Behind the Scenes
You search for a card. We find cards that work similarly. Simple enough. But how do we actually decide what's “similar”? Here's the full breakdown of what happens when you hit search.
We pull from Scryfall's dataset
Every card ever printed. Oracle text, type lines, keywords, mana costs, color identities—all of it. We normalize and index this data so we can compare apples to apples (or Grizzly Bears to Centaur Coursers).
Compare everything against everything
Your target card gets scored against the entire card pool. We use different techniques for different attributes:
Text similarity
TF-IDF vectorization + cosine similarity. Basically: do these cards say similar things in their rules text?
Type & keyword overlap
Jaccard similarity. If both cards share 3 out of 4 keywords, that's a 75% match.
Color identity
Exact matches score highest. Subsets (like mono-red vs. Boros) score well. Overlaps get partial credit.
Stats (CMC, P/T, loyalty)
Distance-based scoring. A 3/3 is more similar to a 3/4 than to a 7/7.
Weight and combine the scores
Not all factors matter equally. Oracle text and type line get the most weight. Color identity and keywords help refine. Stats matter more for creatures than for instants.
* These weights are configurable and may change as we tune the algorithm.
Filter with an adaptive threshold
We sort all candidates by score and apply a dynamic cutoff. The threshold adapts based on how good the top matches are—so you get a focused set of results, not a firehose of mediocre hits.
Translation: if your card is weird and one-of-a-kind, you'll get fewer results. If it's a common effect, you'll see more.
Tag with playstyle hints
Finally, we look at patterns across the results: colors, card types, keywords, and common rules text. Does the card create tokens? Draw cards? Ramp? We tag it. These aren't categories set in stone—they're just hints to help you explore.
Think of these tags like labels on bulk boxes at your LGS: helpful for browsing, not gospel.
Some quick caveats
This tool is not affiliated with Wizards of the Coast. All card data comes from Scryfall, which generously makes it available for projects like this.
Similarity is a heuristic—a useful approximation, not a perfect science. Use these results to spark ideas, not as the definitive answer on what cards are “the same.”
The weights, thresholds, and exact methods will evolve as we refine the algorithm. If something feels off, let us know.
Now go forth and build something spicy.