AI Operating System for Real Estate | n8n, Claude & MCP

People hear “AI agent” and picture something that runs the whole business by itself. honestly? that’s the quickest way to break one

This project was built around the opposite idea. real estate research, lead enrichment, CRM updates, content. same repetitive loop, pulled from multiple sources, eating hours that should go to actual deals

So i developed an AI operating system on n8n, Claude Code agents and MCP tools. everything enters through one front door, a webhook, an email, a form or a scheduled job. then an orchestrator decides who handles it. one agent for research, one for enrichment, one for reporting, one for content. each does one job well

The agents pull live data through MCP servers, plus a RAG layer with vector search, so answers come from the company’s own knowledge and not from guesses.

The part I’m most careful about is the approval gate. nothing important touches the CRM or goes out to a lead until a human says yes. agents do the heavy lifting, a person owns the final call

And because things fail, APIs time out, data comes back messy, I developed error handling, monitoring and retry logic into every step. one bad record doesn’t kill the run

It all connects into HubSpot, Airtable, Supabase, Google Workspace and Slack, so the team keeps working in tools they already use