How I used MCP to compare apartment rent and fees in Austin

Michael P — Mon, 12 Oct 2026 11:43:08 +0000

One Austin apartment showed $1,302 in base rent and a $1,406.62 monthly leasing price. Over its 14-month term, that's $1,464.68 beyond base rent. Utilities still need their own budget.

I opened Zoey's listing for apartment #1-1501 to see what an AI-assisted comparison could actually tell me. At 2Captcha, I used our MCP to read the listing, then checked the figures in the browser. I'd seen the kitchen renderings. Time to look at the bill 🙂

All prices and availability below were checked on October 12, 2026.

Two prices, one apartment

Zoey #1-1501 was a one-bedroom, one-bath apartment with 720 square feet, listed as available now. The $1,302 base rent applied to a 14-month lease. The same page displayed $1,406.62 as its Total Monthly Leasing Price.

I passed the unit URL to scrape_page. MCP returned both figures, the lease term and the apartment number. I opened the page myself and found the same values.

The property's definition includes base rent, mandatory monthly fees and any user-selected optional fees. Variable charges and move-in or move-out costs are excluded. I kept that definition alongside the price. A column labelled "rent" would've left too much room for interpretation.

For an October 12 move-in date, the browser also showed estimated application costs of $250 and estimated move-in costs of $1,487.68. I recorded those separately. They aren't another recurring monthly charge.

What's inside the monthly total

For an itemized example, I opened apartment #3083 at The Asher. It has two bedrooms, so I wasn't treating it as an equivalent alternative to the Zoey one-bedroom. Its fee breakdown was useful, though.

For a 12-month lease, the page showed $2,065 in base rent and a $2,130 monthly total. The $65 difference consisted of $5 for pest control, $35 for trash and $25 for parking required on this floor plan.

I asked parse_marketplace for the apartment number, base rent, monthly total, lease term and fees as separate fields. In that extraction, MCP returned 3083, 2065, 2130 and 12 months, along with the mandatory fee rows. I checked them against the browser calculator.

That gave me a monthly figure of $2,130. Adding the $65 again would've counted the same charges twice. Application fees and potential deposits belong in separate columns.

The offer I kept for a closer look

I kept Zoey #1-1501 for the next round of research, using $1,406.62 as the monthly leasing figure for its 14-month term. Its application and move-in estimates stayed separate, with utilities and deposit conditions still to confirm.

The site advertised up to six weeks of free base rent on selected apartments. I hadn't confirmed that #1-1501 qualified, so I left that concession out of my calculation. "Up to" has plenty of room to do work in a leasing ad.

I've got a specific question ready for the leasing office: "Can you confirm the full pricing for #1-1501 on a 14-month lease, and whether this apartment qualifies for the promotion?" The monthly figure I'll check against their answer is $1,406.62 before variable charges.

Sources
2Captcha MCP: https://2captcha.com/ai/web-mcp
Zoey #1-1501: https://livezoey.com/floorplans/unit-d6a23624f857e627f54d8f2a86382f57/
The Asher #3083: https://www.theasheraustin.com/floorplans/b1/3083/
The Asher calculator: https://www.theasheraustin.com/floorplans/b1/3083/calculator/

Five links, no GPUs: how I used MCP to shop for a local LLM

Michael P — Mon, 12 Oct 2026 09:51:38 +0000

Listings checked on October 5, 2026.

Running a model through an API for every routine task was starting to feel expensive, so I began looking at a local setup. First question: which GPU could I actually buy for it?

I started with 24 GB of VRAM as a shopping filter. More room for model weights and context was the appeal. The exact model, quantization, and context length would still determine what fits, but this gave me a concrete place to start: an RTX 3090 with 24 GB.

I could have asked an LLM for “the best GPU for local AI” and received a very confident top-five list. Useful for learning the specs, less useful when the card in that list is sold out, the price belongs to a different seller, or nobody can tell you exactly which board you're getting.

At 2Captcha, I used our MCP for web search and structured extraction from product pages, then checked the promising page myself. I wanted to get from a search query to an offer I could verify. Fewer tabs and, ideally, fewer expensive guesses.

A surprisingly unhelpful five results

I searched for a 24 GB RTX 3090 in the US with search_web. It returned five links. The first was RTX Corporation. As in aerospace and defense. Another was Raytheon. One was an RTX stock quote. NVIDIA's general RTX page made an appearance too.

Not a single product listing.

That was a pretty good reminder that five search results are not five offers. An LLM will happily compare snippets if you ask it to. The comparison may even look convincing. It still won't tell you whether a specific GPU is available to buy.

So I used search for reconnaissance and moved to direct retailer pages for the actual checks. I chose those URLs myself and passed them to parse_marketplace through the same MCP. The fields I cared about were deliberately boring: exact model, price, currency, stock status, condition, and the source URL. Boring is good when you're about to spend four figures.

From search results to a shortlist

The Walmart listing for a refurbished ASUS RTX 3090 had an attractive-looking $1,139.05 price. The structured result also said out of stock. That's not a deal, that's a number attached to an unavailable card.

On Newegg, MCP surfaced a refurbished Founders Edition with a price of 2,395. I confirmed the $2,395 price on the retailer page. Some pages returned fewer fields or restricted access, so I kept comparing candidates.

The listing I kept for a closer look was a refurbished MSI RTX 3090 Gaming X Trio 24 GB at GotoDirect. Its MCP result included part number G3090GXT24 in the title, plus $1,654.74, USD, and InStock. I opened the retailer page on October 5 and saw the same part number, price, stock status, and refurbished condition.

There's a catch visible right on that page: “Image may differ from original.” So a matching product feed still isn't a substitute for checking the actual unit, warranty, returns, and final delivered price. I wouldn't buy a refurbished GPU on the strength of a stock photo and a green IN STOCK label alone. I'm trying to run a local LLM, not start a small collection of expensive surprises 🙃

A part number beats a top-five list

The MSI listing made my shortlist because it matched the 24 GB filter and gave me an exact part number, price, and stock status I could check against the store. At $1,654.74 refurbished, price and warranty would still need scrutiny before checkout. What I had settled was which offer deserved that scrutiny: G3090GXT24, with a source behind every field.

The first search took me to aerospace pages. The second pass led to G3090GXT24. I'll take the part number.

DEV Community: Michael P