Writing a new integration

Integrations are one of the most important parts of a group chat tool like Zulip, and we are committed to making integrating with Zulip and getting you integration merged upstream so everyone else can benefit from it as easy as possible while maintaining the high quality of the Zulip integrations library.

On this page you’ll find:

Contributions to this guide are very welcome, so if you run into any issues following these instructions or come up with any tips or tools that help writing integration, please email zulip-devel@googlegroups.com, open an issue, or submit a pull request to share your ideas!

Types of integrations

We have several different ways that we integrate with 3rd party products, ordered here by which types we prefer to write:

  1. Webhook integrations (examples: Freshdesk, GitHub), where the third-party service supports posting content to a particular URI on our site with data about the event. For these, you usually just need to create a new python package in the zerver/webhooks/ directory. You can easily find recent commits adding new integrations to crib from via git log zerver/webhooks/.
  2. Python script integrations (examples: SVN, Git), where we can get the service to call our integration (by shelling out or otherwise), passing in the required data. Our preferred model for these is to ship these integrations in our API release tarballs (by writing the integration in api/integrations).
  3. Plugin integrations (examples: Jenkins, Hubot, Trac) where the user needs to install a plugin into their existing software. These are often more work, but for some products are the only way to integrate with the product at all.

General advice

  • Consider using our Zulip markup to make the output from your integration especially attractive or useful (e.g. emoji, markdown emphasis, @-mentions, or !avatar(email)).
  • Use topics effectively to ensure sequential messages about the same thing are threaded together; this makes for much better consumption by users. E.g. for a bug tracker integration, put the bug number in the topic for all messages; for an integration like Nagios, put the service in the topic.
  • Integrations that don’t match a team’s workflow can often be uselessly spammy. Give careful thought to providing options for triggering Zulip messages only for certain message types, certain projects, or sending different messages to different streams/topics, to make it easy for teams to configure the integration to support their workflow.
  • Consistently capitalize the name of the integration in the documentation and the Client name the way the vendor does. It’s OK to use all-lower-case in the implementation.
  • Sometimes it can be helpful to contact the vendor if it appears they don’t have an API or webhook we can use – sometimes the right API is just not properly documented.
  • A helpful tool for testing your integration is UltraHook, which allows you to receive webhook calls via your local Zulip development environment. This enables you to do end-to-end testing with live data from the service you’re integrating and can help you spot why something isn’t working or if the service is using custom HTTP headers.

Webhook integrations

A webhook allows a third-party service to push data to you when something happens. It’s different from making a REST API call, where you send a request to the service’s API and wait for a response. With a webhook, the third-party service sends you an HTTP POST when it has something for you. Your webhook integration defines the URI the service uses to communicate with Zulip, and handles that incoming data.

New Zulip webhook integrations can take just a few hours to write, including tests and documentation, if you use the right process. Here’s how we recommend doing it:

  • First, use http://requestb.in/ or a similar site to capture an example webhook payload from the service you’re integrating. You can use these captured payloads to create a set of test fixtures for your integration under zerver/fixtures.

  • Then write a draft webhook handler under zerver/webhooks/; there are a lot of examples in that directory. We recommend templating off a short one (like stash or zendesk), since the longer ones usually just have more complex parsing which can obscure what’s common to all webhook integrations. In addition to writing the integration itself, you’ll need to create Integration object and add it to WEBHOOK_INTEGRATIONS in zerver/lib/integrations.py; search for webhook in that file to find the existing ones (and please add yours in the alphabetically correct place).

  • Then write a test for your fixture in the tests.py file in the zerver/webhooks/mywebhook directory. You can now iterate on debugging the tests and webhooks handler until they work, all without ever needing to post directly from the service you’re integrating with to your Zulip development machine. You can run just the tests for one integration like this:

    test-backend zerver/webhooks/pagerduty/

    See this guide for more details on the Zulip test runner.

  • Once you’ve gotten your webhook working and passing a test, capture payloads for the other common types of posts the service’s webhook will make, and add tests for them; usually this part of the process is pretty fast. Webhook integration tests should all use fixtures (as opposed to contacting the service), since otherwise the tests can’t run without Internet access and some sort of credentials for the service.

  • Finally, write documentation for the integration; there’s a detailed guide below.

See the Hello World webhook Walkthrough below for a detailed look at how to write a simple webhook.

Files that need to be created

Select a name for your webhook and use it consistently. The examples below are for a webhook named ‘MyWebHook’.

  • static/images/integrations/logos/mywebhook.png: An image to represent your integration in the user interface. Generally this Should be the logo of the platform/server/product you are integrating. See Documenting your integration for details.
  • static/images/integrations/mywebbook/001.png: A screen capture of your integration for use in the user interface. You can add as many images as needed to effectively document your webhook integration. See Documenting your integration for details.
  • zerver/fixtures/mywebhook/mywebhook_messagetype.json: Sample json payload data used by tests. Add one fixture file per type of message supported by your integration. See Testing and writing tests for details.
  • zerver/webhooks/mywebhook/__init__.py: Empty file that is obligatory part of every python package. Remember to git add it.
  • zerver/webhooks/mywebhook/view.py: Includes the main webhook integration function including any needed helper functions.
  • zerver/webhooks/mywebhook/tests.py: Add tests for your webbook. See Testing and writing tests for details.
  • zerver/webhooks/mywebhook/doc.html: Add end-user documentation. See Documenting your integration for details.

Files that need to be updated

  • zerver/lib/integrations.py: Add your integration to WEBHOOK_INTEGRATIONS to register it. This will automatically register a url for the webhook of the form api/v1/external/mywebhook and associate with the function called api_mywebhook_webhook in zerver/webhooks/mywebhook/view.py.

Python script and plugin integrations

For plugin integrations, usually you will need to consult the documentation for the third party software in order to learn how to write the integration. But we have a few notes on how to do these:

  • You should always send messages by POSTing to URLs of the form https://zulip.example.com/v1/messages/.
  • We usually build Python script integration with (at least) 2 files: zulip_foo_config.py containing the configuration for the integration including the bots’ API keys, plus a script that reads from this configuration to actually do the work (that way, it’s possible to update the script without breaking users’ configurations).
  • Be sure to test your integration carefully and document how to install it (see notes on documentation below).
  • You should specify a clear HTTP User-Agent for your integration. The user agent should at a minimum identify the integration and version number, separated by a slash. If possible, you should collect platform information and include that in ()s after the version number. Some examples of ideal UAs are:
ZulipDesktop/0.7.0 (Ubuntu; 14.04)
ZulipJenkins/0.1.0 (Windows; 7.2)
ZulipMobile/0.5.4 (Android; 4.2; maguro)

Documenting your integration

Every Zulip integration must be documented in zerver/webhooks/mywebhook/doc.html. Usually, this involves a few steps:

  • Add text explaining all of the steps required to setup the integration, including what URLs to use, etc. If there are any screens in the product involved, take a few screenshots with the input fields filled out with sample values in order to make the instructions really easy to follow. For the screenshots, use a bot with a name like “GitHub Bot”, an email address for the bot like github-bot@zulip.example.com, and an obviously fake API key like abcdef123456790.

  • Make sure you’ve added your integration to zerver/lib/integrations.py; this results in your integration appearing on the /integrations page. You’ll need to add a logo image for your integration under the static/images/integrations/logos/<name>.png, where <name> is the name of the integration, all in lower case.

  • Finally, generate a message sent by the integration and take a screenshot of the message to provide an example message in the documentation. If your new integration is a webhook integration, you can generate such a message from your test fixtures using send_webhook_fixture_message:

    ./manage.py send_webhook_fixture_message \
         --fixture=zerver/fixtures/pingdom/pingdom_imap_down_to_up.json \

    When generating the screenshot of a sample message, give your test bot a nice name like “GitHub Bot”, use the project’s logo as the bot’s avatar, and take the screenshots showing the stream/topic bar for the message, not just the message body.

When writing documentation for your integration, be sure to use the {{ external_api_uri }} template variable, so that your integration documentation will provide the correct URL for whatever server it is deployed on. If special configuration is required to set the SITE variable, you should document that too, inside an {% if api_site_required %} check.

Hello World webhook Walkthrough

Below explains each part of a simple webhook integration, called Hello World. This webhook sends a “hello” message to the test stream and includes a link to the Wikipedia article of the day, which it formats from json data it receives in the http request.

Use this walkthrough to learn how to write your first webhook integration.

Step 0: Create fixtures

The first step in creating a webhook is to examine the data that the service you want to integrate will be sending to Zulip.

You can use requestb.in or a similar tool to capture webhook payload(s) from the service you are integrating. Examining this data allows you to do two things:

  1. Determine how you will need to structure your webhook code, including what message types your integration should support and how; and,
  2. Create fixtures for your webhook tests.

A test fixture is a small file containing test data, one for each test. Fixtures enable the testing of webhook integration code without the need to actually contact the service being integrated.

Because Hello World is a very simple webhook that does one thing, it requires only one fixture, zerver/fixtures/helloworld/helloworld_hello.json:

  "featured_title":"Marilyn Monroe",

When writing your own webhook integration, you’ll want to write a test function for each distinct message condition your webhook supports. You’ll also need a corresponding fixture for each of these tests. See Step 3: Create tests or Testing for further details.

Step 1: Initialize your webhook python package

In the zerver/webhooks/ directory, create new subdirectory that will contain all of corresponding code. In our example it will be helloworld. The new directory will be a python package, so you have to create an empty __init__.py file in that directory via e.g. touch zerver/webhooks/helloworld/__init__.py.

Step 2: Create main webhook code

The majority of the code for your webhook integration will be in a single python file, zerver/webhooks/mywebhook/view.py.

The Hello World integration is in zerver/webhooks/helloworld/view.py:

from __future__ import absolute_import
from django.utils.translation import ugettext as _
from zerver.lib.actions import check_send_message
from zerver.lib.response import json_success, json_error
from zerver.decorator import REQ, has_request_variables, api_key_only_webhook_view
from zerver.lib.validator import check_dict, check_string

from zerver.models import Client, UserProfile

from django.http import HttpRequest, HttpResponse
from typing import Dict, Any, Iterable, Optional, Text

def api_helloworld_webhook(request, user_profile, client,
                           topic=REQ(default='Hello World')):
    # type: (HttpRequest, UserProfile, Client, Dict[str, Iterable[Dict[str, Any]]], Text, Optional[Text]) -> HttpResponse

  # construct the body of the message
  body = 'Hello! I am happy to be here! :smile:'

  # try to add the Wikipedia article of the day
  # return appropriate error if not successful
      body_template = '\nThe Wikipedia featured article for today is **[{featured_title}]({featured_url})**'
      body += body_template.format(**payload)
  except KeyError as e:
      return json_error(_("Missing key {} in JSON").format(str(e)))

  # send the message
  check_send_message(user_profile, client, 'stream', [stream], topic, body)

  # return json result
  return json_success()

The above code imports the required functions and defines the main webhook function api_helloworld_webhook, decorating it with api_key_only_webhook_view and has_request_variables.

You must pass the name of your webhook to the api_key_only_webhook_view decorator. Here we have used HelloWorld. To be consistent with Zulip code style, use the name of the product you are integrating in camel case, spelled as the product spells its own name (except always first letter upper-case).

You should name your webhook function as such api_webhookname_webhook where webhookname is the name of your webhook and is always lower-case.

At minimum, the webhook function must accept request (Django HttpRequest object), user_profile (Zulip’s user object), and client (Zulip’s analogue of UserAgent). You may also want to define additional parameters using the REQ object.

In the example above, we have defined payload which is populated from the body of the http request, stream with a default of test (available by default in the Zulip development environment), and topic with a default of Hello World.

The line that begins # type is a mypy type annotation. See this page for details about how to properly annotate your webhook functions.

In the body of the function we define the body of the message as Hello! I am happy to be here! :smile:. The :smile: indicates an emoji. Then we append a link to the Wikipedia article of the day as provided by the json payload. If the json payload does not include data for featured_title and featured_url we catch a KeyError and use json_error to return the appropriate information: a 400 http status code with relevant details.

Then we send a public (stream) message with check_send_message which will validate the message and then send it.

Finally, we return a 200 http status with a JSON format success message via json_success().

Step 3: Create an api endpoint for the webhook

In order for a webhook to be externally available, it must be mapped to a url. This is done in zerver/lib/integrations.py.

Look for the lines beginning with:


And you’ll find the entry for Hello World:

  WebhookIntegration('helloworld', display_name='Hello World'),

This tells the Zulip api to call the api_helloworld_webhook function in zerver/webhooks/helloworld/view.py when it receives a request at /api/v1/external/helloworld.

This line also tells Zulip to generate an entry for Hello World on the Zulip integrations page using static/images/integrations/logos/helloworld.png as its icon.

At this point, if you’re following along and/or writing your own Hello World webhook, you have written enough code to test your integration.

First, get an API key from the Your bots section of your Zulip user’s Settings page. If you haven’t created a bot already, you can do that there. Then copy its API key and replace the placeholder “<api_key>” in the examples with your real key. This is how Zulip knows the request is from an authorized user.

Now you can test using Zulip itself, or curl on the command line.

Using manage.py from within the Zulip development environment:

./manage.py send_webhook_fixture_message \
> --fixture=zerver/fixtures/helloworld/helloworld_hello.json \
> '--url=http://localhost:9991/api/v1/external/helloworld?api_key=<api_key>'

After which you should see something similar to:

2016-07-07 15:06:59,187 INFO       POST    200 143ms (mem: 6ms/13) (md: 43ms/1) (db: 20ms/9q) (+start: 147ms) /api/v1/external/helloworld (helloworld-bot@zulip.com via ZulipHelloWorldWebhook)

Using curl:

curl -X POST -H "Content-Type: application/json" -d '{ "featured_title":"Marilyn Monroe", "featured_url":"https://en.wikipedia.org/wiki/Marilyn_Monroe" }' http://localhost:9991/api/v1/external/helloworld\?api_key\=<api_key>

After which you should see:


Using either method will create a message in Zulip:

Image of Hello World webhook message

Step 4: Create tests

Every webhook integration should have a corresponding test file: zerver/webhooks/mywebhook/tests.py.

The Hello World integration’s tests are in zerver/webhooks/helloworld/tests.py

You should name the class <WebhookName>HookTests and have it inherit from the base class WebhookTestCase. For our HelloWorld webhook, we name the test class HelloWorldHookTests:

class HelloWorldHookTests(WebhookTestCase):
    STREAM_NAME = 'test'
    URL_TEMPLATE = "/api/v1/external/helloworld?&api_key={api_key}"
    FIXTURE_DIR_NAME = 'helloworld'

    # Note: Include a test function per each distinct message condition your integration supports
    def test_hello_message(self):
        # type: () -> None
        expected_subject = u"Hello World";
        expected_message = u"Hello! I am happy to be here! :smile: \nThe Wikipedia featured article for today is **[Marilyn Monroe](https://en.wikipedia.org/wiki/Marilyn_Monroe)**";

        # use fixture named helloworld_hello
        self.send_and_test_stream_message('hello', expected_subject, expected_message,

    def get_body(self, fixture_name):
        # type: (Text) -> Text
        return self.fixture_data("helloworld", fixture_name, file_type="json")

In the above example, STREAM_NAME, URL_TEMPLATE, and FIXTURE_DIR_NAME refer to class attributes from the base class, WebhookTestCase. These are needed by send_and_test_stream_message to determine how to execute your test.

When writing tests for your webhook, you’ll want to include one test function (and corresponding fixture) per each distinct message condition that your integration supports.

If, for example, we added support for sending a goodbye message to our Hello World webhook, we would add another test function to HelloWorldHookTests class called something like test_goodbye_message:

    def test_goodbye_message(self):
        # type: () -> None
        expected_subject = u"Hello World";
        expected_message = u"Hello! I am happy to be here! :smile:\nThe Wikipedia featured article for today is **[Goodbye](https://en.wikipedia.org/wiki/Goodbye)**";

        # use fixture named helloworld_goodbye
        self.send_and_test_stream_message('goodbye', expected_subject, expected_message,

As well as a new fixture helloworld_goodbye.json in zerver/fixtures/helloworld/:


Once you have written some tests, you can run just these new tests from within the Zulip development environment with this command:

./tools/test-backend zerver/webhooks/helloworld

(Note: You must run the tests from the top level of your development directory. The standard location in a Vagrant environment is /srv/zulip. If you are not using Vagrant, use the directory where you have your development environment.)

You will see some script output and if all the tests have passed, you will see:

Running zerver.webhooks.helloworld.tests.HelloWorldHookTests.test_goodbye_message
Running zerver.webhooks.helloworld.tests.HelloWorldHookTests.test_hello_message

Step 5: Create documentation

Next, we add end-user documentation for our webhook integration. You can see the existing examples at https://zulipchat.com/integrations or by accessing /integrations in your Zulip development environemtn.

There are two parts to the end-user documentation on this page.

The first is the lozenge in the grid of integrations, showing your integration logo and name, which links to the full documentation. This is generated automatically once you’ve registered the integration in WEBHOOK_INTEGRATIONS in zerver/lib/integrations.py, and supports some customization via options to the WebhookIntegration class.

Second, you need to write the actual documentation content in zerver/webhooks/mywebhook/doc.html.

<p>Learn how Zulip integrations work with this simple Hello World example!</p>

<p>The Hello World webhook will use the <code>test<code> stream, which is
created by default in the Zulip development environment. If you are running
Zulip in production, you should make sure this stream exists.</p>

<p>Next, on your <a href="/#settings" target="_blank">Zulip
settings page</a>, create a Hello World bot.  Construct the URL for
the Hello World bot using the API key and stream name:
  <code>{{ external_api_uri }}/v1/external/helloworld?api_key=abcdefgh&amp;stream=test</code>

<p>To trigger a notication using this webhook, use `send_webhook_fixture_message` from the Zulip command line:</p>
<div class="codehilite">
./manage.py send_webhook_fixture_message \
> --fixture=zerver/fixtures/helloworld/helloworld_hello.json \
> '--url=http://localhost:9991/api/v1/external/helloworld?api_key=<api_key>'</pre>

<p>Or, use curl:</p>
<div class="codehilite">
  <pre>curl -X POST -H "Content-Type: application/json" -d '{ "featured_title":"Marilyn Monroe", "featured_url":"https://en.wikipedia.org/wiki/Marilyn_Monroe" }' http://localhost:9991/api/v1/external/helloworld\?api_key\=<api_key></pre>

<p><b>Congratulations! You're done!</b><br /> Your messages may look like:</p>

<img class="screenshot" src="/static/images/integrations/helloworld/001.png" />

These documentation blocks should fall alphabetically. For the integration-lozenge div this happens automatically when the html is generated. For the integration-instructions div, we have added the div between the blocks for GitHub and Hubot, respectively.

See Documenting your integration for further details, including how to easily create the message screenshot.

Step 5: Preparing a pull request to zulip/zulip

When you have finished your webhook integration and are ready for it to be available in the Zulip product, follow these steps to prepare your pull request:

  1. Run tests including linters and ensure you have addressed any issues they report. See Testing for details.
  2. Read through Code styles and conventions and take a look through your code to double-check that you’ve followed Zulip’s guidelines.
  3. Take a look at your git history to ensure your commits have been clear and logical (see Version Control for tips). If not, consider revising them with git rebase --interactive. For most webhooks, you’ll want to squash your changes into a single commit and include a good, clear commit message.
  4. Push code to your fork.
  5. Submit a pull request to zulip/zulip.

If you would like feedback on your integration as you go, feel free to submit pull requests as you go, prefixing them with [WIP].