MongoDB has had its fair share of criticism throughout the years. The web-scale proposition has been met with skepticism by many developers. The counter argument is that scale is not needed most of the time and we should focus on other design considerations. While this may be true on several occasions, it's a false dichotomy and in an ideal world we would have both. MongoDB is as close as it can get to combining scalability with features and ease of use/time to market.
MongoDB's schema-less nature is also a big point of debate and argument. Schema-less can be really beneficial in many use cases as it allows for heterogeneous data to be dumped into the database without complex cleansing or ending up with lots of empty columns or blocks of text stuffed into a single column. On the other hand, this is a double-edged sword as a developer may end up with many documents in a collection that have loose semantics in their fields and it becomes really hard to extract these semantics at the code level. What we can have in the end if schema design is not optimal, is a plain datastore rather than a database.
Lack of proper ACID guarantees is a recurring complaint from the relational world. Indeed, if a developer needs access to more than one document at a time it's not easy to guarantee RDBMS properties as there are no transactions. Having no transactions in the RDBMS sense also means that complex writes will need to have application level logic to rollback. If you need to update three documents in two collections to mark an application level transaction complete and the third document doesn't get updated for whatever reason, the application will need to undo the previous two writes, something that may not be exactly trivial.
Defaults that favored setting up MongoDB but not operating it in a production environment are also frowned upon. For years, the default write behavior was write and forget, sending a write wouldn't wait for an acknowledgement before attempting the next write, resulting in insane write speeds with poor behavior in case of failure. Authentication is also an afterthought, leaving thousands of MongoDB databases in the public internet prey to whoever wants to read the stored data. Even though these were conscious design decisions, they are decisions that have affected developers' perception of MongoDB.
There are of course good points to be made from criticism. There are use cases where a non relational, unsupporting transactions database will not be a good choice. Any application that depends on transactions and places ACID properties higher than anything else is probably a great use case for a traditional RDBMS but not for a NoSQL database.