Resources · n8n integration

n8n Anthropic Chat Model nodeConfigure Anthropic Chat Model in n8n.

The n8n Anthropic Chat Model node is the sub-node that hands Claude models to a root node. It has no main input and never runs on its own: it plugs into the Model port of an AI Agent or a chain. Version 2 exposes 10 parameters, from token limits to thinking modes.

Try the Anthropic Chat Model node on n8nSee the 10 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Anthropic Chat Model node actually do?

A chat model node hands a large language model to another node. This one does that for Anthropic's Claude family of chat models: it holds the model choice and the sampling settings, then feeds them to whatever root node it is wired to. A sub-node is a node that attaches to a port of a bigger node instead of sitting in the main flow, so this one produces nothing by itself. That surprises people who drop it on a blank canvas and press Test.

The first shape is the one most builders land on. An AI Agent reads an incoming message, decides what to do, and needs a brain to do it with. You wire this node to its Model port, pick a model in Model, and cap the answer with Maximum Number of Tokens so a long reply never blows past what you want to paste back into a Gmail draft. A token is a chunk of text, roughly a short word, and the cap counts the generated ones only.

The second shape is quieter and just as common: no agent at all, a chain that classifies or extracts. Feed it rows read from Google Sheets, set Sampling Temperature low so two runs on the same row give the same label, and the model behaves like a rule you did not have to write. Expressions such as {{ $json.comment }} carry the field from the previous node into the prompt of the root node, not into this one.

The third shape is reasoning that takes its time. Turn Enable Thinking on, leave Thinking Mode on adaptive, and raise Effort when a step keeps going wrong on a hard document. Cost follows: the provider of the model bills its own API calls under its own terms, and n8n adds nothing on top.

When would you pick something else? Any other chat model sub-node fits the same port, so switching to OpenAI is a matter of unplugging one node and plugging another, with the same root node above it. Stay with this one when you want the Claude family and the thinking parameters it exposes.

The limits are worth knowing before you build. This node has no main input, so it cannot be the first node of a workflow, cannot call a tool by itself, and cannot be tested in isolation: you run the root node and read the output there. One port takes one sub-node type, so a root node gets one chat model, not two. Our n8n review covers where the platform as a whole fits.

Connect

What do you need to authenticate?

  1. 01

    Create the Anthropic credential in n8n

    Open the Credentials menu in n8n and create the credential this node asks for. A credential is the stored login n8n reuses instead of retyping a secret in every node, and this one is created once and reused in every workflow that talks to Claude. The key itself comes from an Anthropic account, on Anthropic's side.

  2. 02

    Or run on Gateway credits instead

    On n8n Cloud there is a second route. Select Use Gateway credits in the node's credential field and the node runs without an Anthropic account and without your own Anthropic API key. Handy when you want a workflow running today and the account paperwork is still in progress. Self-hosted instances go the API key route.

  3. 03

    Select the credential, then the model

    Back in the node, pick the credential you just saved, then set Model. The field takes a value from the list or a model ID typed by hand, which is what you use when a model is newer than the list in front of you. Nothing else in the node is required for a first run.

Connections

What does this node plug into?

n8n splits AI work in two: a root node sits in the main flow and receives the items, and sub-nodes attach to its ports, one type of sub-node per port. This node is a sub-node, so the only question is which port it goes on.

Output (what it plugs into)

  • Modelai_languageModel
01ModelThis is the output side of the node: it connects upward, to the Model port of a root node that needs a language model to run.

This is the output side of the node: it connects upward, to the Model port of a root node that needs a language model to run.

Key parameters

  • Required: the root nodes that expose this port do not run without a model on it.
  • AI Agent and Basic LLM Chain: the two starting points, an agent that decides between tools, or a single call with one prompt.
  • Information Extractor and Text Classifier: the same port on nodes that return structured output instead of free text.
Use cases
one credential and one model choice can serve several root nodes in a workflow, each with its own copy of this node.
Parameters

What can you set in the node?

The Anthropic Chat Model node has 10 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Model

model

What you see in n8n

Notes & use cases

The only required field of the node. It decides which Claude model answers, and it is a resource locator, meaning you either pick from the list n8n loads or paste an identifier yourself.

Key parameters

  • Model: required, chosen from the list or given as an ID when the model you want is newer than the list.
Use cases
a support agent and a nightly summarizer rarely need the same model, so keep one copy of this node per root node instead of sharing one across the workflow.
02

Maximum Number of Tokens

options.maxTokensToSample

What you see in n8n

Notes & use cases

Sets a ceiling on the length of what the model generates. The count covers the completion, not the prompt you sent, so a long conversation history does not eat into it.

Key parameters

  • Maximum Number of Tokens: optional number, the maximum tokens to generate in the completion.
Use cases
an answer that lands in a Slack message wants a low ceiling; a full report wants a high one. Raise it when replies stop mid-sentence, which is the usual symptom of a cap set too tight.
03

Sampling Temperature

options.temperature

What you see in n8n

Notes & use cases

Controls how random the wording gets. Lower values give less random completions, and as the temperature approaches zero the model becomes deterministic and repetitive.

Key parameters

  • Sampling Temperature: optional number, the randomness of the sampling process.
Use cases
classification, extraction and anything you compare between two runs belongs near zero. Draft copy where repetition reads badly belongs higher. Move this one or Top P, and watch the output change before touching anything else.
04

Top K

options.topK

What you see in n8n

Notes & use cases

Trims the tail of the distribution. It removes low probability responses from the choices the model considers for the next token, and it defaults to -1, which disables it.

Key parameters

  • Top K: optional number, off at -1, otherwise the number of token choices kept.
Use cases
reach for it when output wanders into odd vocabulary that temperature alone did not fix. Most workflows never touch it, and leaving it disabled is a perfectly normal answer.
05

Top P

options.topP

What you see in n8n

Notes & use cases

Another way to narrow the same choice, by probability mass rather than by count. At 0.5, half of all likelihood-weighted options are considered, and lower values ignore the less probable ones.

Key parameters

  • Top P: optional number, the probability mass the completion samples from.
Use cases
the documentation is explicit that you generally alter this or the temperature, not both. Pick one knob per node and note which one you moved, or the next person debugging the workflow will move the other.
06

Enable Thinking

options.thinking

What you see in n8n

Notes & use cases

A switch that turns thinking mode on for the model. Off, the model answers directly; on, it works through the problem before producing the answer the root node receives.

Key parameters

  • Enable Thinking: optional boolean, thinking mode on or off.
Use cases
multi-step decisions inside an agent, or documents where the first answer keeps missing a condition buried in the text. Leave it off for labeling and short rewrites, where it buys nothing.
07

Thinking Budget (Tokens)

options.thinkingBudget

What you see in n8n

Notes & use cases

Caps the thinking itself. It sets the maximum number of tokens the model may spend before it starts writing, separate from the cap on the answer.

Key parameters

  • Thinking Budget (Tokens): optional number, the maximum tokens used for thinking.
Use cases
useful on a batch that runs on every row of a table, where a generous budget multiplied by hundreds of rows is a bill you feel at the provider. Set it deliberately rather than leaving it to chance.
08

Thinking Mode

options.thinkingMode

What you see in n8n

Notes & use cases

Says how extended thinking is configured, and it is the parameter that decides whether the budget above matters at all.

Key parameters

  • Disabled (disabled): no extended thinking.
  • Adaptive (Recommended) (adaptive): Claude decides how much to think, and you steer it with Effort.
  • Manual (Deprecated) (manual): the legacy fixed-budget mode, rejected by Opus 4.7 and later.
Use cases
new workflows go adaptive. If an older workflow suddenly fails after a model change, this deprecated mode is the first thing to look at.
09

Effort

options.effort

What you see in n8n

Notes & use cases

The dial that goes with adaptive thinking. It tells the model how hard to work on the request rather than fixing a number of tokens.

Key parameters

  • Low (low) and Medium (medium): everyday steps inside an agent.
  • High (high), X-High (xhigh) and Max (max): the settings you escalate to.
Use cases
start low, move up one step at a time on the cases that actually fail, and keep the rest of the workflow cheap. Jumping straight to the top hides which step was the problem.
10

Stream Responses

options.streaming

What you see in n8n

Notes & use cases

Changes how the answer travels, not what it contains. On, the model streams over Server-Sent Events, a connection the server keeps open to push text as it is written; off, one payload arrives at the end. The final output shape is unchanged either way.

Key parameters

  • Stream Responses: optional boolean, streamed or single payload.
Use cases
worth turning on for a chat interface where a person is watching words appear, pointless for a workflow that writes a row and moves on.
Need help

Need help automating Anthropic Chat Model with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the n8n Anthropic Chat Model node free to use?
Yes on the n8n side. The node ships in n8n's AI package, so there is nothing to install and nothing extra to pay, on n8n Cloud as well as on a self-hosted instance running the Community Edition under the Sustainable Use licence. What is not free is the model itself: the provider bills its own API calls under its own terms, and n8n adds nothing on top of that. On n8n Cloud, Gateway credits are the other way to run the node, without your own Anthropic account. Either way the workflow you build is identical.
02What do you need to authenticate before the node runs?
One credential, and that is the whole list. The node needs an Anthropic credential created once in n8n's Credentials menu from an Anthropic account, and then reused by every workflow that calls Claude. On n8n Cloud there is an alternative: select Use Gateway credits in the node's credential field and the node runs without an Anthropic account and without an API key of your own. After that, only Model is required. One warning: since this node never executes alone, a bad credential shows up when you run the root node above it, not when you press Test here.
03What are the limits of this node?
It is a sub-node, which sets most of them. There is no main input, so it cannot start a workflow and cannot be run or tested by itself. It calls no tools and keeps no memory: those are other sub-nodes on other ports of the same root node, and one port accepts one sub-node type, so a root node gets a single chat model. The page describes version 2 of the node, the highest in the catalogue, and an older workflow may show an earlier version with fewer options. Manual thinking mode is deprecated and rejected by Opus 4.7 and later.
04What is the minimum to plug in, and where do you start?
Two nodes and a wire. Put a root node in the main flow, connect this one to its Model port, and you have a working setup. Start with Basic LLM Chain if you want one prompt and one answer, because there is less to go wrong and the output is easy to read. Move to AI Agent when the model has to choose between several actions. In the node itself, set Model, leave everything else alone for the first run, then add Maximum Number of Tokens and Sampling Temperature once you see what the output looks like.
05n8n or Make for this kind of AI workflow?
It depends on where you want the workflow to live. Make is hosted by Make, with no self-hosting option, and it bills per operation, which is predictable and easy to start with. n8n runs on Cloud or on your own server, so the data and the model calls stay where you put them, and an AI step that loops does not turn into a per-operation counter. The visual logic differs too: n8n's ports and sub-nodes are more to learn at first, and they are what lets you swap one chat model for another without rebuilding the workflow around it.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations