n8n

Section 2 — Is this for you?

n8n is the automation tool people pick when they want more control than Zapier or Make allow — and don't mind getting their hands a little dirty. It looks like a visual workflow canvas, but it hides a real code layer underneath. You can drop in a JavaScript node whenever the UI stops being enough.

It's also the one platform on this list you can self-host for free, indefinitely, with no execution limits. That single fact changes the math on almost every big decision: data residency, cost ceilings, compliance, vendor risk.

n8n is the right pick when:

• You have a developer in the loop (or you are one), and you want something more flexible than pure no-code
• You need to self-host — for GDPR, HIPAA preparation, air-gapped environments, or just to avoid per-execution billing
• Your workflows include real code steps — API calls, data transforms, custom logic
• You're building AI agents and want MCP server support that actually works in production
• You want predictable pricing at scale — Cloud plans bill per execution, not per step

n8n might work, but watch for:

• Your team is non-technical — expect a 2-4 week learning curve before people feel productive
• You need a signed BAA from the vendor — n8n Cloud does not sign one (self-hosting on HIPAA-eligible infra is your own job)
• You want to embed automation in a SaaS product you sell — the Sustainable Use License blocks that unless you buy an Enterprise/embed license
• You need a US-hosted Cloud option today — n8n Cloud is EU-only by default (Azure Germany West Central)

Look elsewhere when:

• You just want "connect Gmail to Sheets" and never touch code — Zapier is a better match
• You need a signed BAA with zero DIY — Power Automate under Microsoft's BAA is safer
• You want the absolute cheapest multi-step math — Make credits are hard to beat for simple high-volume workflows
• You want pay-per-compute-second billing — Pipedream is the pick for that model

Section 3 — The real story

n8n started in 2019 in Berlin, when Jan Oberhauser shipped it as an open-source project on GitHub. The pitch was simple and slightly contrarian: automation platforms should be self-hostable, not locked behind SaaS pricing. Fair-code, not proprietary.

The project grew quietly for a few years, then exploded. By 2025 it had 230,000+ active users, 3,000+ enterprise customers (Vodafone, Delivery Hero, Microsoft all on the list), and ARR that had grown 5x in twelve months.

Funding tracked the growth. A €55M Series B in March 2025 (Highland Europe, €300M valuation) turned into a $180M Series C seven months later (Accel lead, with Nvidia's NVentures joining, $2.5B valuation). New offices opened in New York and London. The team still ships code almost every day.

The identity stayed consistent through all of this. n8n is still fair-code, still self-hostable, still developer-friendly. The Cloud product exists as a convenience layer — not as the thing you have to buy.

What "fair-code" actually means

n8n uses the Sustainable Use License. It's not OSI-approved open source, but for most buyers the difference doesn't matter.

You can:

• Self-host n8n for free, forever
• Use it internally for any commercial purpose
• Modify the code, run it on your own infrastructure
• Charge for consulting or implementation work around it

You can't:

• Embed n8n inside a SaaS product you sell to customers
• Resell "n8n as a service" to external users
• Let untrusted external users trigger your workflows as a paid service
• Fork it and rebrand without a license

For the 95% of buyers who just want to run their own automations, this is indistinguishable from regular open source.

Section 4 — How it works

The execution model

n8n runs on Node.js. Every workflow is a graph of nodes — triggers and actions — and when a trigger fires, n8n walks the graph, passing data from one node to the next.

One full run of a workflow is called an execution. This is how Cloud billing works: one execution = one run, no matter how many nodes or steps are in the workflow. A 40-node workflow and a 4-node workflow both cost one execution.

Self-hosted is different — executions are unlimited by default. The hosting cost is whatever your server or container bill is.

The canvas

The visual builder looks like Make's — a 2D canvas with nodes you drag around and connect. But n8n's canvas is tuned for developers first. Every node exposes a raw JSON view, and most of them have parameters you can fill in with JavaScript expressions like {{ $json["email"].toLowerCase() }}.

The killer node is the Code node. Drop one in anywhere, write plain JavaScript (or Python, as of n8n 1.x), and you get full access to incoming data and the HTTP client. For anything the UI can't express, you just write the code.

Triggers: instant, polling, and scheduled

• Instant triggers fire on webhooks. Most apps that support n8n have webhook-based triggers.
• Polling triggers check an API on a schedule. The floor is 1 minute on self-hosted and most Cloud plans.
• Schedule triggers run on a cron-style timer. You can use full cron syntax, not just "every X minutes."
• MCP Server Trigger exposes the workflow as a tool any MCP-aware AI client can call.

Branching with IF and Switch nodes

n8n has IF nodes (two-branch split) and Switch nodes (n-branch split based on a value). Combine them with Merge nodes to rejoin branches after they've done different work.

There's no hard limit on branching depth. The only practical ceilings are memory and execution timeout, and on self-hosted you control both.

Section 5 — Where it shines

Self-hosting that actually works

This is the headline. n8n isn't "self-host the open core and pay for the real product." The community edition includes all 400+ native integrations, all code nodes, all branching primitives, and unlimited executions. You give up SSO, RBAC, audit logs, Git version control, and multi-environment — and those are the features enterprises actually pay for.

For a mid-size team running serious automation volume, the self-host/Enterprise split is an honest deal. The community edition isn't crippleware.

Code when you need it

Every other platform on this list treats code as an escape hatch. n8n treats it as a first-class primitive. The Code node, the Function node, and JavaScript expressions everywhere mean you don't have to choose between "no-code" and "give up and write a script."

This matters more than it sounds. Most real-world automations need 1-2 steps of custom logic — data reshaping, conditional formatting, a weird API call. In n8n you just write it. In Zapier or Make you either build a five-step contraption or reach for a custom function module.

AI-native, not AI-bolted-on

n8n has 70+ AI nodes built on LangChain. The AI Agent node lets you build LangChain-style tool-using agents that can call *any* n8n node as a tool. Vector store nodes for Qdrant, Supabase, PGVector, Azure AI Search, and a simple in-memory store ship with the platform.

If you're building an AI agent that needs to look something up, update a record, send an email, and summarize the result — that entire loop lives natively in n8n. No external orchestration layer, no glue code.

MCP support that's production-grade

n8n shipped both a MCP Server Trigger (expose workflows as tools) and an MCP Client Tool (consume external MCP tools inside workflows). Both are GA in 2026. You can run an n8n-hosted MCP server your Claude client talks to, or have an n8n workflow call Claude's tools as an MCP client.

Among automation platforms, this is the most complete MCP story. Only Make's cloud-hosted MCP server comes close, and Make's is read-only for the AI — n8n's MCP tools can do anything a workflow can do.

Per-execution pricing caps your bill

Cloud plans bill per execution, not per node or per step. A 40-node workflow and a 4-node workflow cost the same. This is a huge difference from Zapier (per successful task = per action) and from Make (per module run).

For any workflow with more than ~5 steps, n8n Cloud is cheaper per run than Zapier. And if the bill still hurts, you move to self-hosted and the unit cost drops to whatever your server runs at.

Community + ecosystem

5,834 community nodes (as of January 2026), 500+ npm-published community packages, an active Discord, and a template library with 1,500+ pre-built workflows. If someone in the world has integrated a tool with n8n, there's a good chance you can find their node.

Section 6 — What breaks most often

The learning curve is steep

Teams migrating from Zapier consistently report 2-4 weeks before they feel productive in n8n. Month one is frustrating — the canvas has more concepts, JavaScript expressions feel scary if you haven't used them before, and debugging returns JSON you have to read.

Month two is usually the turnaround. Teams that push through tend to end up faster in n8n than they ever were in Zapier, because the primitives are more powerful. But the first month is real.

Memory pressure on big data

n8n loads an entire dataset into memory during execution. If a node returns 50,000 items, all 50,000 sit in RAM until the workflow finishes. Large Postgres queries, big CSV imports, and mass-record updates are the common offenders.

The standard workaround is the SplitInBatches node — chunk the data into batches of 100-500 items and process them sequentially. This is the n8n equivalent of Make's "split scenarios with a cursor" pattern.

Rate-limit and loop ceilings

Common complaints from community users:

• Rate-limit hits around 130 items when looping through an API
• Process occasionally stops at 30-40 loops when iterating over 500+ items
• Some upstream APIs throttle n8n harder than Zapier because requests look less "spread out"

These aren't n8n bugs — they're interactions with real API limits. The fix is almost always SplitInBatches + explicit wait nodes.

No HIPAA BAA on Cloud

n8n Cloud is SOC 2 Type 2 certified, but the company does not sign Business Associate Agreements for HIPAA. If your workflow touches PHI and you need vendor-signed HIPAA coverage, Cloud is out.

Self-hosting on HIPAA-eligible infrastructure (AWS, Azure, GCP healthcare tiers) is the workaround — but the BAA is between you and the infrastructure provider, not n8n. That's more DIY than some healthcare teams want.

Cloud is EU-only by default

n8n Cloud runs on Azure Germany West Central. If you specifically need US data residency, you either wait for multi-region Cloud (in preparation as of early 2026) or self-host in US-based infrastructure.

For EU-based buyers this is actually an advantage — data never leaves the EU without you doing something unusual. For US buyers with strict "no data leaves the US" contracts, it's a hard stop.

The Sustainable Use License trips up embedders

If you're building a SaaS product and want n8n to be the automation engine behind it — so your customers trigger workflows you built — the Sustainable Use License blocks that unless you buy an embed license. This catches startups off guard who assume "open source means I can build on top of it."

The fix is straightforward (talk to n8n sales, get an Embed plan), but if you're mid-prototype and haven't budgeted for it, it's a surprise.

Queue mode has setup tax

Default n8n runs in "regular mode" — one Node.js process handles everything. To scale past a few hundred concurrent executions you flip on queue mode, which uses Redis to distribute work across multiple worker processes.

This works great once it's running. Getting it running requires Redis, configured environment variables, separate worker containers, and a main process that only handles UI. Not hard, but not a one-click flip.

Section 8 — Getting started reality

The fastest path: Cloud

Go to n8n.io, sign up for the Starter plan ($20/month, 2,500 executions), and you're in the canvas in under two minutes. Cloud is the right answer for the first 30 days even if you plan to self-host later — it removes the "did I configure this correctly?" anxiety while you learn the product.

The other path: self-host with Docker

One command:

`` docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n ``

Open http://localhost:5678, create an owner account, and you're running a private n8n instance. For production, you'll want Postgres (not SQLite), a reverse proxy with HTTPS, and persistent volume mounts — but the 30-second Docker run is enough to try the product.

Your first workflow

Click Workflows → Add workflow. You get an empty canvas with a "trigger" placeholder. Click it, pick On app event (Gmail, Slack, Airtable, whatever), and n8n walks you through connecting the account the first time.

Add a second node with the plus-button on the right side of the trigger. Pick an action — say, "Send a row to Google Sheets." Map the fields using the expression editor ({{ $json.email }} pulls the email field from the trigger's output). Click Execute workflow to test. When it works, flip the toggle in the top-right to activate it.

Where people get stuck

• Expressions are not JavaScript-y enough. You write {{ $json.field }} in parameter fields, not bare $json.field. Forgetting the braces returns a literal string.
• The item model confuses Zapier migrants. n8n passes arrays of items between nodes, not single records. If you came from Zapier, the "n8n thinks every output is a list" model takes a week to feel natural.
• Credentials live per-user, not per-workflow. If a teammate built a workflow with their Gmail credential and you import it, you need to re-auth with your own account.
• Self-host time zone bugs. The default container runs UTC. If you don't set GENERIC_TIMEZONE to your local zone, cron-style schedules run at the wrong time.

Free learning resources

The official n8n docs are strong. The template library at n8n.io/workflows has 1,500+ importable examples — searching "Zapier replacement" or "AI agent" both return good starters. The community Discord is active and helpful when you get stuck.

Section 9 — Migration translations

From Zapier

• Zap → Workflow. Both are graphs of triggers and actions. n8n's workflows can branch, loop, and call sub-workflows in ways Zaps can't.
• Task → Execution. Zapier bills per successful action. n8n bills per workflow run. A Zap that uses 5 tasks is 5 Zapier billings; the same work in n8n is 1 execution. Your bill usually shrinks significantly.
• Paths → IF / Switch + Merge. Zapier Paths become IF or Switch nodes in n8n. The big difference: n8n lets you merge branches back together after they've done different work.
• Looping by Zapier → SplitInBatches. SplitInBatches handles batches of data. No 500-iteration ceiling like Zapier's Looping, but you'll want to understand the chunk-and-continue pattern.
• Code by Zapier → Code node. n8n's Code node is more capable: full npm packages (on self-host), full access to workflow data, and no 10,000-character limit.
• Sub-Zap → Execute Workflow node. n8n has real sub-workflows. You call one workflow from another and pass data in and out — much cleaner than Zapier's webhook-chain pattern.

No automated import tool exists. Plan on rebuilding workflows manually.

From Make

• Scenario → Workflow. Very similar visual model. The biggest shift is billing — Make bills per module run, n8n bills per workflow run. Scenarios with lots of modules are much cheaper on n8n.
• Router → Switch node. Similar fan-out behavior. The Merge node is n8n's equivalent of Make's Aggregator.
• Iterator → SplitInBatches. Both loop over arrays. n8n's SplitInBatches handles the continue-where-you-left-off pattern a bit more explicitly.
• Tools modules → Code node or expressions. Make has specific utility modules for JSON, regex, date math. In n8n you typically do those inline in expressions or in a Code node.
• Data Store → Built-in workflow statics or external DB. n8n has simple workflow-level variables, but for persistent key-value state most teams use a Postgres or Redis connection.

From Power Automate

• Flow → Workflow. Trigger + actions — same concept.
• Premium connectors → Native or HTTP Request. Everything in n8n's catalog works on every plan. No premium tier gates.
• Dataverse → Postgres or external store. n8n has no equivalent to Dataverse. Teams usually wire up their own Postgres or Supabase instance.
• Environments → Git version control + separate instances. Enterprise n8n has Git-based version control and multi-environment support. On self-host, separate Docker instances per environment is the common pattern.
• Premium per-user pricing → Per-execution. The math usually goes in n8n's favor at any meaningful scale.

From Pipedream

• Workflow → Workflow. Very similar model. Both are code-friendly.
• Credit (compute seconds) → Execution. Pipedream bills compute time; n8n bills per full workflow run. Long-running workflows get much cheaper on n8n.
• Pipedream code step → Code node. n8n's Code node runs Node.js or Python, full npm access on self-host. Similar capability.
• Pipedream connector auth → n8n credentials. The UX is comparable. Both have good OAuth flows for major apps.

Section 10 — AI & MCP readiness

The AI Agent node

n8n's AI Agent node is built on LangChain. It runs a tool-using agent that can call any other n8n node as a tool. The pattern: the agent receives a task, decides which tools to use, calls them, reads the outputs, decides what to do next. Loop until done.

Because every n8n node can be wrapped as a tool, you can give an agent access to Gmail, Slack, HubSpot, your Postgres database, custom code, and more — all with the same wiring. This is the most direct "AI agent takes real action in your stack" implementation of any automation platform.

70+ AI nodes

Beyond the agent, n8n ships with dedicated nodes for OpenAI, Anthropic, Google Gemini, Azure OpenAI, Cohere, Hugging Face, Ollama, and more. There are chain nodes (summarize, extract structured data, question-answer over docs), memory nodes (Postgres, Redis, buffer), and vector store nodes (Qdrant, Supabase, PGVector, Azure AI Search, simple in-memory).

MCP Server Trigger

The MCP Server Trigger exposes a workflow as a tool any MCP-aware client can call. You build the workflow, set the MCP trigger's path, and point Claude (or Cursor, or any MCP client) at the URL. Every MCP trigger in your instance shows up as an available tool.

This works on self-host out of the box. On Cloud it's GA as of early 2026.

MCP Client Tool

The flip side is the MCP Client Tool, which lets an AI Agent node inside n8n consume external MCP tools. So an n8n-hosted agent can use tools hosted by other platforms, and vice versa. Both directions work.

The summary

n8n's AI story is the most complete in the automation space. You get LangChain-style agents, first-class MCP support in both directions, a wide catalog of model providers and vector stores, and all of it lives in the same canvas as your regular workflows. If you're serious about building production AI agents that *do things*, n8n is probably the most capable choice on this list.

Section 11 — Compliance at a glance

The facts

• SOC 2 Type 2: Yes — n8n Cloud, audited annually
• ISO 27001: Not currently certified
• GDPR: Yes — EU hosting (Azure Germany West Central) by default
• HIPAA / BAA: No on Cloud — n8n does not sign BAAs. Self-host on HIPAA-eligible infra is the workaround.
• Data residency: EU by default on Cloud; unlimited choice on self-host
• SSO (SAML, OIDC, LDAP): Enterprise plan only
• RBAC (role-based access): Enterprise plan only
• Audit logs: Enterprise plan only
• Git version control: Enterprise plan only
• Trains AI on your data: No — n8n doesn't use customer workflow data to train models
• Self-hostable: Yes — community edition, unlimited executions, no license cost

The fine print

n8n's compliance posture reflects its two-track product. Cloud is SOC 2 Type 2 with EU-only hosting — fine for most commercial use, but if you need HIPAA with a vendor-signed BAA, it's out.

Self-hosted flips the responsibility. You pick the infrastructure, you sign the BAA with AWS or Azure, you configure encryption at rest, you manage backups. n8n itself is not HIPAA-compliant out of the box, but teams regularly run n8n in HIPAA-eligible environments and handle the compliance layer themselves.

The features most enterprise buyers care about — SSO, RBAC, audit logs, Git version control, multi-environment — all live in the Enterprise plan. Starting price is around $20K/year; custom pricing above that depending on seats and support tier. This is where n8n makes its real money.

If you're comparing n8n to Zapier or Make for enterprise rollout, the deal usually comes down to: self-host gives you the compliance and cost control, but Enterprise is where the team-management features live. Most serious n8n customers end up on Enterprise-hosted (n8n-managed dedicated instance) or Enterprise self-hosted.