Original URL: https://getagenttools.com/tools/n8n/ \# orchestration # n8n RUNNING n8n is a workflow automation tool built around a visual canvas. You wire nodes together, each node does one job, and the workflow runs on a trigger, a schedule or a webhook. What makes it worth a review here is the agent layer sitting on that canvas: an AI Agent node with tools, and MCP wired in both directions. - Verdict: **RUNNING**. Visual workflows that carry production pipelines, not demos. - Billing is per workflow run. One execution covers the whole workflow, however many nodes fire. - The AI Agent node takes a chat model plus tools and decides what to call. - Self-hostable under a fair-code licence, or hosted from 20 euro a month billed annually. - Best for teams that want visual ops, skip it if your team prefers code-first. New here? Browse [the full tool list](https://getagenttools.com/tools/), or read the other half of this decision: [the Make review](https://getagenttools.com/tools/make/). If you are deciding whether n8n is the automation layer for your agents, here is the operator view. GetAgentTools lists only what I actually use, with a verdict and no sponsored placements. Verdict first, then the detail. **Verdict: RUNNING.** In one line: visual workflows with an agent layer on top. ## What n8n is and how it works A workflow in n8n is a graph of nodes on a canvas. Something starts the run, usually a webhook, a schedule or an app event. Data flows left to right through the nodes, and each node does one small thing: call an HTTP endpoint, write a row, branch on a condition, run code, call a model. You watch the data move through the canvas while you build, which is the part that makes it teachable. You can run it two ways. The community edition is self-hosted, you install it on your own server and it costs you the server. n8n Cloud is the hosted version where they run the infrastructure and bill you by executions. Both give you the same canvas. One number decides most of the cost question: **an execution is one run of the whole workflow**. The pricing page states it plainly, that it does not matter how many steps are in the workflow or how much data it processes. A workflow with forty nodes and a workflow with four nodes bill the same per run. Remember that line, it comes back in the comparison with Make. ## Why n8n is RUNNING in my stack The AI nodes matured enough to carry production pipelines. Visual ops the whole team can read. For a long time the AI nodes in visual tools were a demo layer. You could bolt a model call onto a workflow, but anything that looked like an agent, calling tools, deciding what to do next, retrying, meant dropping into code. That gap closed. The AI Agent node now sits on the same canvas as the rest of the workflow, the tools it can call are nodes you already know, and the run history shows what it did. The second reason is legibility. **Visual ops the whole team can read.** A Python script is faster for me and opaque for everyone else. When a workflow lives on a canvas, the person who owns the process can open it, see where the data came in, see where it went, and tell me which step is wrong. That is worth more than the twenty minutes the script would have saved. Third: self-hosting is a real option, not a marketing bullet. When a workflow touches data that should not leave your server, you run the same tool on your own box and nothing changes about how you build. ## The agent layer: AI Agent node, tools and MCP The AI Agent node is the piece that matters here. You connect a chat model and at least one tool as sub-nodes, and the agent decides which tools to call to finish the task. The docs are direct about the requirement: at least one tool sub-node, or there is nothing for the agent to do. Since version 1.82.0 there is no agent type dropdown, every AI Agent node runs as the Tools Agent that used to be the recommended pick. n8n workflow: a trigger into an AI Agent node with chat model, memory and tools as sub-nodes, plus MCP nodes The AI Agent node is one node on a normal canvas. A chat model and at least one tool hang under it as sub-nodes, and MCP works both ways: the MCP Client Tool calls outside servers, the MCP Server Trigger exposes your workflows to any MCP client. MCP goes both ways, which is the part people miss. The MCP Client Tool lets your n8n agent call external [MCP servers](https://getagenttools.com/tools/mcp-servers/), so anything already exposed over the protocol becomes a tool for the agent. The MCP Server Trigger flips it: your workflows become an MCP server that outside clients connect to over SSE or streamable HTTP, with bearer token or header authentication. That trigger connects to tool nodes rather than ordinary workflow nodes. Practically, that means n8n can be the tool layer for an agent you run somewhere else. You build the integration once on the canvas, expose it, and [a terminal agent like Claude Code](https://getagenttools.com/tools/claude-code/) can call it without you writing a connector. ## Who n8n is best for **Teams that want visual ops.** If the process you are automating belongs to a team rather than to you, the canvas earns its keep. Marketing, ops and support people can read a workflow, point at the broken step and suggest the missing branch. It also fits when the data has to stay on your own infrastructure, because self-hosting is a supported path rather than a hack. ## When to skip n8n **Your team prefers code-first.** If everyone who will touch the automation writes code, a repo with tests and a scheduler will beat a canvas. You get version control that actually diffs, code review, and no dragging. Visual tools trade precision for legibility, and when nobody in the room needs the legibility, you are paying the trade for nothing. I would also skip it for a single one-off job. Thirty lines and a cron entry beat standing up a workflow tool you then have to keep alive. ## How much does n8n cost? Self-hosted community edition: free, you pay for the server. Hosted plans as listed on the n8n pricing page, billed annually: - Starter, 20 euro a month, 2,500 executions a month - Pro, 50 euro a month, 10,000 executions a month - Business, 667 euro a month, 40,000 executions a month, plus a self-hosted option - Enterprise, custom pricing and custom execution volume Prices move, so treat these as the shape rather than the quote and check [the n8n pricing page](https://n8n.io/pricing/) before you commit. The shape is what matters: you buy runs, not steps, so adding nodes to an existing workflow does not raise the bill. On licensing, be precise. n8n is source-available under the Sustainable Use License, which the project calls fair-code. It grants use, modification and distribution for your own internal business purposes or non-commercial use, and it stops you selling it or running it as a service for others. For a company automating its own work, that licence is a non-issue. If your plan is to resell someone else's automation platform, read it first. ## How to build your first n8n workflow 1. Pick a process that already runs on someone's manual copy-paste. A form filling a sheet, an alert landing in a channel. Boring first. 2. Start hosted unless you have a reason not to. Self-host once you know the tool stays in your stack, so the first week is about the workflow rather than about Docker. 3. Build the plain version with no AI in it. Trigger, two or three nodes, done. Watch the data move through the canvas and fix the shape of it. 4. Only then add the AI Agent node, with a chat model and the smallest useful set of tools. An agent on top of a workflow that already works is debuggable. An agent on top of nothing is a guess. 5. Put a human approval step in front of anything that spends money, writes to production or sends a message to a real person. ## n8n vs Make: which one These two land on the same shortlist, so here is the split I use. The clearest difference is the billing unit, and it decides more than the feature lists do. n8n bills **one execution per workflow run**, whatever the node count. [Make bills per module action](https://getagenttools.com/tools/make/), so a four-module scenario costs four credits every time it fires. Wide workflows with many steps are cheap in n8n and get expensive in Make. Simple two-step scenarios firing rarely are cheap in both, and Make will get you there faster. After that: n8n if you want to self-host, run code inside the workflow, or expose your automations to agents over MCP. Make if you want the shortest path from nothing to a working integration, on someone else's infrastructure, with an app catalogue that covers more SaaS out of the box. I run both, for different jobs. That is not a dodge, it is the honest answer for anyone whose automations are not all the same shape. ## n8n FAQ ### What is n8n? n8n is a workflow automation tool with a visual canvas. You drag nodes onto the canvas, connect them, and each node does one job: catch a webhook, call an API, transform data, run an AI agent. It ships as a hosted cloud product and as a self-hosted community edition you run on your own server. ### Is n8n free? The self-hosted community edition is free. You pay for the server it runs on and you own the upgrades and backups. n8n Cloud starts at 20 euro a month billed annually for the Starter plan, which includes 2,500 workflow executions a month. Check the pricing page for current numbers, they move. ### Is n8n open source? The code is public but the licence is not an OSI open-source licence. n8n ships under the Sustainable Use License, which n8n calls fair-code. You can use, modify and self-host it for your own internal business purposes. You cannot sell it or host it as a service for other people. ### What counts as an execution in n8n? One run of the whole workflow. The pricing page is explicit that it does not matter how many steps the workflow has or how much data it processes, the run still counts as a single execution. That is the opposite of per-step billing, and it is why wide workflows are cheap here. ### Can n8n run AI agents? Yes. The AI Agent node takes a chat model and at least one tool as sub-nodes, then decides which tools to call to finish a task. Since version 1.82.0 there is no agent type to pick, every AI Agent node runs as the Tools Agent that used to be the recommended setting. ### Does n8n support MCP? In both directions. The MCP Client Tool lets an n8n agent call external MCP servers, and the MCP Server Trigger turns your n8n workflows into an MCP server that outside clients can connect to over SSE or streamable HTTP, with bearer token or header authentication. [Back to the list](https://getagenttools.com/tools/)