n8n-code-tool

Create a node of type @n8n/n8n-nodes-langchain.toolCode, read input from the query variable (or _query in Python), and return a string. Connect it to the AI Agent via the ai_tool connection and give it a descriptive snake_case name.

The Code node (n8n-nodes-base.code) receives items via $input.all() and returns [{json:{...}}]. The Code Tool (@n8n/n8n-nodes-langchain.toolCode) is invoked by an AI Agent, receives a query string or object, and must return a string.

This error occurs when you return the workflow item format [{json:{...}}], which belongs to the regular Code node. The Code Tool follows the LangChain tool contract and must return a string, so wrap structured data in JSON.stringify().

No, $fromAI() is not available in the Code Tool sandbox and throws "No execution data available". Read AI-supplied input from the query variable, or use toolWorkflow or HTTP Request Tool if you need $fromAI bindings.

Set specifyInputSchema to true and define the schema via schemaType fromJson (an example JSON, requires v1.3+) or manual (a full JSON Schema). The LLM then passes a validated object as query that you can destructure directly.

Generic names like "Code Tool" and vague descriptions give the LLM no invocation signal. Rename the tool to a verb-y snake_case name, state trigger conditions explicitly in the description, and verify the ai_tool connection to the AI Agent exists.