n8n Review (2026): Is the 3.0 Beta Worth the Upgrade?

n8n shipped a 3.0 beta on October 27, and it is not the Agent Builder update this review has tracked since April. It is a cleanup release: breaking changes, a rebuilt image-editing node, and security defaults that got measurably stricter. The 4.5 rating holds. What changed is the timeline for trusting it in production, and the number you should budget against if you were planning around this review's August pricing figures, because that number moved too.

Here is the short version. If you can run Docker, self-host the free Community Edition and skip the meter entirely, something 3.0 now requires rather than merely recommends. If you cannot, the 50 EUR Pro plan is still the ops-outsourcing option once you clear 2,500 executions a month, though its included AI Workflow Builder credits are a fraction of what this review reported in August. The 3.0 beta is worth reading closely and testing in staging. It is not worth running in production yet, by n8n's own timeline: stable lands in early December. One platform still beats n8n for a non-technical operator, Make, and that has not changed.

n8n 3.0 lands in beta: what breaks, and what gets safer

n8n 3.0.0 shipped as an early-access beta on October 27, 2026, with the stable release scheduled for early December (Introducing n8n 3.0, October 27, 2026). The framing from n8n's own announcement is unusually blunt for a major version bump: this release is not new features. It is internal cleanup that "improves reliability, stability and security," clearing legacy code paths that had to wait for a breaking-change window (n8n blog, October 27, 2026). New features keep shipping on the weekly cadence regardless of version number, which is why this review is still pinned to 2.45.4, the last full stable release before the beta, dated October 26.

Source: blog.n8n.io, captured October 30, 2026.

The practical effects split into three buckets.

Self-hosting goes Docker-only. n8n 3.0 no longer publishes a runnable package to npm, so installs run with npm or npx n8n stop working entirely; n8n ships a one-line setup script as the fast path to a Docker instance, and existing Kubernetes or Helm deployments are unaffected (v3.0 breaking changes). For the solo-dev persona this review has recommended since April, that was already the plan. For anyone still running n8n off a bare npm install, 3.0 forces the move this review already argued for.

Legacy nodes get removed, not deprecated. The Function and Function Item (legacy) nodes, the Item Lists node, the legacy OpenAI node, older Agent modes, the old Cron and Interval nodes (replaced by the Schedule Trigger), and Edit Image node versions 1 and 1.1 are gone in 3.0; workflows that still use them fail outright, and published workflows using a removed node stop activating entirely (v3.0 breaking changes). Every instance on 2.39 or later has a Migration Report under Settings, which lists the affected workflows, auto-fixes many of them as a new version you review before publishing, and flags the rest for a manual edit. Run it before you upgrade, not after.

The Edit Image rebuild is the one most workflows will actually hit. Node version 2 "uses a different image-processing library," per n8n's own changelog, and the migration carries real behavior changes: blur settings convert automatically but the Sigma parameter is gone, text placement can shift, the Get Information operation returns a normalized result structure, PNG and TIFF output now ignore the Quality setting, and a resize to zero or negative dimensions fails instead of silently succeeding (v3.0 breaking changes). If your workflows touch image blur, resize, or metadata, test the output by hand after migrating, not just the fact that the workflow still runs.

Security defaults tightened across the board: encryption key rotation is on by default, Slack credentials now require a signing secret so a forged request can't impersonate a Slack trigger, the default SSRF block list is wider, decompression limits on the Compression node are lower to block oversized "zip bomb" archives, and unverified community packages are off by default (v3.0 breaking changes). None of these change what an existing, correctly configured workflow does. They close holes a misconfigured one could fall into.

Change in 3.0 beta	Who it affects	What to do before upgrading
Docker-only self-hosting	Anyone on npm or npx n8n	Move to Docker via the one-line setup or your own Compose file
Edit Image node v1/v1.1 removed	Any workflow with a blur, resize, or text node	Run the Migration Report, then re-test blur, resize, and PNG/TIFF output by hand
Legacy nodes removed (Function, Item Lists, Cron, Interval, legacy OpenAI, older Agent modes)	Older workflows built before 2025	Run the Migration Report; most have a direct, auto-migrated replacement
Slack credential requires a signing secret	Anyone using Slack triggers	Add the signing secret before upgrading, or the trigger stops authenticating

Source: n8n v3.0 breaking changes and n8n blog, October 27, 2026.

What changed since August, and what did not

The August update to this review covered 2.35.5, the release that gave the Agent Builder test runs, per-tool human review, and local token counting. Those three features are the reason the agent side of n8n stopped being a bare loop you had to armor yourself. They have not regressed or changed meaningfully in the ten weeks since, through 2.45.4 and into the 3.0 beta: test runs still execute an agent against generated input before it touches live traffic, and the per-tool review gate still pauses only on the tools you mark, routing an approval request to Slack, Telegram, or n8n Chat.

What did move: the two features this review flagged as "2.36 pre-release" in August have shipped. The MCP auto-expose toggle is live and documented at Settings > Instance-level MCP, under "Auto-exposing new workflows" (n8n MCP docs, fetched 2026-10-30). The per-tenant credential isolation this review also flagged for multi-tenant Cloud agents has not shown up anywhere in n8n's changelog or docs between then and the 3.0 beta; if that is your blocker, treat it as unresolved, not merely delayed.

Pricing: current euros, approximate dollars, and a credit allowance that shrank

Execution caps and per-seat pricing held flat through October: Starter 20 EUR/mo, Pro 50 EUR/mo, unchanged since July (n8n pricing, fetched 2026-10-30). What did not hold flat is the AI Workflow Builder credit allowance, the budget for using n8n's AI chat assistant to generate and edit workflows. This review cited 2,300 credits on Starter and up to 13,700 on Pro in August; the pricing page today lists 50 and 150.

Source: n8n.io, captured October 30, 2026.

Plan	Price (EUR, annual, n8n pricing)	USD approx	Executions/mo	AI Workflow Builder credits/mo	Hosting
Community Edition	0	$0	Unlimited	n/a	You self-host
Starter	20 EUR/mo	~$22/mo	2,500	50	n8n cloud
Pro	50 EUR/mo	~$54/mo	10,000	150	n8n cloud
Business	667 EUR/mo	~$720/mo	40,000	Coming soon, not included today	n8n or self
Enterprise	Custom	Custom	Custom	1,000 (cloud only)	n8n or self

Prices, execution caps, and AI Workflow Builder credit allowances per n8n's pricing page, as of 2026-10-30. The USD column converts at roughly 1 EUR = 1.08 USD, an approximate rate for budgeting, not billing.

Two things are worth naming directly. First, the credit cut is real and dated: whatever changed n8n's calculus between August and October, 50 credits a month on Starter covers meaningfully fewer AI-assisted workflow builds than 2,300 did, and if you were budgeting around this review's earlier figure, re-budget now. Second, Business still does not include any AI Workflow Builder credits, listed as "coming soon," which is a gap worth knowing about before you pay 667 EUR a month expecting it.

Executions remain the real wall, and the arithmetic has not changed. One workflow polling an API every fifteen minutes runs about 2,880 times a month, clearing the Starter cap by itself before any other automation fires. Pro's 10,000 absorbs most small-team loads; past that you are on Business at 667 EUR or self-hosting. A managed VPS is still the escape route: the Community Edition is a Docker image with no license fee and no execution meter, and a Cloudways droplet in the 11 EUR/mo range handles patching and backups without you also running a server by hand.

Agent Builder: production-ready yet?

Closer than in August, still not there. n8n's own documentation settles this without hedging: "Agents are in Preview and may change in future releases. Avoid relying on them in production workflows" (n8n MCP docs, fetched 2026-10-30). That line has not changed since this review first flagged the Agent Builder as a Preview feature, through the ten weeks of weekly releases that followed and the 3.0 beta.

What's real inside that Preview: test runs execute an agent workflow against generated input so a broken tool node fails in the editor, not in front of a live user. Per-tool human review lets you gate only the irreversible tools. Say your agent reads a CRM, drafts a reply, and sends an email; enable review on the send-email tool only, and the agent runs the read and draft unattended, pauses before the send, and drops an approval request into Slack showing the exact recipient and body. A reviewer approves or denies, and on denial the model is told no and continues (n8n HITL docs). Token counting estimates the per-run budget before you execute, so a context-heavy run is visible before it bills, not after.

What's still missing is the reliability layer the Preview label is warning you about. The AI Agent node hands you the loop; it does not absorb malformed model output for you, and a dependable production agent still means building the retry path yourself. Turn these features on in staging. The vendor is telling you, in writing, not to bet a customer-facing SLA on them yet.

MCP integration depth

n8n's MCP server is no longer carrying the hedge its Agent Builder still does. Instance-level MCP access, the feature that lets Claude, ChatGPT, or another MCP client search, run, and build n8n workflows from a single connection, shows no Preview label in n8n's current docs, and the auto-expose toggle this review tracked as pre-release in August is live: enable it once at Settings > Instance-level MCP and every new workflow becomes available to connected clients without a per-workflow step (n8n MCP docs, fetched 2026-10-30). Exposing an agent, as opposed to a workflow, to an MCP client still carries the same Preview warning as the Agent Builder itself, which is the one place the hedge survives.

If you are wiring n8n into an agent stack that speaks MCP, the 2026-07-28 stateless MCP spec is the context worth reading first, because it changes how those server connections hold state. No other automation platform in this review's coverage lets a client auto-register new workflows to an MCP server with a single settings toggle. That gap is the clearest reason an agent-first team picks n8n over the alternatives.

The verdict, per persona

The rating stays 4.5. n8n's core economics (self-hosting deletes the per-execution meter, the MCP server is first-party and now out of Preview) have not moved, and the 3.0 beta is a cleanup release, not a reason to upgrade the score or downgrade it. The pick, by buyer:

Solo dev, comfortable with Docker. Run the free Community Edition on a managed VPS. An 11 EUR/mo Cloudways box gives you unlimited executions and the full Agent Builder for zero license cost, and 3.0 making Docker mandatory for everyone else just catches up to where this persona already was.

Ops team, 2,500-plus executions a month, no one to own uptime. Buy Pro at 50 EUR/mo (pricing), but plan around 150 AI Workflow Builder credits a month, not the 13,700 this review cited in August. If you have even one engineer who can run a container, self-host instead and put the difference into monitoring.

Non-technical operator who will not touch a container. Make still wins. The Agent Builder's test run and per-tool review gate were the friction reducers that could have closed the gap, and n8n's own docs still mark that feature Preview eleven weeks after it shipped. Revisit this call once the Agent Builder label changes, not before.

Enterprise with multi-tenant agents. The credential-isolation fix this review has been waiting on since August has not surfaced in n8n's changelog or docs. If that is your blocker, the answer is still "not yet," and self-hosted Business or Enterprise on infrastructure you control is the safer path until n8n documents a fix.

Agent-first team that wants an MCP-native workflow. n8n is the clearest choice in the category. The MCP server's auto-expose toggle and its graduation out of Preview status are real, shipped advantages nothing else in this category matches.

What to watch

Two triggers would move this review again. First is n8n 3.0 reaching stable, due early December per n8n's own announcement; that is when the Docker requirement, the Edit Image rebuild, and the tightened security defaults stop being beta-only and start being what every production instance runs. Second is the Agent Builder's Preview label coming off, which is the signal a customer-facing team needs before trusting agent workflows past staging. Until then, 2.45.4 is the version to run, and the pick stands.

Ready to start? Stand up the self-hosted Docker image from n8n if you want the pricing argument to hold, or run it managed on Cloudways to get the unlimited-execution economics without owning the ops stack.

Review history

This review is updated in place as the product changes. Earlier versions:

• October 30, 2026: Is the 3.0 Beta Worth the Upgrade? (current version)
• August 20, 2026: Is the 2.35 Agent Builder Worth the Upgrade?
• July 2, 2026: 2.26 Is Out, the MCP Server Is Real, and Pricing Moved to Euros
• May 1, 2026: The Open-Source Automation Platform Worth the Setup Time