: The coreMQTT agent described on this page is functional but not yet part of the coreMQTT library or in the main FreeRTOS.org download because it does not yet meet our code quality criteria and hasn't yet been fully tested. Source code is available on GitHub
MQTT Agent and Demos using coreMQTT
Including Over the Air (OTA) updates
coreMQTT is an MIT licensed open source C MQTT client library for microcontroller and small microprocessor based IoT devices. Its design is intentionally simple to ensure it has no dependency on any other library or operating system, and to better enable static analysis including memory safety proofs. That simplicity and lack of operating system dependency (coreMQTT does not require multithreading at all) means coreMQTT does not build thread safety directly into its implementation. Instead thread safety must be provided by higher level software. This webpage demonstrates a coreMQTT extension that provides that higher level functionality in the form of an MQTT agent (or MQTT daemon). While the implementation demonstrated here is currently specific to FreeRTOS, there are not many dependencies on FreeRTOS, meaning the implementation can easily be adapted for use with other operating systems.
Implementation overview and usage model
The MQTT agent is an independent task (or thread of execution). It achieves thread safety very simply be being the only task that is permitted to access the MQTT library's API. Isolating all MQTT API calls to a single task serialises access, removing the need for semaphores or any other synchronisation primitives.
When using the agent, if an application task wants to perform an MQTT operation, for example publishing a message, it calls the MQTT agent's MQTTAgent_Publish() API instead of calling coreMQTT's MQTT_Publish() API. MQTTAgent_Publish() packages the information required to complete the Publish operation into a structure, then sends that structure over a queue to the MQTT agent task. The MQTT agent task receives the structure, then it calls the underlying MQTT library's MQTT_Publish() API on behalf of the application.
Each API that sends a command to the MQTT agent allows the application writer to optionally specify a callback function, and a parameter to be passed into the callback (called the callback context), for the agent to execute when the resultant MQTT operation completes. That way, the application task can opt to enter the Blocked state (so not consuming any CPU time) to wait for the callback's execution, or alternatively continue executing while the MQTT operation is in process. See the example at the end of this page.
Each API that sends a command to the MQTT agent also allows the application writer to specify the maximum time the calling task should wait in the Blocked state for space to become available in the queue used to send commands to the MQTT agent, should the queue be full at the time of the MQTT agent API call. Again, see the example at the end of this page.
The MQTT agent and associated
example project are functional but not yet complete. Be aware the agent does not
yet comply with our code quality standards and is not yet fully tested. The APIs are unlikely to change before its official first release.
initialises the TCP/IP stack before starting the FreeRTOS kernel scheduler.
When the network connects, the network event hook creates a single RTOS task. That
task optionally creates multiple other tasks, each of which connects to the
MQTT broker before sending and receiving MQTT packets of various different size
and at various different Quality of Service (QoS) levels. The original thread
then becomes the MQTT agent thread.
constants define the created demo tasks. The links in the descriptions go to comments that provide more information at the top of each implementing source file.
#define democonfigCREATE_LARGE_MESSAGE_SUB_PUB_TASK [1 or 0]
Set to 1 to create the task that runs the MQTT demo implemented in
large_message_sub_pub_demo.c, or 0 to omit that task from the build.
#define democonfigNUM_SIMPLE_SUB_PUB_TASKS_TO_CREATE [n]
Sets the number of instances to create of the task implemented in simple_pub_sub_demo.c, which can be 0.
#define democonfigCREATE_CODE_SIGNING_OTA_DEMO [1 or 0]
Set to 1 to include Over the Air (OTA) update functionality, or 0 to omit OTA. There is a separate page that describes using OTA with this demo.
#define democonfigCREATE_DEFENDER_DEMO [1 or 0]
Set to 1 to build the AWS Device Defender demo, or 0 to omit. The instructions for the Device Defender demo can be found here.
Obtaining the source code
in the FreeRTOS GitHub account
demonstrates the use of an agent on top of coreMQTT. It also demonstrates how
to submodule other FreeRTOS libraries into a project. Do not use the "Download Zip"
link in Gitub to obtain the code as the zip file will not include the submoduled
libraries. Instead use one of the following Git commands to clone the repo and
its submodules onto your local machine:
To clone using HTTPS:
git clone https://github.com/FreeRTOS/Lab-Project-coreMQTT-Agent.git --recurse-submodules
git clone email@example.com:FreeRTOS/Lab-Project-coreMQTT-Agent.git --recurse-submodules
If you have downloaded the repo without using the --recurse-submodules argument, you need to run:
git submodule update --init --recursive
At the time of writing the project uses the
FreeRTOS Windows port, the FreeRTOS+TCP
TCP/IP stack, and builds using the free
Community Edition of Visual Studio.
Its directory structure is however structured to enable the addition of projects for
other development tools soon.
Source code organisation
The git repo is organised as follows:
| +-VisualStudio Contains the Visual Studio project for this demo
| +-AWS Contains a sub-module for the OTA library along with an OTA PAL port for Windows.
| +-FreeRTOS Contains sub-modules of the FreeRTOS libraries used by the demo
| +-ThirdParty Contains submodules of third party libraries used by the demo
+-configuration-files Contains configuration files for the demo and libraries
+- . Contains source files that implement the various demos
The demo uses the FreeRTOS+TCP TCP/IP stack, so follow the instructions provided for the TCP/IP starter project to ensure you:
- Have the pre-requisite components installed (such as WinPCap).
- Optionally set a static or dynamic IP address, gateway address and netmask.
- Optionally set a MAC address.
- Select an Ethernet network interface on your host machine.
- …and importantly test your network connection before attempting to run the MQTT demo.
Each demo project has its own configuration settings. When you are following the network
configuration instructions, make sure to apply the settings in the MQTT demo project, rather than
the TCP/IP starter project. By default the TCP/IP stack is configured to use a dynamic IP address.
Configuring the MQTT broker connection
The MQTT agent can connect to any MQTT broker, either locally or remotely, and either using TLS or in plain text. The Configuring an MQTT Broker section on the non-agent plain text demo documentation page
details some local and remote plain text MQTT broker options. The Configuration an MQTT Broker section on the server authentication
and mutual authentication
pages detail some TLS encrypted MQTT broker options.
Additional notes for AWS IoT users: If you are going to connect to AWS IoT then you can create the necessary cloud and device side configurations
by following the instructions provided in the Using the AWS IoT Message Broker section
of the page that documents the separate mutual authentication demo. That section references scripts that can be found in the /lib/AWS/tools directory
of the MQTT Agent repository.
Once you have decided on the broker to connect to, the code snippet below the following bullet points describes the compile time configuration constants that
configure the MQTT connection. These constants are in /Source/configuration-files/demo_config.h.
Place any keys required for TLS connections in the same file.
Plain text connections are useful to learn how MQTT works and to debug connections because
the MQTT conversation can be viewed in a network sniffer such as WireShark.
However, real IoT devices should not use an unencrypted plain text connection. Never send private
data over a plain text connection - we highly recommend using encrypted and mutually authenticated connections for all IoT devices.
in a header file is for convenience of demonstration only. We strongly recommend
that real devices store keys in a secure location, such as a secure element or enclave.
Further, we recommend real IoT devices access keys and other crypto objects via an API
that does not expose the keys, such as PKCS #11
or PSA APIs.
#define democonfigCLIENT_IDENTIFIER "Thing1"
#define democonfigMQTT_BROKER_ENDPOINT "192.168.0.100"
#define democonfigMQTT_BROKER_PORT ( 8883 )
#define democonfigUSE_TLS 1
MQTT agent configuration constants
The following code snippet shows the compile time constants relevant to the MQTT
agent, along with their default values should they be left undefined.
The MQTT agent does not have its own configuration file, but it will use any macros defined in core_mqtt_config.h
. Macros defined there will be used in place of these defaults.
#define MQTT_AGENT_MAX_OUTSTANDING_ACKS ( 20 )
#define MQTT_AGENT_MAX_EVENT_QUEUE_WAIT_TIME ( 1000 )
Example of using the MQTT agent API
Important notes on the following code:
The MQTT PUBLISH payload and topic string must remain valid until the
MQTT PUBLISH has been acknowledged by the MQTT broker. In the example
below the topic string is a static const, so it will always be valid
In this example the task that calls
MQTTAgent_Publish() waits to be notified that the MQTT PUBLISH command has been acknowledged by the
server. It is optional for the task to pass the completion callback which executes upon receipt of the acknowledgment, but it is useful
as it also signals when the buffers used to hold the MQTT publish information can be reused.
static const char * const pcTopicName = "/my/topic/x";
static MQTTAgentContext_t xAgentContext;
static void prvCommandCallback( void *pxCommandContext,
MQTTStatus_t xReturnStatus )
CommandContext_t *pxApplicationDefinedContext = ( CommandContext_t * ) pxCommandContext;
pxApplicationDefinedContext->xReturnStatus = xReturnStatus;
void ExampleOfCallingAgentPublish( char *pcPayload, uint16_t usPayloadLength )
xPublishInfo.qos = MQTTQoS1;
xPublishInfo.pTopicName = pcTopicName;
xPublishInfo.topicNameLength = ( uint16_t ) strlen( pcTopicName );
xPublishInfo.pPayload = pcPayload;
xPublishInfo.payloadLength = usPayloadLength;
xCommandContext.xTaskToNotify = xTaskGetCurrentTaskHandle();
xCommandInformation.cmdCompleteCallback = prvCommandCallback;
xCommandInformation.pCmdCompleteCallbackContext = &xCommandContext;
xCommandInformation.blockTimeMs = mqttexampleMAX_COMMAND_SEND_BLOCK_TIME_MS;
xCommandAdded = MQTTAgent_Publish(
if( xCommandAdded == MQTTSuccess )
xReturn = xTaskNotifyWait( 0,
if( xReturn != pdFAIL )
MQTT agent API
At this time the MQTT agent API is documented in the mqtt_agent.h
header file and at the MQTT agent API References
Copyright (C) Amazon Web Services, Inc. or its affiliates. All rights reserved.