JWT Decoder

How it works

1. 1

   Paste the token, Bearer prefix and all

   It decodes as you type. If what you pasted is not a JWT, this jwt decoder says what it is instead — a JWE, an opaque session ID, an API key, or a token that lost its last segment on the way out of a log.

2. 2

   Read the expiry line first

   Expired forty seconds ago, not valid for another two minutes because of clock skew, or no exp claim at all. This is the answer to most of the questions that bring people to a jwt debugger, so it is the first thing on the page rather than a number you have to convert yourself.

3. 3

   Then read the findings

   alg set to none, a header pointing at an attacker-controllable key URL, a lifetime measured in months, or an email address sitting in a payload that anyone holding the token can read. Each one says why it matters rather than just naming it.

Frequently asked questions

Do you store or send my token anywhere?

No. The decoder is a few kilobytes of JavaScript that runs on your machine — there is no server call, nothing is logged and there is no account. This matters more here than on most pages: a JWT from a live session is a working credential until it expires, so pasting one into a website is closer to pasting a password than to pasting a config file.

Why does this not verify the signature?

Because verifying it means handing over the key. For HS256 that is your signing secret, and a jwt validator that asks you to paste a production signing secret into a text box on the internet is asking for something worse than the problem it solves. For RS256 you would need the issuer's public key, which tells you the token was signed by whoever holds the private key — something your own backend already checks on every request. Everything this page reports can be established without either.

Is the payload encrypted?

No, and this is the single most common misunderstanding about JWTs. The payload is base64url, not encryption — a signature stops it being changed, not read. Anyone holding the token can see every claim, including your users, any proxy on the path, and anything that logs an Authorization header. So this jwt decoder lists any claim whose name suggests personal data, because that data is effectively public to everyone the token touches.

My token is valid but the API rejects it. What now?

In the order they actually happen: the token expired between issuing and use; nbf is in the future because the issuing server's clock is ahead of the verifier's; the aud claim names a different service; the iss does not match what the verifier expects; or it is an ID token being sent where an access token was expected. The panel shows all five, which is why it is more useful as a jwt debugger than a plain decode.

What does alg: none mean?

That the token is unsigned. A library that reads the algorithm out of the token rather than pinning the one it expects will happily accept a payload anyone edited — the attack described in CVE-2015-9235, which still ships in real code. It is flagged as a danger for that reason. If you see it on a token from a production system, the token is not the problem.

Why does it warn about jku and x5u in the header?

Those header parameters tell the verifier where to fetch the key. If the verifier follows the URL without an allowlist, whoever controls the token controls the key, and therefore controls the signature. A jwk in the header is the same problem in a shorter loop: the token supplies the key used to check the token.

What is the difference between an ID token and an access token?

An ID token describes who the user is and is meant for your application. An access token authorises a call and is meant for an API. They look identical, which is why sending the wrong one is such a common bug — Cognito's token_use claim exists precisely to tell them apart, and a jwt validator on the API side should be checking it.

Can it tell me which provider issued the token?

Yes, from the iss claim. Auth0, AWS Cognito, Firebase, Keycloak, Okta, Supabase, Microsoft Entra ID, Clerk, Stytch and WorkOS are recognised by their issuer URLs. It is a small thing, but useful when a token has arrived from a system you did not set up.

Does it work as a jwt token decoder for tokens from a header or a log line?

Yes. A leading Bearer is stripped automatically, and surrounding whitespace is ignored. As a jwt token decoder it also handles base64url properly — the dash and underscore characters that appear in most real signatures, and that decoders built on plain base64 quietly fail on.

Why does it warn about the token size?

Because browsers cap a single cookie at about 4 KB, and several proxies and servers cap request headers by default too. A token that grows past that limit starts failing in ways that look like an auth bug and are not. Large role or permission arrays are almost always the cause.

Why does it warn about long lifetimes?

A bearer token is valid until it expires, and nothing checks back with the issuer in between. That means a leaked token with a thirty-day lifetime is a thirty-day problem unless you maintain a denylist. Short access tokens plus a refresh token exist to make the window small.

What if I paste something that is not a JWT?

You get told what it is. Five segments means a JWE, which is encrypted and cannot be read without the key. A single opaque string is a session ID rather than a token with anything inside it. Two segments usually means the signature was truncated on copy. "Invalid token" would send you hunting for a typo that is not there.

Is there a jwt token decoder for expired tokens?

This one. Expiry is a claim inside the payload, not a property of the encoding, so an expired token decodes exactly like a valid one — the difference is only that a verifier will refuse it. Being able to read an expired token is usually the whole point of opening a decoder.

Why is this free, and what is Tekk?

Tekk is a spec-driven development platform for people building software with AI coding agents. This page costs nothing to run because nothing runs on our servers, so there is no signup wall, no run limit and no upsell inside the tool.