The API and MCP: your bankroll's research, in your own tools covered how to connect an agent (Claude Code, Codex, Cursor, or any MCP client) to EV Blacksite over the API or MCP. This entry is the next step, and it comes with a warning label: automating your own betting, at your own risk.

To be clear about what this is: it's about automating the mechanics of placing bets. Getting good at betting itself, the bankroll discipline and review loop that actually make you money, is a separate craft (the craft of staying solvent). An agent that places bets flawlessly on a bad process just loses money faster.

The boundary, stated plainly

EV Blacksite is an analytics product. It gives you the signals, the sizing, the recalibration, and the grading, plus every tool you need to do the research. It does not hold your sportsbook logins, it does not hold your funds, and it cannot place a bet for you. Whether a wager actually gets placed is your choice and your action, in your own tooling. Nothing you build on top of it is something EV Blacksite runs or manages for you.

That boundary is the whole point of this page. Automating the full loop, so that signals turn into placed bets without you logging in every night, is genuinely useful: it makes your betting passive instead of a nightly chore. It is also the one place where you are handing an agent access to your money, so the details matter more than anywhere else on the platform.

Keep it deterministic

The instinct is to point a smart agent at a book and say "figure it out." Don't. Screenshotting a page, reading it with vision, and clicking around by interpretation is slow, burns tokens, and is unreliable in exactly the spot you cannot afford it: typing a stake and confirming a bet.

Build it the other way. The agent's job is to fetch the opportunity and do the math. The book interaction should be a deterministic script, one per book: known page, known fields, known buttons. Spend the AI on pulling signals and recalibrating, never on guessing which text box is the stake.

I run my own betting bots and built this the hard way. What follows is that playbook distilled, not a copy-paste bot.

The playbook, one bet at a time

You tell the agent which book you're on, it pulls your signals, and it works each one in turn.

1. Pull the opportunity. Ask EV Blacksite what's live at that book. It comes back with your top signals: say, Team A moneyline +135, stake $100.
2. Open the exact bet at the book and check the price first. Pull up that game, that market, that side. Markets move, and the line you were shown may not be the line on offer now. Compare the book's current odds to what the signal said.
3. If the price moved, recalibrate. If the book is no longer +135 but +132, don't just take it. Run it through the recalibrator (the API and MCP both expose it): the odds are +132 now, is this still +EV, and what should I stake? It reads your strategy and answers. Maybe it's still +EV and you bet $102 instead of $100. Maybe the edge shrank and you bet $85. If it's gone negative, skip it and move to the next signal.
4. Fill the slip. Enter the side and the stake. On some books the amount you type nudges the price again (a bigger stake gets a worse fill), so if the odds shift when you enter the number, recalibrate once more. If they hold, you're clear.
5. Place it. Once the numbers are confirmed and still +EV, place the bet.
6. Close the loop back to EV Blacksite. Log the bet with log_bet, every time. That feeds the bet tracker, grades it against the close, and sharpens the confidence on your strategies for the next run (Auto-resizing: what it is and when to turn it on). A bet you placed but never logged is a blind spot in your own record.

Then loop to the next signal. Anything that reads negative, skip.

Start tiny, and watch every bet

Do not turn this loose at full size on day one.

• Shrink the stake to a test size, 5 to 10% of what you'd normally bet. If your unit is $100, run it at $10 while you prove it out.
• Watch every single bet at the start, one at a time. Say "place the next one," then watch it walk the whole process. You are hunting for the small, expensive mistakes: it forgot to re-check the price, it typed the number into the to-win box instead of the stake, it opened the wrong game, it filled the wrong field. Those are the errors that cost real money, and they are easy to miss if you jump straight to hands-off.
• Keep a human on the trigger. A good pattern: let the agent do everything up to the filled-in slip, then you eyeball the numbers and click place yourself. You get most of the automation and keep the final say over your own funds.

This is not "hey Claude, automate my betting." It is detail work, and you are the one on the hook if the agent invents a number or does the wrong thing. I learned these lessons by making them: wrong game, wrong fields, wrong text boxes. Be explicit, and be sure.

Which books you can actually automate

Not every book works, and picking the wrong one is how accounts get limited.

• Web-based only. If a book is phone-app only with no real web version, this approach doesn't apply. Some web books also enforce geolocation, so your computer has to pass a location check confirming you're in a legal state.
• Start with API-based venues. The cleanest path is a book with a real trading API, because it hands you exactly the tools and documents exactly how to use them. Kalshi and Polymarket are API-based. ProphetX has an API too, but access is gated: you request it, and it isn't open to everyone.
• Mind fees and fill rates on exchanges. A listed price of +133 is the market rate, not a promise. If it can't fill at that price, that price doesn't matter. On top of that, exchanges charge fees, and a taker fee (what you pay to grab whatever is resting on the market, which is what an automated bet usually does) can drag a +EV bet into negative territory fast. So recalibrate after you have the real fill terms and fee, and before you place, then confirm the edge survived.

The set of books you can cleanly automate is smaller than the set that exists. It isn't empty, and with the CFTC and prediction markets pulling more venues toward regulated, API-first models, it's growing.

Do this at your own risk

One more time, because it's the load-bearing sentence: books can limit or ban accounts they catch running automation. That is a real risk you accept when you build this, and EV Blacksite is not responsible if your account gets limited or banned. Your accounts, your books' terms, your jurisdiction, your call.

If you want to try: stick to legal, API-based venues, keep a human in the loop while you learn, and put multiple EV checks in the path, before fill, after fill with fees, and right before you place, so you never fire on a number that has quietly gone negative. Good luck.