Claude works as a model provider for any n8n node that expects a "chat model" input - most commonly the AI Agent node from the previous guide, but also simpler nodes like Basic LLM Chain for single-shot prompts without tool use.
Setting up the connection
Add an Anthropic Chat Model node (search "Anthropic" in the node panel) and connect it to the model input of an AI Agent or LLM Chain node. You'll need an Anthropic API key, created at the Claude Console (console.anthropic.com) - paste it into a new credential the same way you'd set up any API key credential, per the credentials guide earlier in this series. Pick your model version from the node's dropdown once the credential is connected.
Why teams pick Claude specifically here
Claude tends to get chosen for n8n agents built around careful multi-step reasoning, long context handling (useful when an agent needs to work with lengthy documents or conversation history), and instruction-following where you need the model to stick closely to a defined role or set of constraints - support triage, document analysis, and structured extraction are common fits.
Where MCP fits into this picture
MCP (Model Context Protocol) is a separate but related concept, worth distinguishing clearly: MCP is a standard way for an AI model to connect to external tools and data sources, originally built around Claude and Claude Code. n8n's AI Agent node achieves a similar outcome - giving the model access to tools - through its own native tool nodes and sub-workflow pattern, rather than through MCP directly. As of this series' writing, MCP client/server support in n8n exists via community nodes and is an active area of development; check the current node library for the latest first-party support before assuming a specific MCP integration path.
A simple pattern: Claude as a classifier, n8n as the router
A reliable, low-risk way to introduce Claude into an existing workflow: use a Basic LLM Chain node with Claude to classify or summarize incoming data (e.g., "is this support request urgent, normal, or spam?"), then feed that classification into a standard n8n Switch node to route the rest of the workflow deterministically. You get the model's judgment where you need it, without handing over full agentic control everywhere.
Cost awareness
Every Claude node call is a billed API request against your Anthropic account, separate from any n8n execution cost. For workflows processing meaningful volume, keep an eye on prompt length and whether you truly need tool-using agent behavior versus a simpler, cheaper single-shot chain call - the difference in cost and latency can be significant.