Claude AgentInstall

Summary

Runs a Claude agent turn for each incoming item, using the configured model source and user message. It can continue conversations through Redis-backed sessions, configure tools and workspace behavior, and stream agent output and metadata to the workflow.

Inputs

• Model Config Source (required) — Choose model configuration from an upstream claudeModelConfig field, credentials on this node, or inferenceModel/inferenceModelProvider input fields. The selector option lets you change the config field name; credential and input modes expose a model override, and input mode also accepts a provider-to-credential-field map.
• User Message (required) — The current user message sent to Claude.
• System Message — Optional system prompt for a new turn; ignored when resuming a session.
• Options — Configure Redis-backed session continuity and its runtime, workspace directories and skills, permission preset and agent limits, structured output, tool search, hooks, subagents, and MCP servers and tool filters. Session continuation requires a Redis credential; the sidecar runtime also uses the configured sidecar URL.

Output shape

A stream of n8n items containing Claude agent output and stream events.

The source exposes text output, structured output when configured, usage and session metadata, suggestions, and task snapshots through the stream assembler; exact emitted item fields depend on the event and configured output. A Redis session ID enables continuity across executions.

Links

• n8n community nodes installation guide