Ask Claude "what's this house worth?" and it will reason beautifully about a number it doesn't have. Language models know everything about real estate except the actual data — current prices, Zestimates, what sold last month. The Model Context Protocol (MCP) fixed the plumbing problem: agents can call tools. What's left is giving them a tool worth calling.
apillow-mcp is that tool for Zillow property data: an MCP server that lets Claude Desktop, Claude Code, Cursor, Windsurf — any MCP-compatible agent — search listings and look up any US property, live. Install takes about a minute, and if you don't have an API key yet, your agent can start your signup from chat. Queries return in roughly 10–25 seconds because the API fetches live from Zillow.
Install
Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), or open it via Claude → Settings → Developer → Edit Config. If apillow-mcp is not found, use its absolute path as the command (which apillow-mcp / where apillow-mcp):
Claude Code — one command:
Add --scope user to use it in every project instead of only the current directory.
Cursor reads ~/.cursor/mcp.json (global) or .cursor/mcp.json (project) with the same JSON shape. Windsurf uses the same shape in its MCP config file. Full details are in the docs.
No API key? Sign up from chat
This is the part that surprises people. The server ships account tools alongside the data tools, so you can start signup from the conversation. Verification happens by email:
The free tier is 50 properties/month, no card. When the quota runs out, the agent can call upgrade_plan() and hand you a Stripe checkout link. Property queries run entirely in chat; signup needs one click on a verification email, and upgrading opens a Stripe checkout page.
The tools
| Tool | What the agent can do with it |
|---|---|
search_properties(query, type, max_items) |
"Homes for sale in Beverly Hills under $2M" |
search_by_zip(zipcodes, type, max_items) |
"What's on the market in 44124?" |
get_property_by_address(address) |
"What's 1874 Bromton Dr worth, and what did it last sell for?" |
get_properties_by_addresses(addresses) |
"Pull data for every address in this spreadsheet" |
get_property(zpid) / get_property_by_url(url) |
Lookups by ZPID or a pasted Zillow link |
check_addresses(addresses) |
Check which addresses currently have a Zillow listing |
signup(email) |
Create a free account; the key arrives by email after verification |
check_usage · upgrade_plan |
See your monthly usage and get a Stripe upgrade link |
Property results carry price, the Zestimate (for off-market and sold homes; active listings return the list price), beds/baths, square footage, lot size, year built, last sale price and date, listing agent, status, days on Zillow, HOA, the description, and counts of price-history entries, tax-history entries and photos. For the complete record — full price history, photos, comps — use the REST API.
What agents actually do with it
- Deal screening in chat: "Pull these 12 addresses, compare last sale price and Zestimate for the off-market ones, rank them" — one message, no spreadsheet.
- Listing checks: "Pull these 40 addresses and tell me which are for sale right now."
- Watchlist agents: a scheduled Claude Code task that re-checks a ZPID list and flags price cuts or status changes.
- Coding with real data: in Cursor, the fixture data in your tests can be actual property records instead of lorem-ipsum houses.
Why this beats a scraping MCP server
Most "real estate MCP servers" you'll find in directories are hobby projects that scrape Zillow from your machine on every tool call. They work in a demo and then meet Zillow's anti-bot stack — PerimeterX fingerprinting, rotating page structures — from your residential IP. The failure mode is silent: the agent just starts getting empty results.
apillow-mcp is deliberately boring: a thin client over the same hosted API our REST customers use in production. The scraping, anti-bot handling, and off-market sold-data enrichment run on our infrastructure, monitored around the clock; your agent makes an authenticated HTTPS call. Nothing brittle runs locally, and your IP stays out of it.
| Hobby scraper MCP | apillow-mcp | |
|---|---|---|
| Runs from your IP | Yes — blocks land on you | No |
| Sold data on off-market homes | Usually missing | Included (enriched) |
| Survives Zillow layout changes | Until the next one | Provider's problem, not yours |
| Onboarding | Clone, configure, pray | pip install, sign up from chat, verify by email |
| Cost | Free until it breaks | Free tier; from $9.99/mo |
Give your agent property data
pip install apillow-mcp — then ask your agent to sign you up. 50 free properties a month.
Related reading
- Zillow API in 2026: status, access, alternatives
- Zillow ZPID lookup: find the property ID for any address
- Zillow sold data API: sale dates, price history, and comps
- Zillow API Python tutorial
Final takeaway
MCP made it trivial to hand an agent tools; the hard part was always the data behind them. A Zillow MCP server backed by a production API gives Claude, Cursor, and friends real property data — prices, last-sale data, Zestimates for off-market homes — with a one-minute install and a signup you can start from chat.