STAGE 01 · DRAFT
Pick your card
Take cards as you go. The assistant tracks your pool and reads which colors, tribes, and archetypes you’re leaning into.
Magic The Gathering draft assistant
Practice-draft real packs against AI. Every card gets scored, reasoning shown: real 17Lands data, Karsten mana math, weighed against your pool. When the draft ends, it builds your best deck.
Free · runs in your browser · nothing to install
What it is
Open a set and draft real packs against AI, just like MTG Arena. As your pool grows, each pack is ranked by how much every card helps your deck: your colors, your synergies, your curve. When the draft ends, it builds the deck for you. Re-draft as much as you want, then take that instinct into a real queue.
How it works
Three stages, following the shape of a real draft.
STAGE 01 · DRAFT
Take cards as you go. The assistant tracks your pool and reads which colors, tribes, and archetypes you’re leaning into.
STAGE 02 · SCORE
Every card gets one contextual score: base quality, synergy, curve fit, color commitment, removal count, castability. Not a table of stats to weigh yourself.
A table of aggregate win rates
| Card | GIH WR | ALSA | IWD |
|---|---|---|---|
| Black Lotus | 66.4% | 1.2 | +9.8% |
| Dusk-Wing Skirmisher | 54.8% | 6.1 | +1.9% |
| Corona Elemental | 58.9% | 2.4 | +5.0% |
| Hollowfen Ritual | 52.2% | 7.8 | -0.4% |
Now weigh them against everything you’ve already drafted, by yourself.
Black Lotus, for your pool
The same real data, turned into the pick.
STAGE 03 · BUILD
The optimizer weighs every card against real Karsten mana math: sources needed to cast it, what a splash actually costs. What comes out is castable, curved, and built to your colors, not just your best cards stapled together.
Want to see exactly how each number is calculated, from base rating to synergy lift to castability?
Why it’s different
Plenty of tools hand you a number. This one shows its work: every score opens into the math behind it, built from real 17Lands data for the pool you actually have.
Every pick opens into a full breakdown: base rating, synergy, curve, removal, castability. No black box, no mystery number.
A deck optimizer assembles the best deck from your pool with real mana and castability math: splash discipline, fixing, and curve included.
A co-occurrence “lift” signal learns what’s actually played together in winning decks: tribal, archetype, and color synergy, not isolated card win rate.
One score for the pool you actually have, not a spreadsheet of isolated win rates to merge in your head each pick.
Ready when you are
Pick a set, open a pack, and watch the pack rank itself against your pool in real time.
Our model is built on real 17Lands data drawn from winning drafts (decks with 5 or more wins on Magic The Gathering Arena), combined with Frank Karsten’s mana probability models for castability and fixing. 17Lands data is used under CC BY 4.0 and is transformed here (calibrated, aggregated, and re-scaled) rather than shown raw. 17Lands does not endorse this site.
Each card’s core value is data-driven: a calibrated base plus two learned signals. The breakdown panel shows this section first.
Context nudges toward how real winning decks are built. Most are count-based and fade out as your pool reaches the target measured from real 5+ win drafts of this set, so the engine stops rewarding what you already have enough of; the ML Pick Model instead learns from real picks and fades by pick number, not pool composition.
Auto Build searches for the deck that maximizes total value × castability: every card’s contribution is weighted by how reliably your mana can cast it (Frank Karsten’s probability models for source counting over your actual lands and fixers). A splash is charged both for its own inconsistency and for the sources it steals from your main colors, so high-value bombs keep most of their value when splashed and filler does not. Land count and basics are recomputed from the deck’s pips each iteration. Co-occurrence, Open Color, Wheel, and the ML Pick Model are draft-only terms and don’t appear here; deck synergy is carried by the Tribal/archetype term.
No model is finished. Here’s where ours is still wrong: known issues and what’s next.