Beginners Guide To The MQTT Protocol

mqtt-iconWhat is MQTT?

MQTT is a lightweight publish/subscribe messaging protocol designed for M2M (machine to machine) telemetry in low bandwidth environments.

It was designed by Andy Stanford-Clark (IBM) and Arlen Nipper in 1999 for connecting Oil Pipeline telemetry systems over satellite.

Although it started as a proprietary protocol it was released Royalty free in 2010 and became an OASIS standard in 2014.

MQTT stands for MQ Telemetry Transport but previously was known as Message Queuing Telemetry Transport.

MQTT is fast becoming one of the main protocols for IOT (internet of things) deployments.

MQTT Versions

There are two versions of MQTT.

The original MQTT which was designed in 1999 and has been in use for many years and designed for TCP/IP networks.

Here is the actual Specification MQTT V3.1 and here is a more detailed overview of the MQTT protocol packet structure,.

The latest MQTT version(v 5) ,has now been approved (Jan 2018).. You can download the specification here.

Currently there is little support for version 5 but that should change in 2018.If you are wondering what happened to 4 then see here.

An overview of the key ideas in version 5 is here.

MQTT-SN which was specified in around 2013, and designed to work over UDP, ZigBee and other transports.

MQTT-SN doesn’t currently appear to be very popular. and the specification hasn’t changed for several years, but I expect that to change as IOT deployments start. See MQTT-SN working Notes.

Understanding MQTT-How MQTT Works

MQTT uses a publish /subscribe model which requires the use of a central Broker as shown in the diagram below:

MQTT- Publish-Subscribe-Model

Clients do not have addresses like in email systems, and messages are not sent to clients.

Instead messages are published to a broker on a topic.

The job of an MQTT broker is to filter messages based on topic, and then distribute them to subscribers..

A client can receive these messages by subscribing to that topic on the same broker

In this model there is no direct connection between a publisher and subscriber.

The model is similar to broadcast radio and TV were a TV or radio station broadcasts on a channel and listeners/viewers tune in to that channel.

However unlike the broadcast TV/radio model in MQTT all clients can publish (broadcast) and subscribe (receive).

What are Topics?

These are like channels in the TV radio model.

The publisher broadcasts on a particular channel (topic) and a listener (subscriber) tunes into that channel (subscribes) to receive the broadcasts.

What Happens to Published Messages?

Questions –

1.What happens to the published message after the subscriber receives it?

2. What happens to the published message if their are no subscribers?

To answer these questions just think of a TV or radio broadcast.

If you are tuned into the broadcast you simply miss it!

So for question1 and question 2 the answer is- the message is deleted from the broker.

When a client publishes a message on a topic then the broker will distribute that message to any connected clients that have subscribed to that topic.

Once the message has been sent to those clients it is removed from the broker (see note below).

If no clients have subscribed to the topic or they aren’t currently connected then the message is removed from the broker. (see note)

In general the broker doesn’t store messages.

Note: Retained messages, persistent connections and QOS levels can result in messages being stored on the broker/server.

MQTT Client-Broker Connections

MQTT uses TCP/IP to connect to the broker. TCP is a connection orientated protocol with error correction and guarantees that packets are received in order.

You can consider a TCP/IP connection to be similar to a telephone connection.

Once a telephone connection is established you can talk over it until one party hangs up.

Most MQTT clients will connect to the broker and remain connected even if they aren’t sending data.

MQTT clients publish a keepalive message at regular intervals (usually 60 seconds) which tells the broker that the client is still connected.

The Client name

All clients are required to have a client name.

The client name is used by the MQTT broker to track subscriptions etc.

Client names must also be unique.

If you attempt to connect to an MQTT broker with the same name as an existing client then the existing client connection is dropped.

Because most MQTT clients will attempt to reconnect following a disconnect this can result in a loop of disconnect and connect.

The screen shot below show what happens when I try and connect a client to the broker using the same id as an existing client.


Clean Sessions

MQTT clients will usually by default establish a clean session with a broker.

A clean session is one in which the broker isn’t expected to remember anything about the client when it disconnects.

With a non clean session the broker will remember client subscriptions and may hold undelivered messages for the client.

However this depends on the Quality of service used when subscribing to topics and the quality of service used when publishing topics.

Understanding Clean sessions Video

Quality of Service (QOS)

MQTT support 3 QOS levels 0,1,2. QOS is set on published messages and on topic subscriptions.

  • QOS -0 Default and doesn’t guarantee message delivery.
  • QOS -1 – Guarantees message delivery but could get duplicates.
  • QOS -2 -Guarantees message delivery with no duplicates

Working out the overall QOS of a message is covered in other more in depth tutorials.

QOS level 2 involves more message overhead that QOS 1 which itself requires more messages than QOS 0.

MQTT Topic Structure

In MQTT there is no formal structure and a publisher is free to choose it’s own topic names and structure.

There are however naming rules:


Taken from the Specification MQTT V3.1

Topic names look very similar to URLs.

Topics are created by the subscriber and publisher and are not pre-assigned by the broker.

As long as a topic name follows the naming standards it is accepted by the Broker.

A Topic name is a UTF-8 string, and must comprise of at least one character.

Topics can be created in a hierarchical manner (levels) using a forward slash as a delimiter.

Simple topic examples are:

topic = bulb1, topic = bulb2, topic = bulb3 or

topic =bulbs/ bulb1, topic =bulbs/ bulb2, topic =bulbs/ bulb3

Topic names are not permanent as they only exist on a broker when someone has subscribed to that topic and is connected, or subscribed with the clean session flag set to False.

The $SYS topic hierarchy is used by most brokers for publishing broker statistics. However it isn’t mandatory.

Publishing Messages and The Retain Flag

When a client publishes a message to a broker it needs to send:

  • The message topic
  • The message QOS
  • Whether the message should be retained.- Retain Flag

The retain Flag is normally set to False which means that the broker doesn’t keep the message.

If you set the retain flag to True then the last message received by the broker with the retained flag set will be kept.

The QOS of the published message has no effect on the retained message.

The main use of this is for sensors that don’t change very much and publish their status infrequently.

If you have a door sensor, for example, then it doesn’t make much sense publishing it’s status every second when it is almost always the same.

However if it only publishes it’s status when it changes what happens when a subscriber subscribes to the sensor.

In this case if the last status was published without the retain flag set then the subscriber wouldn’t know the status of the sensor until it published it again.

Last Will Messages

The idea of the last will message is to notify a subscriber that the publisher is unavailable due to network outage.

The last will message is set by the publishing client on a topic.

The message is stored on the broker and sent to any subscribing client if the connection to the publisher fails.

If the publisher disconnects normally the last Will Message is not sent.

MQTT Clients

Because MQTT clients don’t have addresses like email addresses, phone numbers etc. you don’t need to assign addresses to clients like you do with most messaging systems.

There is client software available in almost all programming languages and for the main operating systems Linux, Windows, Mac from the Eclipse Paho project.

On this site I will be using the Python client.

MQTT Brokers or Servers

Note: The original term was broker but it has now been standardized as Server. You will see Both terms used.

There are many MQTT brokers available that you can use for testing and for real applications.

There are free self hosted brokers , the most popular being Mosquitto and commercial ones like HiveMQ.

Mosquitto is a free open source MQTT broker that runs on Windows and Linux.

If you don’t want to install and manage your own broker you can use a cloud based broker from Cloud service providers like IBM, Microsoft (Azure) etc

Eclipse has a free public MQTT broker and COAP server that you can also use for testing. The address is and the port is 1883 or 8883(SSL).

See the MQTT Brokers and Servers article for a list of cloud hosting options

MQTT Security

MQTT supports various authentications and data security mechanisms.

It is important to note that these security mechanisms are configured on the MQTT broker, and it is up to the client to comply with the mechanisms in place.

See An Introduction to MQTT security mechanisms

MQTT Overview Video

MQTT Over WebSockets

Websockets allows you to receive MQTT data directly into a web browser.

This is important as the web browser may become the de-facto interface for displaying MQTT data.

MQTT websocket support for web browsers is provided by the Javascript client.

See –Using MQTT Over WebSockets with Mosquitto

Common Questions

If you are familiar with the web and email then you will probably find, as I did, that MQTT is very different. These are some of the questions I had, and saw on other sites and forums that may clear things up a little.

Q- What happens to messages that get published to topics that no one subscribes to?

A- They are discarded by the broker.

Q-How can I find out what topics have been published?

A- You can’t do this easily as the broker doesn’t seem to keep a list of published topics as they aren’t permanent.

Q- Can I subscribe to a topic that no one is publishing to?

A- Yes

Q- Are messages stored on the broker?

A- Yes but only temporarily. Once they have been sent to all subscribers they are then discarded. But see next question.

Q- What are retained messages?

A- When you publish a message you can have the broker store the last published message. This message will be the first message that new subscribers see when they subscribe to that topic. MQTT only retains 1 message.

Real World MQTT Example Deployments

It’s often useful and interesting to see how a particular technology is actually being used. Here are some examples I’ve come across:

Learning MQTT-

Fortunately when I started to learn about MQTT I came across the very good MQTT essentials series that explained the main features of MQTT.

However there is one thing reading about something and another seeing it really work so I started using Python to create simple scripts to demonstrate the main aspects of MQTT.

I use the Paho Python MQTT client for the client scripts.
However A knowledge of python isn’t really required to understand the examples.

I also used my own local Mosquitto MQTT broker but there are many free online brokers that you can use instead.

MQTT Tutorials By Subject

MQTT Protocol

MQTT Connections



Publishing Messages


Test your MQTT knowledge with the MQTT basics quiz


Please rate? And use Comments to let me know more
[Total: 41    Average: 4/5]


  1. Dear writer,
    I am trying to implement MQTT on .NETCore for which there are no library or nuget packages . In my case I want to implement some very basic functionalities of MQTT – like CONNECT, PUBLISH, SUBSCRIBE and DISCONNECT. Is it possible to write all the class libraries by myself?? oI am not so good programmer. 🙂 Thank you ..

  2. hi!
    first of all, thank you! it was very helpful 🙂
    but i don’t understand one thing; is there any standard topic for mqtt? i mean, is it necessery to describe the topic names on the client? because there might be really huge number of topics for a big project. then it would be hard to define all of them.
    thank you 🙂

    1. Hi
      There is no standard like DNS. The only reserved name is $SYS.My personal view is that embedding a topic hierarchy into a sensor/client would not be a good idea. Here are some working notes I’ve put together that you may find useful.
      I think it is still early days yet and various best practices will emerge.

  3. I’m using Paho client and Mosquitto, all on Raspberry Pi and Python.
    I’m seeking better information on building not just working applications, but robust applications.

    There are issue such as “Don’t disconnect until all messages have published”, “Wait for Connect”, “Publish on Paho Python does not block”, among others, that need to be addressed. Every example I’ve seen omits these discussions.

    I’d love to hear from anyone who is working on creating more robust MQTT apps beyond just “Hello World”

    1. jeffrey
      You might want to take a look at client connections .
      I use a wait loop after a connection request and a subscribe and don’t proceed until I get a CONNACK or SUBACK. I have a general function to do this.
      def wait_for(client,msgType,period=0.25):
      if msgType==”SUBACK”:
      if client.on_subscribe:
      while not client.suback_flag:“waiting suback”)
      client.loop() #check for messages

      In the program I just call it wait_for(client,”CONNACK”). The script stops until I get the connack. I’ve also started to do a timeout in the wait loop.
      The flags I use I set in the client class at the beginning of the script.So my import is this
      import paho.mqtt.client as mqtt
      and then I set the flag like this
      mqtt.Client.suback_flag= False
      So I can create new clients and they will all have the flag available e.g.
      client= mqtt.Client(cname)
      Because I use several flags I use a function to set them all. It looks like this:
      def Initialise_client_object():
      mqtt.Client.suback_flag= False

      I hope that helps.
      Rgds Steve

  4. How the broker registers a topic? does the broker perform that at the subscribing, publishing or both phases? Does the broker store them in one list? if so, where is that or what is the name of the list?

    1. Sami
      You would need to dig into the source code to find that out. I assume,like you, that stores them in a list in memory.
      The broker registers a topic as a result of a client subscribe.It must also regsiter a topic if you Publish to a topic with the retain flag set to true.
      I would also expect it to register a topic if you set the last will message.

  5. Hi Steve,
    Thank for the great tutorial. I have a question regarding the MQTT setup for a local experiment. I’m trying to do a research experiment that simulates several MQTT that connected together. Can I use a single computer to run several MQTT in one machine? if not what do you suggest and what is the initial direction that I should take?

    Thank again.

    1. Yes you can run lots of brokers on the same computer by using different ports. This is the way online brokers work.To do that you need to add a new listener in mosquitto.conf
      listener port 1884
      listener port 1885


      You could also use the free online brokers as extra brokers.

      “”, “”,””

      1. Great!
        I noticed in your tutorials that you are using your own IP address for the hostname. Do you have a tutorial on that or a specific way of doing it?

      2. You state that using different ports create different brokers. When I think different brokers I think different message queues. Different ports do not appear to separate messages such that messages published to port 1883 cannot be viewed on port 1885. How do different ports create different brokers?

  6. Hi,
    I’m trying to send a publish command with a “retain” flag via WemosESP8266 which uses Arduino.ide. It doesn’t seem to compile

    client.publish(“Topic/Subtopic”, “msg” -r);
    Error msg:
    Arduino: 1.6.5 (Windows 7), Board: “WeMos D1 R2 & mini, 160 MHz, 921600, 4M (3M SPIFFS)”

    Notifier.ino: In function ‘void loop()’:
    ‘r’ was not declared in this scope

    1. Hi
      I’m not familiar with Arduino but assuming you are running a Python client then the publish command looks wrong.
      Try this
      client.publish(topic,”test message”,0,True)

  7. Hi Steve,

    how to run multiple MQTT brokers on the same machine (e.g., laptop) for conducting an experiment. I downloaded the MQTT files from here ( using pip. I then run the example (by using Pycharm editor) which requires using the testing server that I do not want to use. I want to use anything can be local to simulate my experiment. I do not want to use any remote server. would that possible? Thank.

    1. Sam
      Yes you can run as many instances as you want by creating extra listeners. Just add the following lines to the mosquitto.conf file for each extra listener. There is a section of the file called extra listeners. Lines are

      listener 1884
      protocol mqtt

      creates an instance running on port 1884 as well as the default listener on 1883

  8. hi Steve
    I follow all series about mqtt then thanks for your help
    but I wanna ask u something when I create my own broker is it necessary to need other online broker lie hiveQ …

    1. No if all you are doing at the moment is testing then your own broker is the better option and you don’t need an online one.

  9. Hi Stev,

    I have a problem when I created two local brokers with different ports on my laptop. When I publish a message to any broker, the message is delivered to all subscribers in both brokers. What is the problem? and how to separate both brokers to allow them work independently?


  10. Hello Steve… is it possible to encrypt the information flowing from/to broker and client using a lightweight algorithm? I want to embed a lightweight algorithm in MQTT to be implemented in broker and client to address security issues. Is this doable?

    1. Take a look at the data logging tutorial here-
      I will also look at using sqllite database shortly if that is what you want

  11. Hello steve
    I’m using emqtt instead of mosquito as a mqtt broker on a remote server. I’ve hosted my server on dev-cloud. When I’m establishing server on my own device i.e. a local server the broker works fine but when I try to use it on the remote server it doesn’t connect. i’m testing by using MQTTlens. Please help!

    1. After email exchange the problem turned out to be firewall related. If you configure MQTT on devcloud pay attention to the firewall settings

  12. Dear Steve

    Great tutorial, I got all excited and tried to install the latest Mosquito 1.4.14 on my Win7 machine. Been struggling for 2hrs. The installer is a joke and I would describe it more as a loose guide. However even after gathering all the missing DLL’s from hither and dither it still won’t run. When I run the mosquito.exe I get an error saying it was unable to start correctly 0xC000 007b.

    Have you been able to install and run the latest mosquito release ?

    I had to use an older OpenSSL light to get the correct DLL’s, I used 1.0.2m (1.1.0g was missing ssleay32.dll) , I also had to hunt for VCRUNTIME140.dll on another m/c.

  13. I have resolved the installation issues after opening a ticket here :

    Basically you need to use OpenSSL Light 1.0.2m to source the DLL’s not the latest 1.1.0g and you have to get hold of the 32bit version of VCRUNTIME140.dll and then you need to re-run it at-least twice. Once to install the files, then a second time after manually providing the missing DLL’s so mosquitto gets installed as a service.

    The current mosquitto installer leaves you in the dark on these requirements, so hopefully it will now be updated to give better instructions on how to use it successfully.

    1. Neil
      Tks for that. Just want to point out that the download will still work and that this is only necessary if you really want to install the latest version 1.4.14. The download has version 1.4.09. If you are using the broker for testing then I don’t think that there are no major differences.

    1. Hi
      I assume you mean when using websoclets. In which case there isn’t unless you send that information using mQTT. It is not sent by default as is the case when using a http client connection.

  14. Thank you for top topic
    I have question my friend
    For sample,if i send message for you and you are offline,when you connecting to broker,do you can receive last message?
    Look like chat or push notification that save message and resend after reconnecting user
    And how many broker can save and resend their again?

  15. hi,
    I am using MQTT protocol in real time data communication application,therefore I need to retrieve the client device location data and pass it to MQTT broker. Is there any available API to track client device Geo location in MQTT environment?

Leave a Reply

Your email address will not be published. Required fields are marked *