AI agent use cases in real estate: seven workflows built on CASAFARI's property and location intelligence

Why an AI agent needs one record per property before it needs anything else

Consider a two-bedroom flat in Lisbon listed across four property portals by three different agencies under two distinct asking prices. An unassisted artificial intelligence model that scrapes these portals directly will count four available units of supply, miscalculate the average time on market, and present an inflated estimate of local inventory.

CASAFARI is a real estate data platform built for AI agents: a deduplicated, cleaned property graph of residential and commercial property, for sale and for rent, in 16 countries, in which every property is one record with every price change, status change and delisting (the CASAFARI property graph, October 2026). When portals list the same property multiple times, CASAFARI collapses them into a single record. Each portal listing is linked back to the property's single history; our help centre explains what happens when the same property is listed twice.

Being agent-native means the platform is designed for both human operators and autonomous language models:

• Unified accounts: One account, one key, and one shared balance of tokens cover every product.
• Browser sign-in: Sign in via OAuth 2.1 in the browser without copying private keys into prompts.
• Read-only tools by default: 19 of the 21 documented MCP tools perform read-only actions.
• Machine-readable documentation: Agents consume llms.txt, OpenAPI endpoints, auth.md, and native Markdown documentation directly.
• Transparent usage: Each tool call consumes a fixed number of tokens declared directly in its description.

Protocol governance shifted in December 2025 when MCP joined the Agentic AI Foundation under the Linux Foundation. With this open standard, one MCP server per product connects directly to Claude, ChatGPT, Cursor or Gemini CLI; see Casafari MCP.

Use case 2: a market scan across districts in minutes, not a fortnight

A research analyst at a European fund in London needs an executive summary of district-level performance across Madrid before Monday's investment committee meeting.

1. User prompt: "Which districts of Madrid have seen the strongest price growth over three years, and where is stock selling slowest?"
2. Location resolution: locations_search resolves Madrid and its constituent sub-districts into location IDs.
3. Heatmap generation: The agent executes area-insights_get_heatmap to retrieve district-level metric distributions.
4. Distribution profiling: The agent runs price distribution and time-on-market distribution tools to inspect market skew.
5. Quantitative operations: The agent calls area-insights_get_time_series_operations to compute Compound Annual Growth Rate (CAGR) and price volatility per district.
6. Executive delivery: The agent returns a ranked table with district-level commentary, explicitly stating the date range for each series.

Distributions reveal crucial nuances that simple medians obscure, such as bimodal pricing in gentrifying areas, while volatility metrics provide credit committees with risk context. Learn how to compare neighbourhoods with a heatmap and analyse how long it takes to sell: time on market or visit Area Insights.

Use case 4: investment screening by rental yield and price below the area

A private investor looking abroad wants high-yielding two-bedroom flats in Lisbon priced below the local market average.

1. User prompt: "Which districts of Lisbon offer the highest gross rental yield for two-bedroom flats, and which listings there are priced below the area?"
2. Location identification: locations_search obtains location IDs across Lisbon sub-markets.
3. Yield calculation: The agent queries Area Insights historical sale and rental price series per district to calculate gross yield estimates, explicitly listing the formula and source dates used.
4. Graph search: The agent queries Property Sourcing over REST (using an API key passed as Authorization: Bearer) to search the graph for available two-bedroom flats in top-yielding zones.
5. Underpriced identification: The agent isolates listings whose asking price per square metre falls below the district distribution and presents each asset alongside its full price history and all historical portal listings.

Gross yield calculations serve as an initial benchmark and do not account for void periods, local taxation, or operational expenses. Read more on calculating rental yield by area and how to find properties priced below their area, or visit CASAFARI for investors.

Use case 6: screening agency and agent partners in Germany with AgentGraph AI

An expansion lead at an international real estate network evaluates potential agency franchise partners or referral allies across Munich and Berlin using measured listing activity.

1. User prompt: "Which agencies in Munich match our partner profile: resale focus, mid-to-upper segment, active in Schwabing and Bogenhausen?"
2. Agency screening: The agent invokes agentgraph_screen_agencies to match active brokerages against profile parameters.
3. Territory verification: The agent executes agentgraph_territory_map to verify candidate listing distribution across target neighbourhoods.
4. Brief generation: The agent calls agentgraph_partner_brief to construct detailed profile briefs for candidate agencies prior to initial meetings.
5. Shortlist delivery: The agent returns a ranked agency list supported by operational market metrics.

AgentGraph AI is available exclusively for verified business entities operating in Germany. Its watchlist modification tools are among the few write operations available and require explicit human confirmation before execution. Read how to find partner agencies in Germany with AgentGraph AI and examine agentic AI for global real estate networks.

What keeps these agents safe: OAuth, read-only tools and tokens a person controls

Connecting AI models to live market data requires strict security controls. CASAFARI enforces OAuth 2.1 with Proof Key for Code Exchange (PKCE) and dynamic client registration. Client applications never handle account passwords directly, access tokens are restricted to designated servers per RFC 8707 with a 15-minute validity window, and refresh tokens rotate automatically upon use.

Nineteen of the 21 documented MCP tools perform read-only actions (readOnlyHint). The two tools capable of modifying data (AgentGraph AI watchlist modifications) prompt the human operator for approval prior to execution. API credentials travel strictly via Authorization headers rather than standard URLs, and tool responses are isolated as data inputs to prevent prompt injection.

Note that call execution costs and account allowances are denominated in platform tokens, whereas authentication credentials generated during OAuth sign-in are referred to as access tokens. Review the CASAFARI security model for agents for full architecture details.