Summary
Nate Herk demonstrates an automated workflow repair system that uses Claude Code and n8n’s error workflow feature to detect and fix broken automations without human intervention. When a production n8n workflow fails, it triggers a secondary error workflow that fires an HTTP request to a locally running Claude Code instance (exposed via a tunnel since n8n Cloud can’t reach a local process directly). Claude Code, equipped with the n8n MCP server and custom skill files, analyzes the workflow’s JSON structure, identifies the root cause, applies a fix, and sends a ClickUp notification confirming the repair — all autonomously.
Two live examples are shown. In the first, a validate-each-order node failed because an upstream enrichment node returned all orders as a single array. Rather than following n8n’s own suggestion of inserting a split-out node, Claude Code modified the code node itself to return individual items — a cleaner, more efficient solution. In the second, a structured output parser in an email agent contained invalid JSON in its schema, which Claude Code also corrected automatically.
Herk is candid about current limitations: the system handles logic errors, null values, wrong data types, expression typos, and code node bugs well, but cannot fix expired API credentials or external service outages — though he expects rate-limit recovery to become feasible as models improve. The architecture relies on Claude Code running locally with the n8n MCP server providing read/write workflow access, making this a practical starting point for anyone building resilient n8n automations with Anthropic’s Claude.
📺 Source: Nate Herk · Published January 22, 2026
🏷️ Format: Hands On Build
