Where is my god damn train?
A tale of winter, despair, and the one question your AI agent could never answer. Until now.
Skip the story, jump to setup →You’re on the platform. An Antarctic wind whips through your bones. You clutch a small strong latte for the last warmth on Earth. As the rain begins to pelt you, you shed a single stoic tear: the daily Melburnian ritual.
Wasn’t AI supposed to save us from this? You whisper it into the wind:
“Where is my god damn train?”
For years, your agent had nothing for you. It could write a haiku about trains. It could swear blind that the ancient Comeng you were shivering on was one of the shiny new HCMTs (train joke). It could not, for the life of it, tell you when the next one was actually pulling in.
Now it can.
PTV MCP plugs Melbourne’s live public transport straight into your AI agent. Trains, trams, buses, V/Line, the lot. Ask when the next one leaves, whether it’s running late, whether the whole line’s fallen over today, or how to get from where you are to where you need to be. It answers from live Public Transport Victoria data.
The trick is giving it less to do
Here’s the thing nobody says out loud about AI agents. They’re wonderful improvisers and hopeless clerks. Set one loose to “find the train times” and it’ll go scrape a website, misread the table, mistake yesterday’s timetable for today’s, and hand you something confidently, plausibly wrong. Ask again tomorrow and you’ll get a fresh wrong answer.
The fix isn’t a smarter agent. It’s a dumber one. Every time you swap a guess for a real tool that returns a real answer, you take a little agency away from the model, and the whole system gets steadier for it. That’s all PTV MCP is: your agent stops improvising the timetable and starts looking it up. Same question, same right answer, every time.
Behind the curtain it does the boring, deterministic bits: talks to the PTV Timetable API, turns the machine-speak into plain station names and Melbourne time, and works out the minutes until your train so the model never has to guess. You never see any of it. You just get the answer, and maybe, just maybe, make your train.
Give it to your agent
No install. No API key. The server lives at one address; you point your agent at it and you’re done. Don’t know how? Ask your agent.
https://ptv.southfoundry.co/mcp
Streamable HTTP
Free to use
claude mcp add --transport http ptv https://ptv.southfoundry.co/mcp
{
"mcpServers": {
"ptv": {
"type": "http",
"url": "https://ptv.southfoundry.co/mcp"
}
}
}
Then ask it anything a frozen commuter would
- When’s the next train from Flinders Street?
- Is the Frankston line disrupted this arvo?
- How do I get from Richmond to Box Hill right now?
- What’ll a zone 1 + 2 trip cost me on myki?
Your agent picks the right tools, checks the live timetable, and tells you. One less reason to cry on a platform.
We build these for a living
Got data your AI ought to be using and isn’t? That’s exactly our kind of problem. The South Foundry Co builds AI agents that do real work.
30 minutes, no slides · southfoundry.co · [email protected]