OAuth for all
blog.cloudflare.com
blog.cloudflare.com
If anyone here is interested in providing their own oauth, IAM, rebac permissions, API keys, agent security - check out our open source & commercial products at https://github.com/ory and https://www.ory.com/
I could never understand why there were so much confusion spread around this protocol, almost every junior engineer I worked with would just struggle grasping it, I cannot recommend Scott Brady's blog enough on the topic https://www.scottbrady.io/ it was illuminating to me
I think there's an essential primitive "fear" whenever authN/Z is involved that creates friction for most engineers, they're used to problem solving and this fits within a pre-condition to your problem solving so there's a cognitive tax or something around it
That said I only use cloudflare for piping and none of the compute stuff.
Don't worry you aren't unique in this regard, many other nonworking dev influencers say similar things: never championing for workers but somehow always championing positions that help investors + corporations first and foremost
Having your workers in-house whole other businesses/industries is also not good for workers, because the problem space of that grants their actual employment won't get the required attention.
Iirc Argo and some other routing products are quite expensive.
Cloudflare pricing introduces an additional dimension for when you're architecting software on top of them, but if you do it correctly, your product has the potential to be faster, cheaper, and easier to run than traditional solutions running on multiple geographically distributed VMs. You just can't approach them as if they're just another VM instance provider and expect a similar experience and pricing. What they do and price for is fundamentally different from that.
Most people in CIAM (customer identity, individuals owing their account instead of representing a company) only interact with OAuth client for authentication. They do not give access of their google account to some THIRD PARTY COMPANY.
Your go to a third party web site. They send you to your OAuth provider, like cloudflare. Cloudflare asks you to login if you’re not logged in, then asks if you want to give that party certain permissions. You say yes or no and then click approve and then you get redirected back to the third party site. They get a secure token and can use that to access the services with permissions you approved. If you don’t trust the third party then don’t approve it.
It is like an API key but you never have to touch it. The third party can encrypt it and store it securely and it never has to be copied and pasted. You can use this on backend services that need to access things too. I recently wrote an OAuth client for MCP servers for something I’m building (not gonna advertise here because that’s rude) and it’s very nice once you read the spec.
You picked probably the only semi-straightforward thing about part of one of the OAuth specs, then hand-waved away the other 95% of the necessary related specs, knowledge, and experience for getting an implementation working robustly and securely for a non-trivial use case.
Cloudflare API Keys - You create them and then use those keys directly against cloudflare API's to manage services/infrastructure in your account. How you create the keys is may be a different kind of challenge.
OAuth flow in discussion here - You are using a third party service (which registers themselves as a the client application with cloudflare), this service is going to prompt you for OAuth flow and redirect to Cloudflare, not (only) to authenticate you but it will get a access token on your behalf (your cloudflare account) from Cloudflare. Whatever this THIRD PARTY service uses this token for your behalf is going to incur infrastructure cost for your account.
Sorry if I was rude earlier but saying OAuth is some security flaw made me think that you didn’t understand what it was about; it’s just a way to grant permissions to a third party you trust. If you do then I’m curious why you think it’s flawed.
In case of AWS, you add Github as an IDP (OIDC provider) and associate a role to it.
Github is now authenticating into AWS, scoped to the github repository where its configured and the AWS role it can assume
Its not really a typical OAuth2 or OIDC flow. And yes its better than storing the keys.
Github is not the OAuth client here.
To achieve the task, i.e: a GitHub being able to update an AWS Lambda without storing a secret/key/certificate you minimally need:
1. A way for GitHub to sign a request before calling AWS with it.
2. A way for AWS to verify this request was signed with GitHub.
3. A way to tell AWS what actions a valid GitHub request is authorized to do.
#1 is very easy with a public/private key signature. For #2 OIDC standardizes this part. Every IdP publishes their public keys in a standardized way. For #3 AWS already has a full permissions/roles system, so it makes sense to use that for this. Then you can get a GitHub action to do anything an AWS account can.
To string 1, 2, and 3 together you end up with the confusing flow you described.
To cover the myriad of (sometimes downright stupid) requirements that large enterprises have.
i make a point to implement oauth from scratch, because using the overly complex libraries expose you to bugs such as attacker sending a token which the metadata just says "no encryption or signature. trust me bro", which is actually part of the spec if you combine some options.
while in the real world, if google or apple sends you a token that is not always the same signature cypher (one of a dozen by the spec) you are better of threating as malicious, because it pretty much is. a manual implementation of a token consumer is about 20 lines... including downloading the provider keys and checking it (which most startups never do! allowing anyone to just sign a token as anyone)
It's just design by committee.
I think that most of the "just give me an API key" comments are from a <1% of end-users (developers) that know what an API key is, and are facing a broken OAuth implementation.
Or didn't bother to read the spec to understand why it's non trivial. Things like this are complex because attacks will force it to be.
Also, the broken implementation might be an OIDC implementation that doesn't support client_credentials for example. Seen that many times and that does make it rather awkward to implement a server to server flow...
OpenID piggy-backed on it by layering on new terms to an already complex scheme. The precious, secret information from Big Company in OpenID is just Email and maybe Name and Profile Picture. Then there’s a lot of ceremony for the service using OAuth to securely get that big secret (the user’s Email, which they had to supply in the first place directly to Relying Party).
It's a privacy nightmare.
(Apple login is in nearly every iOS app and most websites)
The only way to preserve privacy while having a central and easy authentication mechanism I can think of is to use IndieAuth[0] which is built on top of OAuth 2.0.
Of course, you will need to be your own provider, using an IndieAuth provider service defeats the purpose, which is what I see most IndieWeb devs are doing.
You will need to own a (sub)domain though.
You may also be interested in the FedCM protocol Google is working on.
I don’t know why anyone wants to use a federated identity to sign into things. Where did the messaging that it’s more secure come from, Google?
Why would I want the headache of having yet another login/password to remember, if (like most people) I haven't figured out password managers? I'd rather just use my Google identity, especially if I don't really care about this particular web app.
I have both Google and Github Oauth flows added to this idea in code, and it works great for my purposes. Prior to coding agents, I wrote the code by hand and went over it a few times to ensure it worked and was safe (for the email token thing at least). I even wired up the Oauth stuff myself, without an agent. It's not hard.
With agents, it's super easy to audit this and also deploy services using them, so I'm not sure why any arguments here mention how hard it is nowadays. It's not hard, but it does expose what site a user is logging into. That's just a easy function for the user though, with known risks.
With Google email, if I use a site with email logins, Google STILL knows I used them. It's just in my email instead of a logfile that I authenticated to a site. I would note that as long as the token is alive, dependent on the provider's choices, Google doesn't know I'm going back again, and has no idea what I'm doing on the site. I'd trust Google over anyone else about this, even Github (as related to Oauth to avoid the nitpick that has been common here recently).
I would hardly call this a "security nightmare" (as someone else said, not you) as Google only knows someone is authenticating to a given URL which they've vetted (a little) during the setup process. Same for Github. If you don't like Oauth, or feel that then every site someone uses should provide an email login fallback.
People trying to remember passwords is a pretty bad security situation.
I'm not an expert but so often folks on here throw criticisms without giving credit to some of the merits of solutions. Nothing is perfect, and progress can still be made. :)
Respectfully, I disagree in a time when all your data is being slurped up and resold constantly I hate any additional costs to my privacy.
> People trying to remember passwords is a pretty bad security situation.
But that's their problem, not mine. I'm an adult and I use a password manager.
One issue I see all the time (for consumer level password managers) is that, for example, their browser has a password manager, but keychain interjects often. Then they don't know where their password was saved. On top of that, Keychain does some magic stuff to pair URLs to passwords, and then there's "Passkeys" built in. When it remembers password(s), passkeys, different URLs (not very human-readable), automatically remembers stuff, injects things: it's a tangled mess.
Go to a coffee shop and ask 20 people of different demographics whether they would prefer to use a password manager app or to log into sites using an existing social or email account. We have to protect the lowest-common-denominator in terms of technical literacy. Scammers are going after elderly, so unless you have a solution to protect them, it's not THE solution.
Privacy nightmare in the real world, "tech" company wet dream in SillyCon Valley
Never want to touch oauth, it's a fucked spec.
Good thing about the OAuth2/OIDC is these things will not put the trust on the bearer of the api key, but on actual identity that needs to have the access.
> Good thing about the OAuth2/OIDC is these things will not put the trust on the bearer of the api key, but on actual identity that needs to have the access.
And... you do not see the myriad of problems with that? What about the OIDC provider going rogue or getting compromised? How do you ensure whatever you use to authenticate with your OIDC isn't compromised? Many identity providers and identity bearers have terrible security practices. "Add a backup email in case you lose your 2FA. Nevermind it's the same email we use for password reset."
Again, I trust zaptheimpaler to keep their secret much better than this whole pretend security theater.
I've never worked at an organization that handled their user's data/privacy/security even remotely close to how I handle my own and I wouldn't even consider myself all that paranoid. I have worked for some companies that really really should care too - there's just no incentive to really care and those in the org that try too do so will get ignored.
The data breach letters I get in the mail a few times a year back me up on this.
In the end there is always some long lived secret. What changes is just where and how it is stored, secured and used.
I bet we can generalize to say that data shows that you will likely fail to properly secure any secret (including the ones used in OAuth2).
EDIT: An example: https://news.ycombinator.com/item?id=37973937
> In the 2025 report, we showed that nearly 70% of credentials confirmed as valid in 2022 were still valid as of January 2025, meaning they have not been rotated or otherwise remediated. When we retested the same dataset in January 2026, the validity rate remained over 64%. That persistence is not a rounding error. It is an operational signal that remediation, not detection, is still the industry’s limiting factor.
> In the end there is always some long lived secret. What changes is just where and how it is stored, secured and used.
Obviously. The point here is to reduce the number of them
Not quite. You shift the trust from the key bearer (the most interested party in all of this) to the identity provider.
I agree that it is too complex though and app to app auth is certainly not a focus. I often still use static common secrets and see no problem with that.
I hate for apps needing to save passwords themselves, even if we have good tools today and the standard bcrypt call is reasonably safe. But then you need to reimplement password reset flows and all that ugly shit. Having that centralised is often
I would recommend self-hosting an OIDC service for that matter. The control you get also allows you to easily comply with some laws like GDPR and cousins, because you need to just purge a user in a single system.
Otherwise I thoroughly feel the frustration with IAM and the big providers. Ain't nobody got time for that and it is never a good and efficient solution.
Seconded. It is fairly easy to set up, and so much easier than the cloud IAM things.
The only catch is, make sure you have some backup access to your OIDC provider in case it goes down. E.g. don’t host it on a server with SSH only accessible through VPN that is authorised using your OIDC provider, etc.
Tailscale’s implementation of OIDC is nice: https://tailscale.com/docs/integrations/identity/custom-oidc
But all that only makes sense if you own a domain name.
I have a hard time believing the venn diagram of "has a need for an auth provider" and "has at least one domain name" isn't just a a small circle almost entirely inside a large one, and the sliver on the outside is not for any reason other than stubborn refusal.
My point was basically: OpenID 1 tried to let people log into websites using domain names. This is still possible to implement (and easier than OpenID, IMO), but people just don’t want to bother with domains.
For API Keys we just launched Ory Talos (https://github.com/ory/talos) - a perfect alternative for when OAuth2 is too much for the use case.
There are use cases and security concerns that legitimize using OAuth2 - with specs like DPoP you can make these flows more secure. In my view the use cases presented here is a good one for OAuth2, but it certainly doesn’t make sense everywhere - complexity makes system harder to secure.
Then implement that on your app... You are just generating a random key and storing a hash + salt.
Auth is hard only applies to auth for many users. For your own auth this is dead simple and made even simpler if you use a half decent framework...
If you are really worried about the implementation being insecure throw one of the many moderately frontier models at it, they are really not bad at finding issues in an auth system that simple.
One good thing GitHub Copilot has it that you can just give it a GH_TOKEN that is valid for 6 months and stop this browser login nonsense.
User shows up at your login flow. You assign them a big random number identifying their user session (this is your "state")
User indicates an identity provider they'd like to use. You probably have a short list you trust.
You ask that provider for the configuration data.
You generate a big random number, that identifies this log in attempt as unique. This is your "nonce".
You send the user, along with the state and nonce, to the trusted third party. (at their "authorization endpoint")
The user proves to the trusted third party they are who they say they are. This isn't your problem.
The user comes back to you with a claimed state and a code (a big random number assigned by the trusted third party).
You check that the user's claimed state matches the state that you assigned them. This ensures that you end up authenticating the same user session as the one that started the login.
You then reach out to the third party directly (to their token endpoint), with state and code in hand, and ask them "yo, a user session with this state just claimed you sent them to me with this code. Who are they?".
And then the trusted third party sends back a token attesting "They are so and so".
The one superfluous step is that, according the spec, you're supposed to then verify the signature of that token. It is unclear to me why this is in the spec, since I just made an https request to the trusted third party. The entire security model here has assumed that trusted third party is trusted.
I love how simple SSH is with it's PK-Auth. The only challenge is session-invalidation and key-management, but that can be surely automated, no?
Expect it. Security is hard and the companies with deep pockets are happy to pay the bill that meets their cybersecurity insurance requirements.
When it’s only used for SSO, it’s extreme overkill.
Or just stick with KeyCloak that offers a full self hosted product... [1]
Yes we have an commercial version because how else can one finance world class open source powering the biggest software names on the planet? It‘s a good thing that Ory has a business model that works, not a bad thing. And by the way, IBM finds ways to charge you for KeyCloak too ;)
Looked at the case study, uses Cockroach which is now commercial, so potentially with the dual costs of Ory and Cockroach licenses, unless you need massive scale, would be too expensive for small/medium and also startups? Unless your sole focus is on enterprises?
And Keycloak also has such a implementation https://www.cockroachlabs.com/blog/deploying-keycloak-on-coc...
If you serve 900m weekly active users, you need this type of distributed database architecture that is expensive to run. But at that point the cost of running it is a fraction of overall infra spend. No start up really needs this level of scale, only Enterprises (hence it‘s gated). Making Cockroach work is more work than just wiring up the SQL, you actually need to deal with it like dynamodb under the hood and use primary keys efficiently, avoid hotspots, and all that jazz.
Most companies (like Cloudflare!) do just fine with Postgres and one of our services. Ory Hydra is written in Go, doesn’t need JVM, very little RAM, doesn’t need caches or start up time due to cold starts. The architecture is different and that makes it cheap and fast to run. From the blog post - they run Hydra on 0.6 vCPU and 200MB of RAM. That’s probably as cheap as it gets!
It‘s a different tool for a different problem than KeyCloak - both have their place.
Glad to see them making use of Hydra for OAuth apps!
What's a "self-managed" Oauth here? What is access is being granted to, who are the clients, who are the partners...?
Anyone care to elaborate?
They're letting you host an OAuth system to approve/deny access to your own resources, so you can build whatever logic you like, rather than waiting on them to allow you to do X under Y conditions. Essentially "log into CloudFlare" -> CF sees you're using this self-managed OAuth -> redirect to your OAuth -> CF trusts your response, and approves access to your account if you approve access.
Anyway good start in the right direction from Cloudflare, yet still long way to go especially compare to the full Ory's offering its built on. Ory's Kratos handles identity, login, registration, recovery, MFA... https://github.com/ory
IMHO full scope should include plans on user store, SAML, multi-tenant org model. Good example - Zitadel https://github.com/zitadel has managed UI for orgs multitenancy, OIDC/PKCE supports, etc you can even partial glue RBAC to it
Subabase offers managed and opensource https://github.com/supabase/auth
Siding "MCP is dead, Skills forever" what bother me about all of them is planning to plug MCPs and rotate keys ... this start hitting the fan very soon
OAuth 2.0 Dynamic Client Registration (RFC 7591) https://datatracker.ietf.org/doc/html/rfc7591
https://modelcontextprotocol.io/specification/2025-03-26/bas...
Any comments greatly appreciated. Especially in multitenant saas and built-in "AI assistants" context
The spec handwaves around this talking about initial access tokens which a client would obtain first in order to register but the details are sparse and probably unworkable when we're talking about every end user being a client.
Ideally i would be able to specify an allowlist of redirect patterns so i could limit it to say, chatgpt or whatever else. But that would be a non-standard behavior so my IAM vendor isn't in a hurry to do it.
Once their revenue from Cloud services overtakes their core offering, bye bye Cloudflare free and so on.
Wait so what do you think their core offering is?
But even to entertain this is crazy, not because of decades of history of capitalist and market enterprise in general, but very specific cases of Technology Companies starting with these kind of feel good ideas and declaring "Don't be evil" or things like " access, safety, and shared prosperity" as their core ideals, turn into absolute panopticon and collaborate with unjust killing of women and children in less than a decade.
The market isn't for free.
Their first products were production grade examples of the SDN that required a lot of bandwidth (DDOS/CDN).
The cloud is a logical continuation.
Their business was always the "internet", see their ticker => NET.
Dev free is part of the marketing cost and would stay under the current leadership.
> After investigation, we discovered that there was an issue in one of the Hydra migrations that corrupted the state of certain valid OAuth sessions, which resulted in the migration marking them as invalid.
Was this one of the open source migration files? While I'm no longer involved in the project, I'd be curious to know if it's been addressed upstream.
Ideally I would like to be able to ask a user for their Google/Microsoft/Apple email address and just drop it into a config file for authorization. The user then autenticates themselves at their G/M/A Id server and gets access. Or is this too simplistic?
Better Auth seems to be the most common recommendation for Typescript applications, but there currently doesn't seem to be an official integration with Workers either from Better Auth or from Cloudflare.
I currently use Supabase to avoid having to set up my own user auth on Workers, but I would much prefer to use D1 etc.
https://github.com/rorz/manual.email/blob/main/packages/db/s...
I find Lucia Auth’s approach more useful in the long run – you have some boilerplate living on your codebase but you own it completely and it doesn’t try to make decisions for you: https://lucia-auth.com/
---
That said, why don’t you use Better Auth with Drizzle and the D1 adapter?
Not sure why they don’t just support DCR or CIMD for this too
probably getting ahead of something the UK and some us states will require soon, as they already require from the sites behind cloudflare.
- They launched Cloudflare Web Analytics in 2020, but it still does not support basic things such as UTM parameters or custom events
- With wrangler (their CLI), you still cannot undeploy a Cloudflare Page
https://github.com/cloudflare/workers-sdk/tree/main/packages...
Wrangler, being cloudflare’s primary CLI tool, is a microcosm of exactly the problem GP was articulating: it’s focused way more on adding new commands than improving existing ones.
Many products, even supposedly “GA” ones, still lack basic operability via wrangler because instead of finishing building out its capabilities to manage existing services, they prioritized adding rudimentary support for new ones.
Abandoning something, and not making the changes you want to see are entirely different things.
It's not handwaving to complain that shipped features are never completed and a functional CLI for fundamental and critical tasks is never delivered.
There are numerous examples, such as Cloudflare claiming Workflows had reached "GA" status before offering a way to delete workflows... not via wrangler, not the dashboard, not the API. They eventually added this many months after declaring it "GA", with no upfront disclaimers, just the capability silently missing.
If they want to move fast and break things (and they do, as their numerous, "CEO declares an emergency" outages have evidenced) they can — but they need to be honest in their communication about the production readiness of their new products.
That is not "life and priorities": it is poor communication/judgement from an infra company, and misleading advertising at worst. I choose to believe it's the former.
Apologies that I didn't notice you weren't CommonGuy, who said "improving them in the future is not really their style". Consider my reply aimed at them. The two of you seem to be saying something quite different. I don't think I necessarily disagree with you
It's full of technical details, but I'm really not sure who they're for. There's nothing particularly novel or impressive. If anything the fact that it took them this long should be embarrassing. They pad it out with a table of stats that are just kind of meh? Congrats I guess for releasing something without burning the house down?
As an on-and-off customer of theirs I tried to quickly skim for some of the details that would impact me, the theoretical end-user, but the vast majority of TFA is just about how they pulled off this apparent feat of engineering.
I'm not trying to be pessimistic, and I don't fault the author (but I question the culture). I honestly don't get who this is for.
For the record this is something they should have had... at least six or seven years ago?
But this is so mundane it bothers me in a way I find surprising. It's more about how they made some questionable choices in the past and how they finally paid off that technical debt. Is it interesting? Perhaps I am just getting old and jaded.
What I find odd is how light TFA is on actual details as to what it is they shipped.
This is the kind of thing I'd ship internally to the org as part of a weekly update or something, but not what I'd expect on a public-facing corporate blog.
Love em., greatest tech company of all time. One stop shop.
I can't keep track of all the new things they do. Something-something-R2? Maybe?