Claude n8n Integration: 5 Free Ways to Automate in 2026

Claude n8n integration connects Anthropic’s Claude models to n8n’s automation platform. Claude reads data, makes a decision, and triggers the next step in your workflow. You can wire it in three ways: n8n’s AI Agent node, a direct API call, or the Model Context Protocol (MCP). This guide covers five practical setups and what each one costs. I’ll also show the exact Claude n8n workflow already running this blog’s email triage.

Key Takeaways

The fastest path is n8n’s native Anthropic Chat Model node, which plugs Claude straight into the AI Agent and AI Chain nodes (n8n, 2026).

Claude Sonnet 5 and Opus 5 now run a 1M-token context window, five times the older 200K limit, so n8n workflows can pass entire documents in one call (Anthropic, retrieved 2026-07-30).

Claude API pricing runs $1 to $5 per million input tokens depending on model tier, plus output tokens at roughly 5x that rate (Anthropic, retrieved 2026-07-30).

For heavier use, the open-source n8n-mcp server lets Claude Code build and edit n8n workflows directly from a terminal.

I send one automation breakdown like this every week. The sign-up is a few sections down.

What Claude n8n Integration Actually Means

n8n doesn’t ship a single dedicated “Claude node.” Instead, Claude plugs in as an Anthropic Chat Model, a sub-node that powers n8n’s LangChain-based AI Agent and AI Chain nodes (n8n, 2026). This sub-node handles the conversation and lets Claude call other n8n nodes as tools mid-workflow. That’s different from a plain API connector, and it’s why the AI Agent node is the default choice for most Claude n8n builds.

Two more paths exist. You can skip the AI Agent framework and hit the Claude API directly with an HTTP Request node. This is useful when you need exact control over the prompt. Or you can go the other direction with MCP. Here, Claude Code or Claude Desktop controls n8n itself, building workflows through plain language instead of a canvas.

This guide covers the AI Agent approach first, since it covers most solopreneur use cases without writing code. If you’re already comfortable with APIs, Method 2 skips the LangChain layer entirely and calls Claude directly.

Before You Start: What You Need

An n8n instance, cloud or self-hosted. What n8n is and how it works covers the basics if you’re new to the platform.
An Anthropic API key from the Claude Console. This bills separately from a claude.ai subscription.
For Method 3 (MCP), Claude Code or Claude Desktop installed locally, plus Node.js to run the n8n-mcp server.
For Method 4, a server where both n8n and the Claude Code CLI can run side by side.

Method 1: Wire Claude into n8n’s AI Agent Node

Add the AI Agent node to your canvas. Attach the Anthropic Chat Model as its language model input, then set your Anthropic API key as an n8n credential. Connect any tool node, such as Gmail, Google Sheets, or an HTTP Request, to the agent’s tool input. Claude decides which tool to call and when, based on the system prompt you write.

This is the pattern behind n8n’s AI Agents and workflows guide, and it’s where most solopreneurs should start. It handles tool-calling, memory, and multi-step reasoning without you writing orchestration logic by hand.

Claude fits this role well because of its context window. Claude Sonnet 5 and Opus 5 run a 1M-token context window on the Claude API, up from 200K on earlier Sonnet models (Anthropic, 2026). In practice, an agent can read a full customer thread, a long PDF, or a chunk of your codebase in one call, instead of chunking it first.

Claude n8n workflow canvas linking an AI agent node to email, calendar, chat, and settings tools

AI Agent vs. AI Chain: Which Node Fits Your Workflow

n8n offers two LangChain-based patterns for using the Anthropic Chat Model: the AI Agent node and the Basic LLM Chain node. They look similar, but they solve different problems.

The AI Agent node is dynamic. Claude decides at runtime which tools to call and in what order, based on the task and the tools you’ve connected. Use it when the right sequence genuinely depends on the input. A support-ticket workflow might need to look up an order, check a policy doc, or escalate to a human.

The Basic LLM Chain node is static. You define the exact sequence yourself, and Claude generates text at one or more fixed points. Use it for simpler jobs, like turning a blog draft into three social captions, where the steps never change regardless of input.

As a rule, pick the AI Agent node when the workflow needs to branch based on what Claude decides. Pick the Chain node when the steps are fixed and Claude is really just generating text, not making decisions. Most Claude n8n workflows that feel slow to build started as an AI Agent when a Chain node would have been simpler and cheaper to run.

Method 2: Call the Claude API Directly with HTTP Request

Sometimes the AI Agent node is more structure than you need. Sometimes Claude just needs to summarize a webhook payload or classify an incoming email. A plain HTTP Request node calling Anthropic’s Messages API is simpler to run and easier to debug.

Set the node to POST to https://api.anthropic.com/v1/messages, add your API key as a header, and pass your prompt in the request body. This route gives you full control over the model, the system prompt, and the token limit. That control matters if you’re processing high volumes and watching cost per run closely.

Method 3: Give Claude Code Direct Access to n8n via MCP

n8n-mcp is an open-source Model Context Protocol server built for exactly this. It gives Claude Code, Claude Desktop, Cursor, and Windsurf structured knowledge of n8n’s full node library, including each node’s real properties and operations (n8n-mcp, GitHub). Instead of hand-writing JSON for a new workflow, you describe what you want in plain language. Claude Code picks the nodes, wires the connections, and hands you a workflow ready to import.

Install it by adding the MCP server to Claude Code’s configuration, pointing it at your n8n instance’s API URL and key. Once connected, you can ask Claude Code something like “build a workflow that pulls new Stripe payments and adds them to a Google Sheet.” It drafts the actual n8n JSON for you.

This is the most powerful option for anyone already comfortable with Claude Code, and it’s where the “n8n claude code” ecosystem is heading. Always review and test what it builds. Treat the output as a strong first draft, not a finished production workflow.

Method 4: Automate Claude Code Itself from Inside n8n

Method 4 runs the other direction. n8n triggers Claude Code, not the reverse. Use n8n’s Execute Command node to run the Claude Code CLI as a scheduled or event-triggered job. This suits tasks like nightly codebase audits, automated PR reviews, or content scripts that need a coding agent’s full tool access, not just a chat reply.

This setup needs a server where n8n and the Claude Code CLI both run, since Execute Command shells out on the same machine. It’s the most advanced of the five methods here, and it’s overkill for most solopreneur use cases. Reach for it only when a task genuinely needs Claude Code’s file-editing and command-running abilities.

Method 5: Community Nodes and Webhooks for Custom Integrations

For everything else, community nodes and generic webhook triggers fill the gaps. n8n’s large integration catalog lets you route a Claude response to Slack, Discord, a custom dashboard, or almost any other tool, without writing code.

A common pattern: a webhook receives an event, an AI Agent or HTTP Request node calls Claude, and a final node routes the response somewhere useful. That shape sits behind most of the practical workflows in this guide.

A concrete version: a form submission triggers a webhook. Claude drafts a personalized reply based on the form fields, and a Slack node posts the draft for approval before it sends. Nothing here needs a dedicated Claude node. It’s the same webhook-to-AI-to-action shape, just pointed at different tools.

What This Actually Costs

Running Claude inside n8n costs two things: your n8n hosting, and Claude API token usage. n8n’s Community Edition is free to self-host, with unlimited executions (full n8n pricing breakdown).

Claude API pricing is metered per token and varies by model. Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens. Claude Sonnet 5 runs an introductory $2 input / $10 output per million tokens through August 31, 2026, rising to a standard $3/$15 afterward. Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens (Anthropic, 2026).

Model	Input (per million tokens)	Output (per million tokens)	Best for in n8n
Claude Haiku 4.5	$1	$5	Fast classification, routing, short summaries
Claude Sonnet 5	$2 (through Aug 31, 2026)	$10 (through Aug 31, 2026)	Most AI Agent workflows, tool use
Claude Opus 5	$5	$25	Complex reasoning, long-document analysis

For most n8n automations, that’s a small line item. A short classification or summarization step on Haiku 4.5 often costs a fraction of a cent per run. Reserve Opus 5 for workflows doing genuinely hard reasoning, and default to Sonnet 5 or Haiku 4.5 for routine automation steps.

The Workflow I Actually Run

I run a live Claude n8n workflow on this blog right now. It triages incoming email before I open my inbox. Claude reads each message with the AI Agent node, decides its category and priority, and routes it accordingly. I wrote up the full build, node by node, in a separate guide.

The trigger is a scheduled check against my inbox. When a new message lands, the AI Agent node reads the subject and body. It classifies the message against a short list of categories from my system prompt: client work, newsletter reply, spam, or something that needs my direct attention. Claude also sets a priority, and a downstream node moves the message into the right folder or pings me for anything urgent.

Building it this way, instead of a rigid keyword filter, means it handles messages that don’t fit a fixed pattern. A vaguely worded subject line from a real client still gets classified correctly, because Claude is reading intent, not just matching text.

The lesson from running it in production: start with Method 1, the AI Agent node. It covers most real automation needs, and you can add Method 2’s direct API calls later for the few steps that need tighter control. I only reached for a raw HTTP Request node once, on a step where I needed to cap output tokens tightly to control cost.

Common Mistakes When Connecting Claude to n8n

Skipping the system prompt. The AI Agent node works best with an explicit, detailed system prompt. A vague one leads to inconsistent tool choices.
Using Opus 5 everywhere. It’s the most capable model, but also the most expensive. Most routing and classification steps run fine on Haiku 4.5.
Not setting a token limit. Without a max_tokens cap, a bad loop or a long document can run up cost fast.
Trusting an MCP-built workflow untested. n8n-mcp drafts a working structure, but always review the node configuration before you trust it with production data.
Hardcoding the API key in a node. Store it as an n8n credential instead, so it isn’t exposed in workflow exports.
Ignoring rate limits at scale. A workflow that fires on every webhook event can hit Anthropic’s rate limits fast. Add a queue or a delay node if you’re processing high volume.
Reaching for the AI Agent node when a Chain node would do. If Claude always does the same steps in the same order, use a Chain node instead. It runs cheaper and is easier to debug than an Agent node re-deciding the same thing every time.

Frequently Asked Questions

Can I use Claude for free inside n8n?

n8n’s Community Edition is free to self-host. Claude itself isn’t free through the API. You pay per token, though cost is often under a cent for a short workflow run. New Anthropic accounts do get a small free credit to test the API.

Which Claude model should I use in n8n?

Start with Claude Haiku 4.5 for fast, cheap tasks like classification and routing. Move to Claude Sonnet 5 for most AI Agent workflows that need tool use or careful reasoning. Reserve Claude Opus 5 for the hardest reasoning tasks, since it costs the most per token.

Does n8n have a dedicated Claude node?

Not a single dedicated node. Claude connects through the Anthropic Chat Model sub-node, which plugs into n8n’s AI Agent and AI Chain nodes. You can also call the Claude API directly with a generic HTTP Request node.

What is n8n-mcp, and do I need it?

n8n-mcp is an open-source Model Context Protocol server that lets Claude Code, Claude Desktop, and similar tools build and edit n8n workflows through natural language. You don’t need it to use Claude inside n8n. It’s useful once you’re building workflows often and want Claude to draft the JSON for you.

Can Claude Code run inside an n8n workflow?

Yes, through n8n’s Execute Command node, which can shell out to the Claude Code CLI. This needs a server where both tools run, and it suits advanced tasks like automated code review, not routine chat automation.

Is Claude better than other AI models for n8n automation?

Both Claude and other providers work well inside n8n’s AI Agent node. Claude’s large context window suits workflows that process long documents in one pass. The better fit depends on your specific task and budget, not a blanket rule.

Conclusion

Most Claude n8n builds should start with the AI Agent node and the Anthropic Chat Model sub-node. Reach for a direct HTTP Request call, MCP, or the Execute Command node only when the AI Agent node’s structure doesn’t fit your task.

Start with the AI Agent node. It covers most real automation needs with the least setup.
Match the model to the task: Haiku 4.5 for routine steps, Sonnet 5 for most agent work, Opus 5 for hard reasoning.
Add MCP or Execute Command once you’re comfortable with Claude Code and want it building or running workflows directly.

Want the packaged, zero-setup version of Claude for business tasks instead of a custom build? Claude for Small Business covers what Anthropic ships out of the box, and how it compares to a custom n8n pipeline like this one.

Sources

n8n, “Anthropic integrations,” retrieved 2026-07-30, https://n8n.io/integrations/anthropic/
Anthropic, “Context windows,” Claude Platform Docs, retrieved 2026-07-30, https://platform.claude.com/docs/en/build-with-claude/context-windows
Anthropic, “Pricing,” Claude Platform Docs, retrieved 2026-07-30, https://platform.claude.com/docs/en/about-claude/pricing
n8n-mcp (Piotr Czlonkowski), GitHub repository, retrieved 2026-07-30, https://github.com/czlonkowski/n8n-mcp

Claude + n8n: Connecting Anthropic’s AI to Your Automation Stack