okay i've now done the mongodb OA twice (first time for an internship, second time for new grad full time 2026 cycle) so i have some comparative data here.
format: HackerRank platform, 2 problems, 90 minutes. not timed per problem which is nice because you can allocate your own time. language is your choice, most people seem to go Python or Java.
difficulty: my honest take is medium-to-hard LC equivalent. first problem is usually something you can brute force and then optimize, like a modified BFS/DFS or a sliding window with some twist. second problem is where they get you, usually a harder graph or dynamic programming problem that requires actually understanding the structure, not just pattern matching.
first time i got: problem 1: find all shortest paths between two nodes in a graph (medium-ish, but the all-paths variant is trickier) problem 2: interval scheduling with dependencies, basically a DAG topological sort with constraints
second time: problem 1: something with string parsing and stack-based evaluation problem 2: a knapsack variant with multi-dimensional constraints, i didn't fully solve it
the thing that tripped me up: partial credit matters a lot. pass 6/10 test cases vs 0/10 is a big difference. so even if you can't get the optimal solution, get your brute force working first and then try to optimize. don't spend 80 minutes on a clean O(n log n) solution you never finish.
they don't do live coding for the OA, which is different from the onsite coding rounds. onsite apparently has a live coding round with an interviewer watching. i haven't gotten to that stage for the full-time role yet but that's what i've read here.
overall i'd say prepare at the medium-hard level. don't skip graph algorithms or DP just because you think they're rare. they're not rare at mongodb.