The store creates a compound unique index on (namespace, key), but namespace is stored as an array. In MongoDB this makes it a multikey index - each array element gets its own index entry - so two documents whose namespaces share any path element and have the same key collide.

store.put(("users", "alice", "preferences"), "food", {...})
store.put(("users", "bob",   "preferences"), "food", {...})

pymongo.errors.BulkWriteError: E11000 duplicate key error
  collection: assistant_demo.memories
  index:      namespace_1_key_1
  dup key:    { namespace: "preferences", key: "food" }

Note namespace: "preferences" - a single string, not the array. That's the multikey entry. Alice's doc already claimed ("preferences", "food") and ("users", "food"), so Bob collides.

Source
https://github.com/langchain-ai/langchain-mongodb/blob/main/libs/langgraph-store-mongodb/langgraph/store/mongodb/base.py#L181-L185

Solution could be to put index on denormalized namespace string (e.g. "users/bob/preferences") instead.