MSK Connect runs Kafka Connect 129 workers for you (Kafka Connect) in distributed mode, against an MSK cluster or any Kafka 129 cluster your VPC can reach. It runs the open-source Connect framework at version 2.7.1 or 3.7.x, so connectors built for newer Connect APIs may need checking. You upload a custom plugin (the connector's JARs or ZIP, such as Debezium 317,608 's PostgreSQL 1,289 connector from CDC with Debezium 3.x) to S3, create a worker configuration (the default uses StringConverter for keys and values), and create a connector with its usual properties and a capacity: a fixed number of workers, or autoscaling between a minimum and a maximum on CPU use.
Capacity is counted in MSK Connect Units (MCUs), each one vCPU and 4 GB of memory, and billed per second. AWS 24 's US East example charges $0.11 per MCU-hour: an S3 sink that autoscales between two and four 1-MCU workers over a working day uses 1,984 MCU-hours in a month, $218.24. BookNest's JDBC sink, one worker of one MCU around the clock, would cost about $80 a month. A connector may have at most 10 workers.
What you lose compared with running Connect yourself is the REST API: connectors are created and changed through the MSK Connect API, console or infrastructure-as-code tools, not by curl 3,008 to port 8083. Logs go to CloudWatch, S3 or Firehose. For landing topics in S3, newer MSK data delivery on Express brokers writes topics straight to S3 buckets or as Iceberg 129 tables in S3 Tables, billed per GB delivered ($10 per TB in AWS's Iceberg example), with no connector at all (compare Streaming into the Lakehouse).