First thing: t212-autopilot doesn't make money. Its own README says the signals it was built around averaged −0.187R over 213 picks, and the one pattern that looked positive got all of it from a single trade. That's why it defaults to the demo account, a £100 ceiling and only 10% on picks. It's a tool for running a plan you've already proved, not a way to make one work. Not financial advice.

With that said, I've written a proper setup guide, and it's tested every week.

A fake broker

You can't let CI near a real brokerage account, demo or not. So the check runs the real CLI against a fake Trading 212 API that answers the reads the CLI makes and records orders instead of placing them. That lets it prove the safety rules, not just that the code imports:

The weekly check against a fake broker

  • plan prints the orders and the broker receives zero
  • T212_MODE=live without --yes-live prints the plan, says REFUSED and exits 2, and the broker still receives zero
  • demo apply sends exactly the planned orders
  • a plain-http picks feed is refused
  • the MCP server answers discovery, but every tool call without the right token gets "not authorised"

Short version 

git clone https://github.com/casareanderson/t212-autopilot && cd t212-autopilot python3 -m pip install -r requirements.txt cp .env.example .env # a DEMO key and secret set -a; . ./.env; set +a python3 -m autopilot check python3 -m autopilot plan --picks picks.example.json

Generate the key on the demo account. A live key gets 401 from the demo host, and that 401 nearly always means the wrong account, not a bad key.

autopilot plan against the example feed

The guide explains the limit you'll hit first: Trading 212's public API has no price endpoint, so ETFs you don't already hold get skipped. Full guide: docs/SETUP.md.

Has anyone actually found an edge they'd automate with real money? Genuine question.

🤖 Drafted with AI assistance from my own homelab notes, logs and repos, then reviewed and edited before publishing.