// the take
coreMQTT is AWS's low-footprint MQTT 5.0 client for embedded C, the same library underpinning the AWS IoT Device SDK and FreeRTOS's own MQTT demos. It's for firmware engineers who need a protocol engine they can drop onto a TCP/TLS stack they already control, not a batteries-included networking stack.
Formal verification via CBMC proofs on the packet (de)serialization paths is rare for a C library and specifically targets the buffer-overflow bug class that bites hand-rolled MQTT parsers on embedded targets. MISRA C:2012 compliance with documented deviations plus Coverity static analysis shows real process discipline, not just a README claim. Zero dependencies beyond the standard library and a transport interface you implement keeps the memory footprint predictable and auditable. It's genuinely production-proven: this is the exact code AWS ships in its IoT SDK and the FreeRTOS+TCP/mbedTLS reference demos, not a side project.
No built-in thread safety: a single MQTTContext_t can't be touched from two tasks without you wiring up MQTT_PRE_STATE_UPDATE_HOOK/POST yourself and getting the locking right, which is an easy place to introduce a race if you're not careful. You write the transport layer yourself - there's no TCP or TLS included, so there's real integration work (sockets, mbedTLS/OpenSSL glue) before a single CONNECT packet goes out. Versioning is oddly split: main doesn't map to one spec version - MQTT 3.1.1 support lives on an old v2.3.1 tag and even MQTT 5.0 usage guidance points at the v5.0.2 tag rather than main, so it's unclear what tracking main actually gets you. Build setup requires either a custom core_mqtt_config.h or a disable macro before anything compiles, which is one more hoop for a first integration.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →