n8n Claude Code Workflows: 2 Ways to Combine Them

n8n Claude Code integration works two different ways, and mixing them up wastes a setup session. Claude Code can build your n8n workflows for you. It reaches into your n8n instance through the Model Context Protocol, or MCP, and writes the workflow itself. Or n8n can call Claude Code as one step inside a workflow you already built. It treats Claude Code like a coding subroutine, not the whole architect. Both methods use the same two tools, in opposite directions. Neither one replaces the other, and most serious n8n Claude Code stacks end up using both. This guide covers each method, when to reach for it, and the exact setup steps for a lean solopreneur stack.

Key Takeaways

• Method 1 (n8n-mcp): Claude Code builds and edits n8n workflows for you through an open-source MCP server that documents n8n’s full node library (n8n-mcp, GitHub, 2026).
• Method 2 (n8n-nodes-claudecode): n8n runs Claude Code as one workflow step, calling the installed CLI directly (johnlindquist/n8n-nodes-claudecode, GitHub, 2026).
• Method 2 needs a self-hosted n8n instance. Unverified community nodes don’t run on n8n Cloud (n8n Blog, 2026).
• Both methods need an active Claude Code login, either a Claude subscription or an API key. Neither is free to run at real volume.
• For the simpler route of using Claude as an AI Agent’s language model instead, our Claude and n8n integration guide covers that pattern.

I break down automation stacks like this every week. The sign-up sits a few sections down.

What n8n Claude Code Integration Means

n8n Claude Code describes two opposite integration directions, not one single feature. In the first direction, Claude Code sits outside n8n and builds it. You describe a workflow in plain language inside a terminal. Claude Code picks the nodes, sets the connections, and hands you something ready to import. In the second direction, Claude Code sits inside n8n as a worker. Your workflow reaches a specific step, hands Claude Code a coding or research task, and waits for the CLI to finish before moving on with the result.

The first turns Claude Code into an architect. The second turns it into a contractor for one step in a bigger process. This is different from wiring Claude into n8n’s AI Agent node as a language model, which is the more common Claude n8n pattern. Our Claude and n8n integration guide covers that route in full, plus a direct API method. This guide stays narrow on purpose. It only covers the two patterns that use Claude Code, the coding agent product, not the general chat model behind the AI Agent node.

Why This Distinction Matters for Solopreneurs

Most solopreneurs don’t need a dedicated developer to build automation anymore. That’s the whole promise of n8n. But once a workflow needs real logic, not just moving data from one app to another, you hit a wall without code. That’s where Claude Code changes the math.

Method 1 removes the wall at build time. You stop hand-placing nodes on a canvas and start describing outcomes instead, the same way you’d brief a contractor. Method 2 removes the wall at run time. A workflow that used to fail on anything unstructured, like a messy customer email or a half-formed pull request, can now hand that mess to Claude Code. It gets a structured answer back. Together, the two methods turn n8n from a tool that only handles predictable steps into one that also handles judgment calls. You never write a custom app to get there.

Method 1: Let Claude Code Build Your n8n Workflows

n8n-mcp is an open-source Model Context Protocol server built for this exact job. It gives Claude Code, Claude Desktop, Cursor, and Windsurf structured access to n8n’s node library, including each node’s real properties and operations. The project documents more than 2,500 n8n nodes, covering both n8n’s core set and its community add-ons, plus templates pulled from real published workflows. Instead of hand-writing workflow JSON from memory, or guessing at a node’s exact parameters, you describe the outcome you want in plain English. Claude Code looks up the right nodes, wires the connections, and hands you a workflow ready to use.

Setup takes a few minutes if Claude Code is already installed. Add n8n-mcp as an MCP server with a single command inside Claude Code’s CLI, using the project’s published config. Two modes exist. In documentation mode, Claude Code only reads node data. It designs the workflow and you import the result by hand. In management mode, you add your n8n instance’s API URL and API key as environment variables. Claude Code can then push the finished workflow straight into your n8n instance, skipping the manual import step entirely. It can also read back your existing workflows to debug or extend them.

This mode shines for a first draft. Tell Claude Code, in plain language, to build a workflow that watches a Gmail label and posts a daily summary to Slack. It picks the Gmail Trigger node, a Filter node, and a Slack node, then wires them in the right order. It uses real, validated node parameters instead of ones it half-remembers. You still review the result. Claude Code is fast at assembling the right shape of a workflow. It is not infallible about your specific business logic, your exact Slack channel naming, or which emails actually count as urgent.

Method 2: Run Claude Code Inside an n8n Workflow

n8n-nodes-claudecode is a community node built by developer John Lindquist. It wraps the Claude Code SDK so a workflow step can call Claude Code directly, the same way it might call any other n8n node. Where Method 1 uses Claude Code from the outside, this method puts Claude Code to work from inside an n8n canvas you already control.

The node needs the Claude Code CLI installed and logged in on the same machine that runs n8n. Install it with npm, either straight into your n8n instance’s custom nodes folder or through n8n’s Community Nodes settings page. Each call can maintain a persistent session, so Claude Code keeps context between steps in a longer workflow instead of starting fresh every time it runs. You can also set exactly which tools and file paths Claude Code is allowed to touch during that session, which matters once it’s running unattended.

This pattern fits jobs that need judgment, not just data movement. Community examples include a workflow that reviews a new pull request and drafts suggested fixes. Others write fresh documentation for a changed API, or port one legacy file at a time to a modern framework. In each case, n8n handles the trigger, the retries, and the audit trail. Claude Code handles the part that actually requires reasoning about code, which is the part n8n’s own nodes were never built to do.

Method 1 vs. Method 2: Which Should You Use?

Question	Method 1: n8n-mcp	Method 2: n8n-nodes-claudecode
Direction	Claude Code builds n8n	n8n runs Claude Code
Best for	Designing a new workflow fast	A judgment-heavy step inside an existing workflow
Where it runs	Your terminal, alongside Claude Code	Inside n8n itself, as a node
Hosting requirement	Any n8n instance, cloud or self-hosted	Self-hosted n8n only
Setup effort	One MCP config command	CLI install plus community node install

Start with Method 1 if you’re building something new and want a working draft fast. Reach for Method 2 once a workflow is already running and needs one smart step Claude Code can own, like reviewing text or generating code. You stop rebuilding that logic by hand every time it runs. Many real stacks use both: Method 1 to sketch the workflow, Method 2 to handle the one step that actually needs judgment.

Step-by-Step: Your First n8n Claude Code Workflow

1. Install the Claude Code CLI if you haven’t already, and log in with your Anthropic account.
2. Add n8n-mcp as an MCP server, using the one-line command from its GitHub README.
3. Optionally set N8N_API_URL and N8N_API_KEY as environment variables, so Claude Code can push workflows straight into your instance instead of leaving you to import them by hand.
4. Open a Claude Code session and describe the workflow you want in plain language, including the trigger, the tools involved, and what “done” looks like.
5. Review the node list Claude Code proposes before you import or deploy it. Check that it picked the right trigger and the right credentials, not just a plausible-looking one.
6. Run the workflow once with a manual trigger and check every node’s actual output, not just whether the run completed without an error.
7. Ask Claude Code to fix any node it flags as misconfigured, using n8n’s own validation, rather than debugging the JSON by hand.

A Realistic n8n Claude Code Workflow

Picture a workflow that watches a GitHub repository for new issues tagged “bug.” A trigger node fires when a matching issue lands. The workflow passes the issue text to a Claude Code step, built with Method 2, and asks it to check the relevant file and draft a fix. Claude Code returns a short diff and a plain-language explanation of what it changed and why. The workflow posts both as a comment on the issue, then pings a Slack channel for a human to review. Nothing merges automatically. Claude Code drafts. A person still approves.

You could design that same workflow’s shape faster with Method 1. Describe it to Claude Code first, inside a terminal, and let it wire the GitHub trigger, the Slack node, and the comment-posting step. Then swap in the Method 2 node for the actual code-reasoning part in the middle. The two methods aren’t rivals here. Method 1 gets you a working skeleton fast, with the boring plumbing already connected. Method 2 fills in the one step that needs real judgment, which is the part a plain n8n node could never do on its own.

What Each Method Can’t Do

Neither method is a fully hands-off system, and treating either one that way is how workflows break quietly. Method 1 is only as good as the instructions you give it. A vague request produces a vague workflow, one that runs without errors but does the wrong thing. It also can’t know your business rules unless you state them, like which customers count as VIP or which errors are safe to ignore.

Method 2 has its own limits. Every call spins up a real CLI process, so it’s slower and costs more per run than a simple node that just transforms data. It also isn’t sandboxed by default. A Claude Code session with broad file access can, in principle, read or write more than you intended. Scoping its permissions matters more here than in a normal n8n workflow. Neither method removes the need to test a workflow before you trust it with production data.

Treat the first week of any n8n Claude Code workflow as a monitoring period, not a launch. Watch the execution log for every run, not just the failed ones. A workflow that “succeeds” but picks the wrong Slack channel, or drafts a fix for the wrong file, still counts as broken. It just fails quietly instead of loudly, which is worse.

Requirements and Costs

Both methods need Claude Code itself, which requires either a Claude Pro, Max, or Team subscription, or a pay-as-you-go API key from the Claude Console. Method 1 adds no extra cost beyond that. n8n-mcp itself is free and open source, and running it locally alongside Claude Code doesn’t add a separate bill. Method 2 adds one more requirement on top. n8n-nodes-claudecode is an unverified community node, so it needs a self-hosted n8n instance, not n8n Cloud (n8n Blog, 2026). Our n8n self-hosted setup guide covers getting an instance running if you don’t have one yet. If you’re new to n8n’s node system generally, our guide to n8n nodes is worth reading first. Both methods assume you already know how nodes connect and pass data along a workflow.

Ongoing cost works differently for each method. A plain n8n node that just moves or filters data costs nothing beyond your n8n plan. A Method 2 step bills through your Claude subscription or API usage every time the workflow runs, the same as any other Claude call. Method 1 barely touches your costs at all, since you only run it while designing a workflow, not every time it executes afterward.

Common Pitfalls to Avoid

• Trying Method 2 on n8n Cloud. It won’t install, since unverified community nodes require a self-hosted instance, and there’s no workaround short of moving your instance.
• Skipping the n8n API credentials in Method 1’s management mode, then wondering why nothing deploys automatically. Without them, Claude Code only drafts a workflow; you still import it by hand every time.
• Letting a Method 2 session carry too much unscoped permission. Claude Code CLI sessions can read and write files on the host machine. Lock down what the workflow’s user account can actually touch before you connect it to anything with real data.
• Assuming both methods bill the same way. Method 1 runs through your normal Claude Code usage. Method 2 spawns a CLI process per workflow run, which adds up fast on high-volume triggers like a busy support inbox.
• Trusting a generated workflow without testing it first. Run every new build on a manual trigger, and check each node’s actual output, not just whether the canvas looks right or the run finished green.

If you’re still deciding whether Claude Code is your primary coding tool at all, our Cursor vs Claude Code comparison walks through how the two compare for day-to-day work, separate from either n8n integration here.

Frequently Asked Questions

Can Claude Code control n8n directly?

Yes, through n8n-mcp’s management mode. Add your n8n instance’s API URL and API key as environment variables. Claude Code can then create, edit, and deploy workflows straight into your account, not just draft them for manual import.

Does n8n-nodes-claudecode work on n8n Cloud?

No. It’s an unverified community node, and n8n Cloud only supports a curated, manually vetted set of community nodes. You need a self-hosted n8n instance to install it. That also means you’re responsible for keeping n8n and the Claude Code CLI updated yourself.

Do I need a Claude Pro subscription or an API key?

Either works. Claude Code authenticates with a Claude Pro, Max, or Team subscription, or with a standalone API key billed per token. Pick whichever matches how much you already use Claude day to day, since switching later is just a re-login.

How is this different from the Anthropic Chat Model node in n8n?

The Anthropic Chat Model node connects Claude, the general chat model, to n8n’s AI Agent and AI Chain nodes. That’s a different product from Claude Code, which is Anthropic’s coding agent built for working inside a codebase or, in this case, an n8n instance. Our Claude and n8n integration guide covers the Chat Model route in full.

Is n8n-mcp official Anthropic or n8n software?

No. It’s an independent open-source project maintained by a developer, not a product built or supported by Anthropic or n8n directly. That’s worth knowing before you rely on it for anything business-critical, since support comes from the open-source community, not a vendor.

Can I use both methods in the same automation stack?

Yes, and it’s a common pattern. Use Method 1 to design and deploy a workflow’s skeleton fast. Then use Method 2 inside that same workflow for any step that needs Claude Code’s actual coding judgment at runtime. That beats rebuilding the same judgment as a rigid set of n8n conditions.

Sources: n8n-mcp, GitHub (czlonkowski/n8n-mcp), 2026. n8n-nodes-claudecode, GitHub (johnlindquist/n8n-nodes-claudecode), 2026. n8n Blog, “Community nodes available on n8n Cloud,” 2026.