n8n Workflow: Screen Contact Activity from CSV (OpenAI GPT-4o)

Screen contact activity from a CSV—using page text and OpenAI in n8n

This n8n workflow screens your contacts by checking what’s visible on their website, Facebook, and Instagram pages (when provided), then uses OpenAI (gpt-4o-mini) to classify each contact as active, inactive, or uncertain—and outputs the results to a new CSV.

What this workflow does

• Manually starts and reads a contacts CSV from your local filesystem via n8n file operations.
• Parses and processes records in batches, preparing each contact for review.
• For each contact, builds browser automation steps to visit:
  • the contact’s website,
  • their Facebook URL (if present),
  • their Instagram URL (if present).
• Extracts visible page text using an MCP Client endpoint (default: http://localhost:8931/mcp) that supports browser tools like browser_navigate, browser_wait_for, and browser_evaluate.
• Cleans and aggregates evidence per channel, then sends an evidence-based screening prompt to OpenAI to return a JSON verdict (active/inactive/uncertain) with confidence and reasoning.
• Writes one output row per contact to an output CSV, including evidence excerpts and timestamps, and aborts if an unexpected runaway loop creates too many rows.

Use cases

• Verify whether leads or partners appear active on their public web/social presence.
• Update CRM-like lists by flagging inactive contacts automatically.
• Deprioritize ambiguous leads as uncertain for manual follow-up.

Technical details

• n8n nodes/parts: Code, Sticky Note, Convert to File, Manual Trigger, Read/Write File, Split in Batches.
• Integrations: OpenAI (gpt-4o-mini) for JSON classification; MCP browser endpoint for extracting visible page text.

Setup note: Provide OpenAI credentials, confirm the model in the OpenAI node, ensure your MCP endpoint is reachable, and set valid input/output CSV paths with matching columns.