed

Open source. Self-host it, audit it, fork it.

One repo, one Docker Compose stack — the exact same app whether you run it yourself with your own AI provider key or let us host it. No feature gap between the two, and nothing to migrate if you switch.

View on GitHub

Pricing without surprises

Self-host

Free forever

  • Unlimited usage
  • Your own AI provider key
  • Full app: chat, dashboard, MCP server
  • MIT licensed, runs on your hardware
Clone the repo
Popular

Hosted Free

$0 / month

  • 200 credits included
  • No API key needed — we run the AI
  • Built-in chat + MCP tokens
  • Connect n8n Cloud or self-hosted
Start for free

Hosted Pro

Coming soon

  • Uncapped usage
  • Everything in Hosted Free
  • Priority support
Sign up to be notified

Questions, answered

Does it work with n8n Cloud and self-hosted n8n?

Both. n8nAI talks to your instance through the standard n8n API — the only difference between Cloud and self-hosted is the URL you paste when connecting.

What is MCP and do I have to care?

MCP (Model Context Protocol) is the open standard that lets AI clients like Claude Code, Cursor, and ChatGPT use external tools. If you use the built-in chat you never touch it; if you prefer your own AI client, you create an MCP token in the dashboard and point your client at our server.

Do I need my own AI API key?

On the hosted version, no — we run the AI and meter usage in credits. If you self-host, you bring your own provider key and usage is unlimited.

How is this different from pasting workflow JSON into ChatGPT?

A general chatbot guesses node parameters and produces JSON you still have to import, fix, and re-import. n8nAI works from n8n's real node schemas, validates the structure before deploying, and creates the workflow directly in your instance.

Is my n8n data safe?

Your n8n API key is stored encrypted and used only to call your instance on your behalf. The project is open source, so you can audit exactly what it does — or self-host so nothing leaves your infrastructure.

Is it really free?

Yes. Self-hosting is free and unlimited under the MIT license. The hosted Free plan includes 200 credits; a paid uncapped plan is coming soon.