Comparison
Croft vs n8n
Both run workflows on a schedule against your accounts. The difference is who builds the graph, and who keeps the thing running.
The short answer
n8n is a node-based workflow builder you can host yourself. You open a canvas, drag nodes onto it, wire them together, and either run the instance on your own infrastructure or pay for n8n Cloud. It has a very large node library and a genuinely open community, and if you want to own the box your automation runs on, it is the obvious choice.
Croft is not a builder. You describe the problem in a sentence and an agent writes the workflow as TypeScript, which runs on our durable runtime with credentials we hold and a human granted. The canvas exists to read what ran, not to author it. There is no instance to operate, no queue to size, and no upgrade to schedule.
Pick n8n if you want to build the graph and own the server. Pick Croft if you want to describe the outcome and never think about either.
What n8n is
n8n is an open workflow automation tool with a visual node editor, several hundred integrations, and a self-hosting story that is the main reason most teams choose it.
Self-hostable workflow automation. This page is our reading of n8n's own published material as of — we deliberately quote no prices of theirs, because a stale price is worse than none. Check their site for the current details.
Croft and n8n, side by side
| Croft | n8n | |
|---|---|---|
| Who writes the workflow | You describe the problem in your own words. An agent writes it through MCP, and it is live in the same conversation. | You do, by dragging nodes onto a canvas and wiring them up. Code nodes cover what the nodes cannot express. |
| What a workflow actually is | TypeScript, compiled to WASM, run in a sandbox with no ambient network and no ambient credentials. | A JSON graph of nodes, executed by the n8n runtime. |
| Where the connections live | Held by Croft, scoped to one workspace, granted by a human and revocable in one click. Never inside workflow code. | Stored in your n8n instance, entered per credential by whoever set that instance up. |
| When a run is interrupted | It carries on. Runs are checkpointed and replayable, and survive deploys, crashes and losing the machine underneath them. | Executions can be retried, but how well an interrupted run recovers is a property of the instance you are running and the queue behind it. |
| What you operate | Nothing. No instance, no workers, no queue, no upgrade window. | The n8n instance, if you self-host — plus the database, Redis and worker processes queue mode needs. Or n8n Cloud, and then nothing. |
| What you are billed on | A flat price per workspace, metered only on runs per week. People, workspaces and workflows are unlimited on every plan. | n8n Cloud is tiered on workflow executions — a whole run, not each step. Self-hosting is free of licence cost under their Sustainable Use Licence and costs you infrastructure and attention instead. |
| What an agent can do | Everything. Every capability is an MCP tool before it is a screen, so Claude can build, run, inspect and fix workflows directly. | A workflow can call a model, and there is an MCP node for reaching tools. Building and changing the workflows themselves is still a person at the canvas. |
| Getting your work back out | Your workflow TypeScript, your tables and your files export at any time. | Workflow JSON exports cleanly. It only runs inside n8n. |
The four differences that matter
You are not the builder
This is the whole difference and everything else follows from it. In n8n, the quality of an automation is bounded by how much time somebody spends on the canvas. In Croft, you say "every Friday, chase whoever hasn't filed their hours, and don't chase the people who have" and the agent writes it, runs it, and fixes it when the sheet's column headings change. Nobody on your team has to learn what a Merge node does.
Nothing to run, and nothing to keep running
Self-hosting n8n is cheap until it isn't. The instance needs a database, backups, a Redis and worker processes once one execution at a time stops being enough, an upgrade cadence, and someone who notices when it stops. Croft's runtime is Elixir on the BEAM and it is our problem, not yours: a workflow scheduled for Sunday runs on Sunday whatever happened to us on Thursday.
Credentials you grant, not credentials you paste
In n8n a credential is a record inside the instance, and anyone who can edit a workflow can point a node at it. In Croft, connections are held centrally, scoped to a single workspace, and a workflow can only touch the ones a human explicitly granted it. Revoking is one click, and the workflow code never sees a secret — there is no ambient network and no ambient credential inside the sandbox to leak into.
Real code, without a repository
A node graph is a config format, so anything it did not anticipate becomes a Code node with the rest of the logic stranded around it. Croft workflows are TypeScript from the start — loops, branches, error handling and retries are just code — but there is no repo, no CI and no deploy: the agent writes it into the workspace and it is live. You can still export every line of it.
Which one you want
Use n8n when
- — You need the workflow on hardware you control, in your own network, or air-gapped.
- — You want to build and see the node graph yourself — that is the part you enjoy.
- — You need one of n8n's several hundred integrations today, and it is a long-tail one.
- — You already have someone who runs infrastructure and is happy to add one more service.
Use Croft when
- ✓ Nobody on the team wants to build or maintain the automation itself.
- ✓ Nobody on the team wants to run a server, and there is no IT department to ask.
- ✓ You want agents like Claude to build and change the workflows directly, over MCP.
- ✓ You want the runs to still be running in two years without anyone having touched them.
Common questions
Is Croft an n8n alternative?
Yes, for teams who want the outcome rather than the canvas. Croft covers the same ground — scheduled workflows that read and write your real accounts — but an agent authors the workflow instead of a person wiring nodes, and Croft runs it for you rather than shipping you something to host.
Can I self-host Croft the way I self-host n8n?
No. Croft is a hosted service. What you can take with you is the part that matters: the workflow TypeScript, your tables and your files all export at any time, so the logic is never trapped in our format.
Does Croft have as many integrations as n8n?
No, and it is not trying to. n8n has several hundred nodes; Croft ships a curated set of managed connectors — email, Google Sheets, Slack, GitHub, Stripe, Attio, X, web search — each with credentials we hold and refresh, plus an allowlisted HTTP connector for anything else. The trade is breadth for connections that a non-technical person can grant and revoke safely.
How does Croft's pricing compare to n8n's?
n8n Cloud meters workflow executions and self-hosting trades licence cost for infrastructure cost. Croft charges a flat price per workspace and meters only runs per week — people, workspaces and workflows are unlimited on every plan, including the free one. Neither of us charges per step, which is the thing that makes Zapier expensive.
The fastest way to tell the difference
Say the thing your team keeps having to do, in your own words. An agent writes the workflow, it runs this week, and you have something to compare against n8n rather than a table to read. The free plan needs no card.