Generate a Notion AI Query Assistant from Any Database Schema Using n8n

Input a Notion database URL → auto-fetch schema → AI generates a tailored workflow JSON → self-validates → returns pasteable n8n workflow

Who

Builders with an n8n instance and Notion API credentials who want to quickly spin up an AI chat assistant for a team Notion database

You get

A complete n8n workflow JSON that can be pasted into the canvas to run an AI chat assistant querying that specific Notion database

You need

n8n · notion-api

Steps

1. Start chat and enter the Notion database URL

   After importing this template into n8n, activate the workflow. Open the public Chat Trigger interface and enter the URL of the target Notion database. The Chat Trigger receives the input and starts the workflow.

2. Fetch and simplify the database schema

   The Notion node extracts the URL and pulls the database schema (id, name, properties). If the URL is invalid or n8n lacks access, the node routes to the error branch. On success, a Set node standardizes fields and a Code node simplifies the properties object (keeping only id/name/type, with multi_select/select options preserved) to reduce token usage for the downstream AI Agent.

   Execution prompt · Source text。n8n Code node 'Simplify properties object' JS Code field

   Copy prompt

   // Loop through each incoming item
   return items.map(item => {
     const inputDatabase = item.json["inputDatabase"];
   
     const simplifiedProperties = Object.fromEntries(Object.entries(inputDatabase.properties).map(([key, value]) => {
       const simplifiedValue = {
         id: value.id,
         name: value.name,
         type: value.type
       };
   
       // Simplify based on type
       if (value.type === 'multi_select' || value.type === 'select') {
         simplifiedValue.options = value.multi_select?.options?.map(option => option.name) || [];
       }
       
       return [key, simplifiedValue];
     }));
   
     // Overwrite the properties object with simplifiedProperties
     item.json.inputDatabase.properties = simplifiedProperties;
   
     return item;  // Return the modified item
   });

3. AI generates the tailored workflow JSON

   The Set input data node feeds the simplified schema (inputDatabase) and the embedded reference workflowTemplate (based on template 2413) into the Generate Workflow Agent. This AI Agent (ReAct mode, default Anthropic Chat Model) is tasked with modifying the reference template to fit the new database schema, outputting valid n8n workflow JSON. The prompt explicitly requires proper quoting, no trailing commas, correct nesting, and preserved linebreaks for copyability. Structured Output Parser + Auto-fixing Output Parser constrain the output format.

   Execution prompt · Source text。AI Agent node 'Generate Workflow Agent' Text/Prompt field

   Copy prompt

   =Your task is to output a modified version of a n8n workflow template so it works with the provided new notion database schema. 
   
   
   # new notion database details
   {{ $json.inputDatabase.toJsonString() }}
   
   # n8n workflow template to use as reference
   {{ $json.workflowTemplate.toJsonString() }}
   
   JSON Output:
   - Ensure valid JSON with properly quoted keys and values, no trailing commas, and correctly nested braces `{}` and brackets `[]`. If unable to format, return an error or a valid example.
   - Output linebreaks so user can copy working JSON

   Execution prompt · Source text。AI Agent node 'Generate Workflow Agent' Options > Prefix field

   Copy prompt

   You are an n8n expert and understand n8n's workflow JSON Structure. You take n8n workflows and make changes to them based on the user request. 
   
   Don't hallucinate. Only output n8n workflow json.

4. Self-validation, error check, and retry

   After generation, Set schema for eval extracts the result and key fields (e.g., searchNotionDBJsonBody). A Switch node checks for a known error pattern: if jsonBody contains the '[object Object]' placeholder (observed ~25% of the time per source notes), it routes to the retry branch. The Add feedback prompt node constructs a feedback message, sending the failed attempt and original task back to Set input data to trigger a retry. Concurrently, a Text Classifier (connected to another Anthropic Chat Model) independently validates whether generatedWorkflow is valid n8n workflow JSON.

   Execution prompt · Source text。Set node 'Add feedback prompt' feedbackPrompt assignment field

   Copy prompt

   =You attempted the below task and outputted incorrect JSON. Below is your incorrect attempt and original task prompt. Try again.
   
   # Incorrect task prompt

   Execution prompt · Source text。Text Classifier node 'Valid n8n workflow JSON?' System Prompt Template field

   Copy prompt

   You are an expert in n8n workflow automation tool. You know whether the json representation of an n8n workflow is valid. 
   
   Please classify the text provided by the user into one of the following categories: {categories}, and use the provided formatting instructions below. Don't explain, and only output the json.

5. Return result to chat

   If validation passes, the Return success to chat node wraps the generated workflow JSON in a copyable code block and returns it, prompting the user to paste into an n8n canvas (v1.52.0+). If the initial URL was invalid or inaccessible, Return error to chat returns an error message instead. Each chat interaction corresponds to one workflow run.

   Execution prompt · Source text。Set node 'Return success to chat' output assignment field

   Copy prompt

   =Created workflow:
   ```
   {{ $json.generatedWorkflow }}
   ```
   
   ☝️ Copy and paste JSON above into an n8n workflow canvas (on v 1.52.0+)

You are a setup assistant familiar with n8n and the Notion API. The user wants to generate a queryable AI assistant workflow for a specific Notion database. Help the user: 1) Confirm the notionApi credential is configured and n8n has access to the target database; 2) Import template 2415 and activate it; 3) Enter the database URL in the Chat Trigger; 4) Wait for generation to complete and copy the returned JSON; 5) Paste the JSON into a new n8n canvas and verify nodes/connections. If generation fails, check URL validity and credential permissions.