Our integrations are one of the most important parts of Zulip, so it's important that they are all implemented to the highest standard of quality that we can write.
Contributions to this guide are very welcome, so if you run into any issues following these instructions or come up with any tips while writing an integration, please open an issue or submit a pull request!
We have several different ways that we integrate with 3rd part products, ordered here by which types we prefer to write:
-
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 add a new handler in
zerver/views/webhooks.py
(plus test/document/etc.). An example commit implementing a new webhook. zulip#324. -
Python script integrations (examples: Trac, 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 tarballs (by writing the integration in
api/integrations
). -
Plugin integrations (examples: Jenkins, Hubot) where the user needs to install a native-language plugin into their existing software, which can be more work, since we may not have Zulip API bindings for the relevant language.
-
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.
-
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.
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/views/webhooks/
; there are a lot of examples in that directory. We recommend templating off a short one (liketravis.py
), since the longer ones usually just have more complex parsing which can obscure what's common to all webhook integrations. -
Then write a test for your fixture in
zerver/test_hooks.py
, and you can iterate on the tests and webhooks handler until they work, all without ever needing to post directly from the server you're integrating to your Zulip development machine. To run just the tests from the test class you wrote, you can use e.g.test-backend zerver.test_hooks.PagerDutyHookTests
See https://github.com/zulip/zulip/blob/master/README.dev.md#running-the-test-suite 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 (see below)!
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/
, not the legacy/api/v1/send_message
message sending API. -
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)
Every Zulip integration must be documented in
templates/zerver/integrations.html
. Usually, this involves a few
steps:
-
Add an
integration-lozenge
class block in the alphabetically correct place in the main integration list, using the logo for the integrated software. -
Add an
integration-instructions
class block also in the alphabetically correct place, explaining all 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. -
Finally, generate a message sent by the integration and take a screenshot of the message to provide an example message in the documentation.
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.