Document Your APIs With OpenAPI

Over the past 5 years I’ve worked on an API first application full time. And over those 5 years we’ve learned a lot about API design. In this post I’m going to focus on documentation, how we approach it, and some of the tools we use.

Document First

Previous projects at our company used documentation generation tools. These tools are useful because they tie into your code and don’t take a lot of work to generate documentation. The documentation also updates as your make changes in code. While there are positives to these tools, there are some drawbacks.

If you use a documentation generation tool, documentation comes after you write code. A good API is designed before you write any code. Take time to think through all the use cases. Sit down with the people that are going to be using your API. Start documenting and iterating on that design.

After you’ve come to an agreement on the documentation, you have an artifact that all parties can start building to. The backend team can start building out the database models at the same time frontend or mobile developers build out their integration. You can even use that documentation to generate mock data to work with.

Another benefit is that your API doesn’t become tightly coupled to your database. While this isn’t necessarily bad, it can cause issues later down the road if your database schema changes.

Stick With The Standards

API documentation has come a long ways in five years. When we started working on our API, the options were Swagger, RAML, Blueprint, SASS solutions, or to build our own. Swagger had some trouble spots for us (or probably a lack of understanding the specification) and the other systems were too complex. Against our better judgment, we wrote our spec using a static site generator. It gave us a nice documentation site, but when it came to tooling, we were on our own.

Had we stuck with Swagger from the beginning and not given up at the first issue we encountered, we would be in a much better spot today. We still have endpoints using our old documentation and making updates to those takes extra effort because of it.

We learned our lesson and started writing our new API specification in OpenAPI. Working with our spec has become much easier. By using OpenAPI, we can use third party tools and libraries that support OpenAPI. You can automatically generate SDKs, run validation, mocking data, and more.

Our Process

Before I talk about our process, we should talk about how we organize our specification. We are potentially working on multiple features at once. To prevent issues with merge conflicts, we split up our specification files and stitch them together. To give you an idea, here is the basic setup of our spec directory.

    _base.yaml # Contains the base structure, info, and common schemas
    animals.yaml # Contains endpoints and schemas related to animals
    internal/ # Contains files for internal use endpoints
        users.yaml # Contains endpoints and schemas related to users

Here is our _base.yaml file. This contains the structure and basic components that most endpoints use.

# _base.yaml
openapi: "3.0.2"
  title: Awesome API
  description: Description for the awesome API
  version: "1.0.0"
  - url:
  - url: http://api.localhost/api/v2
    description: local
    x-internal: true
  - name: Animal
    description: Endpoints for animals
  - name: User
    description: Endpoints for user management
    x-internal: true
paths: {} # other files will get merged here
      description: Error
      # rest of response
  schemas: # Common schemas
      type: object
      # rest of common schemas

Then we create a file for each collection of endpoints. These files contain path and components.

# animals.yaml
      operationId: animals.list
      # rest of the spec
      type: object
        # rest of the schema

To get a single file, we use a npm module that I wrote. It deep merges files matching a pattern and outputs to a single file.

npx openapi-stitcher build "spec/**/*.yaml" openapi.yaml


When we get requirements for a project we’ll start writing the spec. During this we’ve found a UI to be useful, so we’ve built one into our stitcher. Passing the --watch flag will reload the UI when it detects changes.

npx openapi-stitcher serve --watch "spec/**/*.yaml"

We store our specification in our API codebase. This allows us to version our documentation with our API. It also has the benefit of being able to use merge requests to discuss and approve the proposed (or changed) specification.


The OpenAPI specification can be a little hard to start writing. How can you make sure what you’re writing is valid? We use a linter! We were using Speccy, but switched to Spectral. It validates the specification and does some checks for best practices.

We have a step in our continuous integration pipeline that validates our specification is valid, making sure we’re always shipping a quality specification.


When looking to publish your OpenAPI specification, there are tons of options. One of the last steps on our continuous integration pipeline is publishing a UI to an S3 bucket. Now anyone can view the documentation without having to run the UI locally.

We have both public and internal endpoints. Our stitcher tool use to take care of excluding files when creating the spec. But we’ve found a better way to do this with another npm module. OpenAPI allows for custom tags starting with x-. Anything that has an x-internal tag is removed from the final specification.

npx openapi-filter openapi.yaml openapi.public.yaml


Documentation is hard. External documentation is even harder. It’s easy to forget to update as your code changes. To prevent this, we integrate our OpenAPI specification to our functional tests. When we test our endpoints, we use validators to make sure what’s returned, matches what we’ve documented.

Where To Start

If you’re new to OpenAPI, first check out the official site and read up on the specification.

If you want to start writing a spec with little to no knowledge, checkout Stoplight Studio or other GUIs editors. These assist in the process of writing endpoints and model without needing to be familiar with OpenAPI.

For a list of other OpenAPI compatible tooling, check out OpenAPI.Tools.

Comments? I'm @mloberg on Twitter.


Matt Loberg

Matt Loberg

Software Engineer passionate about DevOps and Open Source.