AI back office for a Spanish real estate firm
A residential real estate wholesaler in Valencia, Spain
A back office run by Claude agents: finding underpriced properties, valuing them, contacting sellers, producing contracts and matching deals to investors, with people approving every step that matters.
The problem
The firm buys distressed and underpriced homes and passes the deals to a network of investors. Each property took around 45 minutes of manual analysis, and the rest of the process lived in spreadsheets and inboxes.
The design keeps a strict line: the numbers (valuations, tax, offers, scoring) come from tested Python code, while Claude agents handle the integrations, drafting and coordination, and stop for a person's approval at each gate.
- 45 min to 5 s
- to value a property
- 16,000
- lines of Python across 9 modules
- €3 a day
- operating cost ceiling
What I built
- Nine Claude Cowork projects, one per stage of the business, each with its own schedule, commands and connectors
- A custom MCP server with 20 tools, so the agents call tested business rules instead of working out the maths themselves
- Automated valuations from comparable sales within 500 metres, with a maximum offer worked out in under 5 seconds
- Motivated-seller scoring over Spanish listing text and public auction notices
- Land Registry document checks that flag differences from what the seller declared before any money moves
- Contracts generated and sent for e-signature, invoices synced to Xero, and deal folders filed in Google Drive
A full product specification, Airtable schema, onboarding guides for each role and a runbook, delivered as a PDF onboarding pack.
Tools: Claude Cowork, MCP (FastMCP), Python, Airtable, Slack, WhatsApp Business, Dropbox Sign, Xero, Stripe, Google Drive and Sheets, Docker, Railway
Diagrams and screenshots
