For agents
Everything here is machine-readable.
Point a client at the files below and it can work out the URL grammar, the one JSON endpoint and the fact that there is nothing to log in to. Start with llms.txt. Take the OpenAPI description when you need scores for exact dates.
The files
Take whichever your client can parse.
llms.txt
Start here
The URL grammar, the questions this site answers and when a plain page beats the API.
Open fileOpenAPI 3.1
The JSON endpoint
Parameters, response schemas and structured errors for one destination across exact dates.
Open fileauth.md
Nothing to sign up for
No key, no token, no registration flow. This is the document that says so in writing.
Open fileRFC 9727
API catalogue
A linkset tying the API to its description, its documentation and its metadata.
Open fileARD 1.0
Resource catalogue
An inventory of the agent-facing files here, each one listed with the questions it answers.
Open fileJSON
Skills index
Versioned skills for dated, monthly and comparison questions, each with a digest so a client can spot a changed file.
Open filePages for browsing, API for exact dates
Use the prerendered pages when the question is broad: what a country covers, how a destination behaves across the year, what a month looks like or how two places compare over one window. They are meaningful HTML that works without JavaScript, and several of them answer in Markdown if you ask for it with an Accept header. llms.txt names which ones do.
Use the JSON endpoint when you need the score and the factor breakdown for one destination across exact dates. Every payload carries its source_version, so the same inputs on the same version give the same answer. Cache against it and come back when the version moves.
No key, no catch, one limit
Nothing here wants an account, a token or an Authorization header. The API is read-only, and the only thing between a client and the data is a fair-use limit on /api/: 20 requests every 10 seconds per IP, then a 429 from the edge. The prerendered pages carry no limit at all.
There is no MCP server. The pages and the one JSON endpoint already cover the jobs WenToGo does, and a third protocol would only be a new way of saying the same thing.
We sweat the data.You enjoy the days.
