Blogging in Lisp


OAuth2 and Mastodon

Submitted by Bill St. Clair on Mon, 01 Jul 2019 22:15:18 GMT

This page explains my thinking of doing OAuth2 authorization for an Elm web frontend talking to a Mastodon API backend.

The OAuth2 Authorization Code Grant Flow Dance

OAuth2 has a number of authentication methods. For end users, using web apps, the Authorization Code Grant Flow is common. It lets the user authenticate with the service she's using, such that the client code never sees her userid or password, then the client code fetches a token it can use to authenticate API requests. If the user notices the client misbehaving, she can go to the service's web site, and remove permission for that client to access her account.

There are three computers involved:

  1. The user's computer, often via a web browser.
  2. The client application's web server, accessed via the redirectUri.
  3. The API service's web server, accessed via the authorizationUri, the tokenUri, and the apiUri.



Mastodon requires /oauth/authorize and /oauth/token as the OAuth 2 endpoints and /api/v1 as the base of the REST API URLs.

There is some information that the API service uses to identify the client application:

  1. The application name
  2. The application Url (optional)
  3. The clientId
  4. The clientSecret

The clientId can be present in the webapp, on the users's computer, since without the clientSecret, it cannot be used for anything. The clientSecret is kept secret, on the redirect server.

Steps in the dance:

  1. The user clicks a "Login" button in her web browser.
  2. The webapp redirects to the authorizationUri, passing the clientId, the redirectUri, scope descriptors, saying what the client app will be allowed to do, and a state string.
  3. The authorizationUri puts up a form, requesting userid and password.
  4. If the user logs in successfully, the authorizationUri server forwards the user's browser to the redirectUri, passing the clientId, state, and an authentication code that it generates and remembers.
  5. The redirectUri server posts the clientId, clientSecret, and authentication code to the tokenUri, and receives back a token.
  6. The redirectUri server then uses the state to put up a web page for the user to interact with the server via the API.
  7. When the user does something that requires an API call, the token is passed along, for authentication and identification of the user.

The Status Quo

First, I'll explain from where I started, back in December of 2017, before adding Mastodon to the mix.

I published the billstclair/elm-oauth-middleware in the public Elm repository. It works with Google, Facebook, GitHub, Gab Legacy, and likely any other proper implementation of the OAuth2 Authorization Code Grant Flow (but I only tested those four). It is running at It contains both server and client code.

The elm-oauth-middleware server expects application definitions to be mostly static, with the tokenUri, clientId, and clientSecret stored in a JSON file on the server, which is queried periodically, and reloaded if it changes. This allows hot changes to the applications, with a text editor on the server.

I store the clientId, and redirectUri in the Elm client code, compiled to JavaScript in the browser, again loaded from a JSON file that ships with the client application. It redirects to the authorizationUri, passing the clientId, redirectUri, scope, and some state. The authorization server (Google, Facebook, GitHub, Gab, etc.) prompts the user for ID and password, and redirects to the redirectUri with an authorization code and some Base64 encoded JSON state. That server posts the code to the tokenUri to get a token, which it returns to the client code, by using the state to go to a URL on a redirectBackHost that are validated from its configuration file. Validation of that redirectBackHost is my invention.

This is a non-standard use of the Authorization Code Grant Flow. Usually, the token stays on the server, and it uses it to make API calls and then populate HTML for the client browser. Since my clients are all in Elm, and work with no HTML generation by a server, other than an initial static HTML file that loads the Elm JavaScript, that client needs to have the token, and make calls itself to the apiUri. The client typically stores the token in JavaScript localStorage, so that it doesn't have to request it again every time the user goes to its web site.

Enter Mastodon

Normal Mastodon servers have a Your Applications page, linked from </> Development in the left column of the preferences page. This allows you to create a standard, static, OAuth2 application, giving an Application name, Application website, Redirect URI list, and allowed Scopes, and receiving a clientId, clientSecret, and token.

This is fine if your application is targeted to a small set of Mastodon servers, but the nature of Mastodon is hundreds of federated servers, so the API has a POST /api/v1/apps call to create a new client_id and client_secret.

My idea for using this is to have the Elm app send its own URL as the Redirect URI, use POST /api/v1/apps to get a client ID and secret, then redirect to https://<mastodon-host>/oauth/authorize, so the user can log in, get back the authorization code when restarted as the redirectUri, then POST to https://<mastodon-host>/oauth/token to turn that code into a token. The only thing I don't now yet is whether that final POST will pass CORS muster. The API calls have to, but that one doesn't. If it doesn't, then that part of the dance needs to be moved to a server, with the clientId and clientSecret passed in the state, so that the server doesn't need any state itself.

This is likely an unusual use of the POST /api/v1/apps call. I think it's expected that the redirectUri host will save the association of the <mastodon-host> and a clientId/clientSecret pair, so that it doesn't need to request a new one unless a new user specifies a never-before-seen server. I plan to cache the clientId/clientSecret pair (and the most recently issued token) in localStorage on the user's machine, but store no state on the server, even if I need one to get around CORS.

Add comment   Edit post   Add post

Comments (1)

It works!

Submitted by Bill St. Clair on Tue, 23 Jul 2019 14:22:50 GMT

I have implemented this idea in billstclair/elm-mastodon. It works against Mastodon and Pleroma servers. It is live at

Edit comment

Add comment   Edit post   Add post

Previous Posts:

ZAP Meme
stackoverflow April Fools 2019
Image Sizing in elm-ui
How do I run elm-test in Elm 0.19?
CLtL2 on Kludges
iPhone Xs Pricing
New Elm Package: billstclair/elm-port-funnel
Building the Elm compiler