Ed 
Try this right now. Open ChatGPT. Type in “what’s for sale in 78738 under $700K?” and hit enter. Go ahead, I’ll wait. Now open Zillow or Redfin and look up the same thing.
The lists won’t match. They won’t even be close.
ChatGPT just confidently invented addresses for you. Some of those houses don’t exist. Some of those prices are from two years ago. Some of those “features” were never in the actual listing. And the model said all of it with the same calm authority it uses for everything else.
Welcome to the AI data gap. Lets talk about why it happens and what I built to fix it.
Why ChatGPT and Claude Confidently Make Up Listings
Here is the part nobody explains in plain English. Large language models like ChatGPT and Claude are trained on a giant snapshot of the internet from months ago. When you ask them about “right now” they don’t actually know anything about right now. They pattern-match what listings USED to look like in their training data and produce something that LOOKS like a listing.
That is not a bug. It is a consequence of how the technology works. These models are extraordinarily good at generating plausible text. They are not databases. They have no live connection to anything unless you give them one.
So when you ask “what’s for sale in 78738 right now,” the model does what it always does. It produces plausible-looking text. The addresses look real. The prices look reasonable. The bed and bath counts are believable. None of it has to be true.
This is what AI researchers call hallucination. I call it confidently making stuff up. Same thing.
The reality is, an LLM without a live data connection is basically a really articulate friend who hasn’t read the news in six months but really really wants to help you. That friend is going to give you an answer. The answer is going to sound good. You should not buy a house based on it.
Why You Can’t Just Scrape Zillow
Ok so the obvious question is, why can’t AI companies just point their models at Zillow or Realtor.com and pull live data? Couple reasons. Some technical, mostly legal.
MLS data (the actual database of every listed property) is owned by the local Multiple Listing Service, which is in turn governed by NAR rules and standardized through RESO (the Real Estate Standards Organization). It is not public data. It is licensed data. Brokers pay to access it. We agree to a stack of rules about how we can use it, who can see what, and what we are allowed to display.
Public sites like Zillow, Redfin, and Realtor.com get the data through something called IDX (Internet Data Exchange). That is a feed brokers can opt into to display listings on their own websites. There are display rules. There are attribution rules. There is no version of “we just scrape it.” That is a terms-of-service violation, and depending on the MLS, it is also a contract violation by the broker who provided the feed.
Direct programmatic access (the kind you need to wire up an AI) comes through VOW (Virtual Office Website) feeds or RESO Web API endpoints. Both require broker membership, both require signed agreements, and both come with real consequences if you misuse the data. We are talking broker discipline, fines, getting kicked out of the MLS, and in some cases lawsuits.
So when an AI company wants to give ChatGPT real-time access to “what’s for sale,” the answer is not “scrape Zillow.” That path leads to a cease and desist letter. The answer is “partner with a licensed broker who already has a VOW or RESO feed.”
That brings us to MCP.
What MCP Is (in Plain English)
MCP stands for Model Context Protocol. It is an open standard that Anthropic released in late 2024, and it has since been adopted by basically everyone who matters in AI. OpenAI added support. Google did too. It is becoming the way AI models talk to outside data.
Think of MCP as a USB port for AI. You know how USB was a big deal because you could plug anything into anything? MCP is the same idea for AI tools. Build an MCP server for a service (your calendar, GitHub, a customer database, a real estate MLS) and any MCP-compatible AI client (Claude, ChatGPT in developer mode, Cursor, Perplexity Pro, Cline) can plug in and use it.
When the AI is connected through MCP, it stops guessing. It queries. The model asks the MCP server “give me 3-bed homes in 78738 under $700K with a pool” and the server returns actual rows from a real database. The model then writes the answer using real facts. No more hallucinated addresses.
That is the whole shift. The same AI you have been using suddenly knows real things, because someone gave it a real connection.
What I Built
I’m a realtor. I’ve been selling homes in Austin for the last 19 years, and at some point along the way I became a builder too. Code, databases, automations, whatever the job needs. Most days I’m not sure if I’m a realtor who codes or a developer who happens to sell houses, but the answer is probably both.
I also have broker-level VOW access to the Austin MLS through Unlock MLS (formerly ACTRIS). That means I can pull real listing data programmatically, under a licensed feed, fully within the rules.
So I did the obvious thing. I wrapped the VOW data in an MCP server.
Now when I open Claude and ask “show me 3-bed homes in 78738 under $700K,” I get real answers with real addresses and real prices. Not invented ones. Real ones. The first time I did it I just sat there for a minute. It worked. I wrote about that moment here because it felt like a small turning point. The AI was no longer guessing about my market. It was looking at my market.
That’s not that hard right. The technology has existed for months. The data has existed for decades. The only thing missing was a realtor willing to wire them together.
What This Means for Buyers
If you are house-hunting and using AI to help, the takeaway is pretty simple. Stop relying on ChatGPT or Claude or any other LLM to tell you what is for sale, on its own. Without a live data connection, it is making things up. Politely. With great confidence. Still making things up.
Instead, use AI tools that are wired to live data. Ask them comparison questions. Ask them analysis questions. The things AI is genuinely good at are things like “compare these three houses on schools, commute, and price per square foot,” or “of these eight properties, which has the best lot for a future pool,” or “summarize the price history of every house that sold in this neighborhood in the last year.”
Those are great AI questions. AI is actually good at that, when it has the data.
What it cannot do, no matter how smart it sounds, is invent reality. It will not know about the listing that hit the market this morning unless someone tells it. It will not know about the price drop last week unless someone tells it. The “telling it” part is the entire game.
If you have read either of my pillar guides (How to Buy a House with AI and How to Sell a House with AI) this is the missing piece those guides keep pointing at. AI plus live data equals a real assistant. AI minus live data equals a very articulate guesser.
How to Use Austin MLS MCP
This whole thing is live at Austin MLS MCP. The setup page walks through how to connect your AI client of choice, whether that is Claude Desktop, ChatGPT in developer mode, Cursor, Perplexity Pro, Cline, Windsurf, or Gemini CLI. There is a setup guide for each one.
To be clear about what this is. It is the same MLS data you would see on NeuhausRE.com, just delivered programmatically to your AI. You are not buying MLS data from me. You are paying a retainer to work with me as your buyer’s agent and the MCP access comes with that relationship. Same data, displayed two different ways. The Active Buyer retainer is $200/month and applies as a credit against commission at closing. If we work together I want you to have the best tools. This is one of them.
If you want a deeper dive on what MCP is in general (not just for real estate), I wrote about that here: What Is MCP? Why Every Real Estate Agent Should Care.
Where This Is Going
A few predictions, hedged appropriately because the AI space is moving stupidly fast.
National MLS coverage is coming. The MLS landscape has been consolidating for years and 2026 is accelerating it. Once a few large MLSes adopt MCP-compatible feeds, “AI that knows what’s for sale anywhere in the country” stops being a vision and becomes a Tuesday afternoon project for whichever brokerage wires it up first.
More MCPs are coming for everything else in real estate. Deed records. TCAD and other appraisal districts. Public records. Mortgage rate feeds. Permit data. The same trick that worked for MLS works for any structured dataset. The barrier was never the technology. The barrier was someone deciding to build it.
AI itself becomes a research partner instead of a confident guesser. That is the real shift. Right now most people think of AI as a slightly weird chat experience that occasionally makes things up. In two years it will feel more like a junior associate who reads everything, remembers everything, and has live access to every relevant database. Not because the model got smarter. Because the model got connected.
I would argue the brokerages and agents who refuse to engage with this are going to look like the ones who refused to put listings on the internet in 2002. Same energy. Same outcome.
The Realtor’s Take
AI without live data is a parlor trick. AI with live data is the new normal. Buyers who use both will outperform buyers who use neither, every time, no big deal right. Agents who use neither are going to find themselves competing against agents who use both, which is a very short and very ugly fight.
I’m not saying this because I built one. I built one because I believe it. The technology is here, the data exists, the connection between them is finally a solved problem. Now it is just about who shows up and uses it.
Frequently Asked Questions
Want to See It in Action?
If you want to use AI that actually knows what is for sale in Austin instead of inventing it, head over to Austin MLS MCP and pick your AI client. Setup takes about ten minutes. Once it is wired up, you have a research partner that actually knows the market.
If you want to talk through whether this is the right tool for your search, or you just want to discuss the Austin market in general, reach out here. I am happy to walk you through what we built and how to use it.
