Connect FindHomes once, then just ask your AI to search, value, and compare South African property. It does the searching, the maths, and the market analysis for you.
Included with FindHomes Plus. No API keys, no code — sign in once and ask.
These are based on real interactions — one message in, deep analysis out, in whichever AI tool you use.
Private Sale Valuation
Dual-Commute Search
Branded Market Report
Your assistant chains multiple data sources and tools together in a single response: from location lookup to rendered map to branded report.
"Is R12.5M fair for 33 Alma Road?" Compares against recent sales, factors in views, condition, solar, parking, and more.
Search like you talk, or point your assistant at a photo of a property you love — it understands both.
"15 min drive from Sea Point High during morning traffic". Searches within drive-time zones, even dual-commute overlaps.
Historical sales data, price growth trends, days on market, sale rates, and months of supply for any suburb going back years.
Gross yield, cap rate, rental benchmarks, and negotiation strategies. Know if an investment property stacks up before you commit.
"Durbanville vs Brackenfell vs Kraaifontein": side-by-side median prices, R/m², liquidity, and growth potential.
Scrapes a development's marketing page, extracts unit prices, and benchmarks against resale market data. Finds the best-value units.
Professional reports with your company's branding. Cover pages, market stats, and listing tables delivered straight into your chat.
"Is now a good time to sell?" Combines FindHomes data with live web research on rates, trends, and new developments.
Three steps, about a minute. You never leave the AI tool you already use.
The AI integration is included with FindHomes Plus — nothing extra to buy.
Sign in with your FindHomes account when your assistant asks. No API keys to copy, no code to write.
Ask for a search, a valuation, or a suburb comparison in plain language. Your assistant does the rest.
FindHomes connects to your assistant using MCP (the Model Context Protocol) — an open standard, supported by ChatGPT, Claude, and Cursor, that lets an AI assistant securely use external tools on your behalf. You don't need to understand any of it to use FindHomes: sign in once and ask. Building your own app? The developer details are further down the page.
Pick the AI tool you use and we'll show only the steps for it. Sign in once, then ask about South African property.
Building your own app or agent? See the full tool reference and rate limits.
Name a suburb or describe a commute, then search listings, historical sales, rental yield or a suburb comparison — all in one conversation. Commute zones are resolved on the server, so the model never handles polygons.
The tools you will use most. Each accepts suburb names or a commute directly, so a single call usually does the job.
search_propertiesSearch residential for-sale listings in South Africa. Scope by suburb names (locations), location ids, a commute zone, the overlap of two or three commutes, or nothing at all for a nationwide search. Filter by price, bedrooms, bathrooms, flatlets, parking, property type, status, agency, pet-friendliness and a semantic visual_search_description; sort by price, R/m² or listing date. Every result carries a findhomes.co.za URL.
Try asking
“Find pet-friendly 3-bed apartments in Sea Point under R5M with mountain views.”
search_property_salesHistorical sale transactions from deeds data, with optional statistics (average, median, min, max, std-dev) and market activity (days on market, absorption, price reductions). Needs a location scope: suburb names, location ids, a commute zone or a polygon. Filter by sale date, price, land or floor size and property type.
Try asking
“Compare median sale prices for 3-bed houses in Durbanville vs Brackenfell over the last 24 months.”
property_detailsFull details for up to 20 listings by FindHomes property id: rooms, sizes, features, agent and agency contact details, the listing description and the canonical URL.
Try asking
“Give me the agent contact details and full description for the first two results.”
rental_comparablesComparable rental listings and a gross-yield estimate for a for-sale property. Pair it with property_details and search_property_sales for valuation and investment questions.
Try asking
“What rental yield could I expect on this Observatory flat at R1.8M?”
screen_locationsCompare up to 10 suburbs side by side: median sale price, R/m² land, days on market, percentage under offer, percentage expired, months of supply and sales count over a window (default: the last 24 months). Use it before deep-diving into one suburb.
Try asking
“Screen Wynberg, Plumstead, Claremont and Kenilworth for a first-time buyer.”
Optional previews and disambiguation helpers. The search tools resolve names and build commute zones on the server, so these are for checking a zone or picking between two Kensingtons.
location_resolveUsually unnecessary: the search tools accept suburb, area and region names directly. Call it to inspect candidates, or after a result with status needs_disambiguation, to see id, name, parent, type, live listing count and similarity per input.
Try asking
“Which Kensington do you mean — Cape Town or Johannesburg?”
location_isochronePreview a travel-time or distance isochrone around a destination: geocode result, area in km², bounding box and vertex count. It returns no polygon — the model passes the same commute fields to the search tools and the server resolves the zone. arrive_by (driving only) defaults to 09:00.
Try asking
“How big is a 25-minute morning-drive zone around UCT, arriving by 09:00?”
isochrone_intersectPreview the overlap of two or three commute zones — for example both partners' workplaces — ranked by area, or a NO_OVERLAP status with suggestions. Then the same commutes array goes to search_properties or search_property_sales.
Try asking
“Where can we both live within 30 minutes of Century City and Claremont?”
Only used by clients that render MCP App widgets, such as ChatGPT.
render_property_resultsIf your client renders MCP App widgets (ChatGPT), this displays search_properties results as property cards. Other clients present the results with their URLs directly.
Try asking
“Show those results as cards.”
We rate-limit per signed-in user, not per client, so the cap is yours to use across ChatGPT, Claude, and any other MCP client at the same time.
60 tool calls / minute
Plenty for human-paced conversation. The cap is on tools/call only — control-plane methods like tools/list and initialize are free.
1,000 tool calls / day
The day window resets at midnight UTC. A power user typically runs 50–100 tool calls per day; the cap is here to catch runaway loops, not throttle real work.
When you cross 80% of the daily cap, we send one email so you can decide whether to slow down or get in touch about a higher limit. You won't get spammed — exactly one warning per day, even if you keep tool-calling after.
The tool call returns HTTP 429 with a Retry-After header telling the client exactly how many seconds to wait. ChatGPT, Claude, and Cursor handle this transparently — you'll see a short pause, then the call resumes.
Need a higher cap for an internal tool, an evaluation harness, or a partner integration? Email founders@findhomes.co.za and tell us about your use case.
We only see your tool calls — we don't see anything else, and you can revoke us at any time.
These are the only OAuth scopes we accept today. We do not use scopes to gate destructive actions; sensitive operations (like saving searches) are gated by your FindHomes account permissions on the server side.
Tool-call audit rows are stored on FindHomes infrastructure for support, abuse prevention, and rate limiting. We never sell or share this data with third parties. The full policy is in our privacy policy.
Connect FindHomes to ChatGPT, Claude, or Cursor, sign in with your account, and let your assistant handle the searching, the maths, and the market analysis.