THE ONE WHO READS THE WHOLE RECORD
Marshall MQv0.4.0A native macOS client for Kafka and RabbitMQ — open it, see what's in your queues, find the message that broke, and never be told a zero that was really an "I don't know".
Fresh from the lab: what’s new in v0.4.0
He read every message. Twice. He has notes.
i made you a chart. i make a lot of charts.
01 / MORE THAN A PRETTY ICON
7 ways to make your day easier.
Yes, we got a little carried away.
Marshall compares the shape of every message he reads and names the ones that disagree — “totalCents is a string in 2 of these, a number in the other 12” — on the exact line. No model, no schema registry, no network: it is in the data, and it is usually why your dead-letter queue filled up.
Where a topic's messages actually sit across its partitions: the one holding 90% of the traffic, and the one holding nothing while its neighbours are full. A hot partition looks like a slow consumer from the outside; this is what it looks like from the inside.
On Kafka, reading is a seek: no consumer group joined, no offset committed, no topic created by accident. Point him at production and nothing notices.
Go on. Open one.
There is no way to look at a RabbitMQ message without consuming it, and even putting it straight back flips its redelivered flag for good. So reading there is never automatic — he asks first, once per broker, in plain words.
DLQs are recognised and sorted to the top, and the moment a broker answers he reports what he found — because a queue that filled up overnight is the reason you opened the app.
A consumer group that never committed has unknown lag. Fourteen messages across three partitions cannot carry a proportion. Marshall says “can't see” and “too few to judge” instead of a comforting number he can't back up.
He reads a sample and explains it in plain words — through the AI CLI already signed in on your Mac (Claude Code or Codex), so there is no API key and no tokens to buy.
A closer look at what makes Marshall MQ tick.
Marshall MQ is a native macOS client for Kafka and RabbitMQ. Topics with the depth they actually retain, consumer lag per group, dead-letter queues surfaced first, and every message readable with its key, headers and payload. He reads and never disturbs: on Kafka he never joins a consumer group, never commits an offset, and never creates a topic by asking about one — and on RabbitMQ, where looking at a message does leave a mark, he tells you before, not after. He compares the shape of every message he reads and names the ones that disagree, draws a chart of where your messages actually sit across the partitions, reads Avro through your Schema Registry, opens a consumer group with the verdict before the numbers — and when he can't see something, he says so: a group that never committed has unknown lag, not zero.
02 / LET’S MAKE THIS OFFICIAL
A few practical details.
No awkward first-date questions.
THE COMPATIBILITY CHECK
For macOS.
Made to feel right at home.
YOUR NEXT FAVORITE TOOL IS RIGHT HERE.
Free, native, and he never touches what he reads. Meet the rest of the family too.
THERE’S MORE WHERE THAT CAME FROM
Your AI agent, living in your menu bar.
Your servers. Your cloud. Zero complexity.
Your whole restaurant. One happy place.
One beautiful client for every database.
Kubernetes, without the 500 MB suit.
The smart notch for macOS.
Memory that remembers what matters.