Auto-Enrich HubSpot and Google Sheets Leads with Bright Data MCP in n8n

Trigger on new HubSpot contacts or Google Sheets row changes → AI Agent calls Bright Data MCP tools to search and fill → structured output with confidence scoring → write back to sheet or update CRM

Who

Sales ops, marketing ops, or RevOps teams already using n8n with access to HubSpot, Bright Data, and OpenAI accounts

You get

Automatically backfill LinkedIn, company info, social handles, and other fields for new leads; route results by confidence to overwrite, append for review, or update HubSpot contacts

You need

n8n · google-sheets · hubspot

Steps

1. Configure HubSpot authentication

   In HubSpot, create a Private App (simple) or an OAuth App (advanced), then copy the Access Token or Client ID / Client Secret / Developer API key. Create the matching HubSpot credential in n8n and attach it to the HubSpot nodes. For Private App, paste your n8n Webhook node URL into HubSpot's webhook settings so HubSpot can push new-contact events.

2. Collect Bright Data credentials

   Log in to Bright Data and note your API token; optionally collect the Web Unlocker zone name and Browser API auth string (username:password).

3. Host the Bright Data MCP SSE server

   Choose one deployment method so n8n can reach Bright Data MCP over HTTP SSE: Method 1 (recommended): Run Supergateway on a separate cloud server (DigitalOcean, Heroku, Render, etc.). For NPM, set env vars API_TOKEN, WEB_UNLOCKER_ZONE, BROWSER_AUTH and use the start command: npx -y supergateway --stdio "npx -y @brightdata/mcp" --port 8000 --baseUrl http://localhost:8000 --ssePath /sse --messagePath /message After deploy, access https://your_server_url/sse. Method 2: Co-locate Supergateway on the same server as self-hosted n8n. Set the same env vars, run the command above or set it as a pre-deploy command, then use http://localhost:8000/sse. Method 3 (easiest): Use Smithery.ai. Visit https://smithery.ai/server/@luminati-io/brightdata-mcp/tools to sign up, create an API key, configure env vars in a profile, and retrieve your SSE HTTP URL.

   Setup guidance · Source text。Supergateway NPM start command; use in standalone server or self-hosted n8n startup script

   Copy prompt

   npx -y supergateway --stdio "npx -y @brightdata/mcp" --port 8000 --baseUrl http://localhost:8000 --ssePath /sse --messagePath /message

   Setup guidance · Source text。Supergateway Docker start command

   Copy prompt

   docker run -it --rm -p 8000:8000 supercorp/supergateway \
       --stdio "npx -y @brightdata/mcp /" \
       --port 8000

4. Prepare and connect Google Sheets

   Ensure your sheet includes columns: row_id, first_name, last_name, email, status (plus any fields you want filled). Share the sheet with your n8n service account or connect via OAuth. In n8n, add a Google Sheets Trigger node and set it to watch that sheet for new rows or status changes.

   Setup guidance · Source text。Sticky Note next to Google Sheets Trigger node; reminder to write 'needs more enrichment' in the status column to trigger a retry

   Copy prompt

   ## Watch for the updated record in Google Sheets based on the column 'status'
   Write 'needs more enrichment' in the status column if you want to retry the enrichment

5. Import the workflow and configure node credentials

   Import the template workflow JSON into n8n. Configure credentials for: - HubSpot: choose App Token or OAuth credential - Google Sheets Trigger / Google Sheets: link to your actual sheet - OpenAI Chat Model / OpenAI (Judge): enter your API key and pick a model (Scraper AI Agent defaults to gpt-4.1-mini; Judge and Auto-fixing default to gpt-4o-mini) - Bright Data MCP Client: enter your SSE server URL (e.g., https://your_server_url/sse or http://localhost:8000/sse) Review the Structured Output Parser JSON Schema and adjust it to match your sheet columns if needed.

6. Run the AI Agent to enrich leads

   When a new HubSpot contact appears or a Google Sheets row has empty status or status = 'needs more enrichment', the workflow triggers. The Scraper AI Agent receives the lead data and follows its system instructions to search and fill.

   Execution prompt · Source text。Scraper AI Agent node text field (user message); auto-injects the current lead JSON at runtime

   Copy prompt

   =Here's the lead's data:
   
   {{ $json.toJsonString() }}

   Execution prompt · Source text。Scraper AI Agent node systemMessage field; defines search strategy, source requirements, and output format

   Copy prompt

   =Prompt: Person and Company Enrichment Task
   
   Objective:
   Enrich the lead’s information as completely and accurately as possible. Avoid fabricating data and maintain strict adherence to the correct tool schema for each tool used in the MCP client.
   
   ⸻
   
   1. Person Enrichment Instructions:
   - Start with a Google Search:
   Prioritize locating the individual’s LinkedIn profile URL using the available information.
   - Query Guidelines:
   - If the email address is available, use it in your search query.
   - If not, use the person’s first name and last name.
   - Always append keywords like “linkedin”, “facebook”, “instagram”, or other relevant social media platforms to refine the search.
   - LinkedIn-Based Enrichment:
   Once the LinkedIn URL is found, use it to gather available details from their profile.
   
   ⸻
   
   2. Company Enrichment Instructions:
   - Email Domain Extraction:
   If the person’s email is not from a generic provider (e.g., gmail.com, outlook.com, hotmail.com, icloud.com), extract the domain and use it in your search.
   - Query Composition:
   - Combine the extracted domain with any existing value in the “company” field of the lead’s details.
   - Do not rely on or reuse company data obtained from other tool outputs.
   - Focus on sourcing reliable company information from LinkedIn and Crunchbase.
   
   ⸻
   
   Output Format:
   
   Return the enriched data in valid JSON format using the original key structure. Exclude any fields that remain empty. Example output:
   
   {
     "first_name": "John",
     "last_name": "Doe",
     "address": "123 Maple Street, Apt 4",
     "instagram_handle": "johndoe38"
   }

7. Confidence scoring and write-back

   The AI Agent output is validated by the Structured Output Parser; if it fails, the Auto-fixing Output Parser attempts repair. Then the Judge node uses gpt-4o-mini to compare original and enriched data, outputting a confidence score (0–1) and remarks. Routing logic: - confidence ≥ 0.85: overwrite the original Google Sheets row; within this branch, if confidence > 0.89, the No Operation → Set Desired Confidence Score path additionally updates the HubSpot contact (including custom properties instagram_handle, facebook_handle, and linkedin_url) - confidence < 0.85: append a new row (row_id appended with timestamp) and set original row status to 'human review needed' - if Scraper output does not match original first_name/last_name/email at all (If/If1/If2 false branch): via Not matched route, update status only to 'no information found' After each operation, a Wait node (named 'Space requests by 5 seconds' with empty parameters in workflow JSON; actual delay depends on n8n Wait node default behavior or name implication) spaces out processing before the next item.

   Execution prompt · Source text。Judge the accuracy of scraped data node system message; defines scoring criteria and output format

   Copy prompt

   Task:
   Evaluate the accuracy of the scraped data compared to the original source data.
   
   Instructions:
   - Assess how closely the scraped output matches the original data in terms of content, structure, and completeness.
   - Provide a confidence score based on the overall accuracy.
   - Explain your reasoning in concise, clear remarks.
   
   Output Format (JSON):
   
   {
     "confidence": [decimal between 0 and 1, e.g., 0.98 for 98% accuracy],
     "remarks": "A brief explanation justifying the confidence score, highlighting any discrepancies or issues found."
   }

   Execution prompt · Source text。Auto-fixing Output Parser node prompt field; fallback instruction when format validation fails

   Copy prompt

   Instructions:
   --------------
   {instructions}
   --------------
   Completion:
   --------------
   {completion}
   --------------
   
   Above, the Completion did not satisfy the constraints given in the Instructions.
   Error:
   --------------
   {error}
   --------------
   
   Please only return the first_name, last_name, email, and scrape_error (put the error details in this field)

You are an n8n workflow setup assistant. The user wants to replicate the "HubSpot + Bright Data MCP + Google Sheets" AI lead enrichment template. First confirm prerequisites: n8n instance (cloud or self-hosted), HubSpot account with Private App or OAuth access, Bright Data MCP API token, OpenAI API key, and a Google Sheet with columns row_id, first_name, last_name, email, status. If the user chooses self-hosted Bright Data MCP SSE, confirm whether they have an extra server or prefer Smithery.ai. Then guide them to import the workflow JSON, configure credentials, and test with a single record. Do not generate or send any passwords or tokens on behalf of the user.