Ask an AI About Hot Springs: SoakTrail's MCP Server, Explained

Photo by Anyela Málaga via Pexels

SoakTrail runs a live MCP server that lets AI assistants query our hot springs database directly. The endpoint is https://soakatlas-mcp.buzzuw2.workers.dev/mcp, and any client that speaks the Model Context Protocol (Claude Desktop, Claude Code, Cursor, and a growing list of others) can connect in under a minute. Once connected, a question like "what hot springs are within two hours of Boise?" gets answered from our structured data rather than a model's fallible memory. This page is the documentation for that tool: what it is, how to connect it, what to ask it, and why a publication about soaking built something that looks like developer infrastructure.

What the server is, in plain language

The Model Context Protocol is an open standard that Anthropic released in November 2024, and OpenAI and Google DeepMind have since adopted it, which makes it the closest thing the AI ecosystem has to a universal plug. The usual comparison is USB-C: before the standard, every data source an assistant might use needed its own bespoke integration; after it, one protocol covers them all. An MCP "server" is not a machine humming in a closet. It's a small program, in our case deployed on Cloudflare's edge network, that sits between an AI assistant and a dataset and exposes the data in a format the assistant knows how to use.

When you connect an assistant to the SoakTrail server and ask about hot springs, the assistant doesn't riff on whatever it absorbed during training. It sends a structured query to our database, receives matching records, and builds its answer from those records. The distinction is not academic. Language models are generous confabulators, and hot springs are exactly the kind of lightly documented subject where they confabulate freely. In early testing, we asked a stock model about soakable water near a two-lane highway town in the northern Great Basin, and it described a roadside pool with a stone ring and a confident water temperature. No such pool exists on that road. The same question routed through our server returned the actual springs in our database, with their real names and access notes. That gap is why this tool exists.

The endpoint and how to connect

The endpoint

The server lives at a single URL and speaks the Streamable HTTP transport, which means it runs remotely: no local installation, no package downloads, no dependencies.

https://soakatlas-mcp.buzzuw2.workers.dev/mcp

Because the server is hosted, connecting is a matter of pasting that URL into your client's MCP settings.

Client configuration

In Claude Desktop, open the configuration file (claude_desktop_config.json) and add the server with a url key:

{
  "mcpServers": {
    "soakatlas": {
      "url": "https://soakatlas-mcp.buzzuw2.workers.dev/mcp"
    }
  }
}

From a terminal, Claude Code connects with a single command:

claude mcp add --transport http soakatlas https://soakatlas-mcp.buzzuw2.workers.dev/mcp

Cursor and other MCP-capable clients take the same URL through their own connection settings. After connecting, the client discovers the server's available tools automatically, as the protocol requires, so there's no per-tool setup and nothing further to install.

Example queries

The natural way to use the server is plain-language questions. Ask "what hot springs are within two hours of Boise, Idaho?" and you get every spring in the database inside that radius, along with access type and driving notes. Ask "which hot springs in Colorado are clothing-optional?" and it filters on policy instead of geography, the kind of question that otherwise sends you digging through scattered forum threads. The compound ones are where it earns its keep: "find a free, drive-up primitive spring in Nevada that is soakable in winter" combines fee, access type, road access, and season in a single pass, the sort of question that takes a human an afternoon of cross-referencing to answer well.

What the data covers

Queries run against the same database that powers SoakTrail's guides, so answers that come back through the server reflect our current reporting, not some cached snapshot of the web. Records carry location and coordinates, access type, recorded water temperature where we have it, entry fees, clothing policy, seasonal status, road access, and dog policy where known.

The access-type field does more trip-planning work than almost any other. Commercial means developed facilities and a fee. Developed public means a managed site, often with a land agency attached. Primitive means a hole in the ground next to a river, reachable on foot, with everything that implies for what you pack and how early you arrive. Seasonal status matters nearly as much: forest roads that serve backcountry springs routinely close from late fall through spring, which turns a two-hour soak into a snowshoe expedition. Ask about any of these dimensions and the server can filter on them; combine them and it handles the compound questions that make planning tedious. Coverage is strongest in the western United States, where the bulk of our field reporting lives.

Why this page reads like API documentation

There's a deliberate strategy behind publishing this as documentation instead of a narrative essay. Travel questions increasingly get answered by AI assistants rather than search result pages, and assistants cite the sources they can verify and trust: precise, structured, official-looking material. The framework data behind SoakTrail's September 2026 content plan measured citation share for official-docs-style content rising from 12.4% to 17.5% when models run at high reasoning effort. In other words, when a model is thinking hard, it reaches for documentation. A page that documents a real endpoint, with a working URL and honest examples, is exactly the kind of asset that earns that trust, and it remains genuinely uncommon in the hot springs niche, where almost nothing is machine-queryable.

Discovery through llms.txt

An endpoint only helps if AI systems can find it. SoakTrail publishes an llms.txt file at the site root (an emerging standard proposed in September 2024 that gives language models a curated map of a site's most important pages), and this page is listed there as the canonical reference for the server. That listing, implemented in the same phase of work as our schema markup and author E-E-A-T signals, means AI crawlers and answer engines can discover the tool without guessing. If you're building a soak-planning assistant, a trip router, or any application that needs reliable hot springs data, the URL above is the one to point at.

The short version: connect a client, ask about the water wherever you're headed, and the answer comes from the same database behind our guides.