The build loop

Every n8n workflow built through this kit follows the same 8-step loop. Subagents enforce it; you should know what they're doing.

Why a fixed loop?

Hand-written n8n workflows from memory fail constantly — property names drift between node versions, validation rules are subtle, and nodeType formats differ by tool. The loop eliminates guesswork by pulling real schemas from the n8n-mcp server and validating at every step.

The 8 steps

1. Invoke the matching skill — before any tool call, load the skill that encodes the relevant patterns. Skipping this is the single biggest cause of broken workflows.
2. Discover nodes — search_nodes({query}) → get_node({nodeType, detail: "standard"}).
3. Pull real schemas — never hand-write configs from memory. get_node returns the canonical property list for the installed node version.
4. Validate each node — validate_node({nodeType, config, profile: "runtime"}). Fix errors one node at a time before assembling the workflow.
5. Create the workflow — n8n_create_workflow({name, nodes, connections}). Use the full prefix (n8n-nodes-base.slack) for the type field here.
6. Validate the whole workflow — n8n_validate_workflow({id}). Catches connection errors, branch mismatches, and cross-node issues.
7. Iterate — n8n_update_partial_workflow({id, intent, operations}). Always pass intent (why you're making this change) — it's not cosmetic.
8. Activate — n8n_update_partial_workflow({id, operations: [{type: "activateWorkflow"}]}).

The nodeType format gotcha

n8n-mcp has two tool categories that use different prefixes for the same node:

Tool	Prefix	Example
search_nodes, get_node, validate_node, validate_workflow	short	nodes-base.slack
n8n_create_workflow, n8n_update_partial_workflow	full	n8n-nodes-base.slack

search_nodes returns both in its results — use nodeType for search/validate, workflowNodeType for workflow tools.

LangChain nodes follow the same pattern: nodes-langchain.agent vs @n8n/n8n-nodes-langchain.agent.

Validation profiles

Four options, from lenient to strict:

• minimal — only checks required fields. Use when debugging connection structure.
• runtime — checks values and types. Default and recommended.
• ai-friendly — reduces false positives from AI-generated configs.
• strict — maximum safety, use before production hand-off.

Smart parameters

When connecting IF or Switch nodes, use semantic names instead of sourceIndex:

// IF node
{ type: "addConnection", source: "IF", target: "Handler", branch: "true" }

// Switch node
{ type: "addConnection", source: "Switch", target: "Case A", case: 0 }

When a template already exists

Before step 2, search the 2,700+ community templates:

search_templates({ query: "your use case" })

If a template fits, n8n_deploy_template drops it into your instance — skips steps 3-6 entirely.