I was trying to work out when to leave Cabo for Austin. Somewhere around the sixth identical Outsite search, I started wondering why I was still doing it by hand.

This is my first trip to Cabo. I rode along the Pacific, took a boat past El Arco at sunset, and kept finding good reasons to leave a few days later. That was part of why I wanted to keep the dates loose. I was planning a trip. Somehow I ended up building software.

My dates were still moving, which sounds harmless until every extra day changes the flight and every flight moves the Austin stay. Then a different set of rooms opens up. I wanted to know whether I could leave Cabo on a Tuesday, stay in Austin for a week, and get a room I liked.

Outsite's main date search starts with fixed dates. Mine kept moving, and the answer I wanted often sat one layer down, inside the calendar for a particular room.

These were the two places at either end of the search. Todos Santos looked like somewhere I could happily stay put. Travis Heights had the desk, daylight, and Austin location I was trying to line up.

Property photos from Outsite Todos Santos and Outsite Austin Travis Heights.

So I did what I apparently do when a travel site makes me repeat myself: I made a small MCP server. ChatGPT can use it to search public Outsite locations, quote exact dates, and inspect an individual room calendar across a wider window.

This is an unofficial, read-only project. It reads anonymous data from Outsite's public booking site. The server has no Outsite login, and its tools cannot book a room or make a payment.

What it does

It does a few specific things, all based on searches I kept doing by hand:

• Find an Outsite house from a city, neighborhood, country, or house name.
• Check the public room and rate combinations quoted for exact dates.
• Read the room description, bed type, size, and occupancy attached to each public quote.
• Inspect one room type across a wider calendar and find contiguous open periods.

Every result includes its source URL and the time it was fetched.

Install it in ChatGPT

ChatGPT calls this a custom app in the current UI. I still think plugin is the better name.

1. Open Settings, then Apps.
2. Enable Developer mode. Its location varies a little by plan and workspace role.
3. Choose Create app.
4. Name it Outsite Research (Unofficial).
5. Set the server URL to https://outsite-mcp.kristianeboe.me/mcp.
6. Choose No Auth.
7. Review the tools and connect the app.

Open a new chat, select the app from the tools menu, and ask it to check a stay.

OpenAI documents the current flow in Developer mode and MCP apps in ChatGPT. ChatGPT will warn you that this is an unreviewed server. That warning makes sense. You can inspect the source or run your own copy if you prefer to control the endpoint.

A test from my Austin search

I used this prompt:

Use Outsite Research (Unofficial) to find public availability and quoted rates at Austin Travis Heights for September 12 to 19, 2026. Return room names, rate types, and totals.

When I ran it on August 25, 2026, the app found Woodland at $554.98 and Monroe at $582.60 on the public member weekly rate. It returned guest rates for the same rooms too. Both rooms showed one available for the full week.

The prompt I care about more is:

Check each room calendar at Austin Travis Heights and show me contiguous stays of four to seven nights during the first two weeks of September.

This is the part I built it for. The room window becomes an input to the flight search, instead of a constraint I discover after choosing a flight.

I can also follow up with:

Open the public source pages for these rooms. Compare the photos and descriptions, then note the bed layout, desk, natural light, and bathroom setup.

That second step uses the agent's web or browser access. The MCP supplies the room data and the source page.

What it cannot promise

Rates and availability can change after a result is fetched. A rate labeled "member" on the public site does not prove that your account can book it. The server does not check membership eligibility.

Outsite's GraphQL API is undocumented. I added response validation, short caching, timestamps, and clear errors, but an upstream change can still break the tool.

The code is available under the MIT license at kristianeboe/outsite-mcp. The hosted server is the easiest way to try it. If you would rather not trust a random travel tool I built while planning my own trip, run your own copy. That is fair.

At some point I should probably stop turning travel planning into software projects and book the trip. I am not promising that will happen.

More is coming

The Outsite MCP is the top of the iceberg. I write about the things I build, the places they take me, and whatever rabbit hole comes next.

Keep me posted

Occasional emails. Unsubscribe whenever you want.