AgentCore Identity OAuth 2.0 discovery: what changes
Amazon Bedrock AgentCore Identity now accepts OAuth 2.0 discovery URLs, not only OIDC ones. Who should care, what it costs and how to try it.
On November 13, 2026, AWS announced that Amazon Bedrock AgentCore Identity supports OAuth 2.0 discovery URLs. In short: you can now point AgentCore at an authorization server that publishes plain OAuth 2.0 metadata, without needing OpenID Connect (OIDC). If your identity provider was ruled out for AgentCore because it lacks OIDC, that blocker is gone. If you already use an OIDC provider, nothing changes for you.
What was announced, and what was there before?
AWS's announcement says AgentCore Identity now accepts OAuth 2.0 authorization server discovery URLs in two places (the inbound guide names Runtime and Gateway as what the authorizer protects):
- Inbound: the JWT authorizer that protects an AgentCore Runtime or Gateway.
- Outbound: the OAuth credential provider your agent uses to get tokens for other services.
The page states the earlier limit plainly: "Previously, AgentCore Identity accepted only OIDC discovery URLs." Authorization servers that implement OAuth 2.0 but not OIDC publish their metadata at a different well-known path, so they could not be used.
- Your identity providerpublishes metadata at a well-known URL
- Discovery URLOIDC or OAuth 2.0 (new)
- AgentCore Identityreads keys and endpoints
- Runtime or outbound calltokens validated or obtained
What is a discovery URL, and how do the two types differ?
A discovery URL is where a server publishes a JSON document describing itself: its issuer, token endpoint and the keys used to sign tokens. A client reads it instead of being configured by hand.
AWS's inbound authorizer guide says the URL must contain /.well-known/openid-configuration or /.well-known/oauth-authorization-server, optionally followed by a path. The second is the path defined by RFC 8414, OAuth 2.0 Authorization Server Metadata.
One detail is easy to get wrong. RFC 8414 says that when the issuer has a path, the well-known string is inserted before that path. The RFC's example: https://example.com/issuer1 maps to https://example.com/.well-known/oauth-authorization-server/issuer1. OIDC Discovery appends the suffix instead. For issuers without a path, the two agree. Copy the exact URL from your provider's documentation rather than building it by hand.
Who should care, and who can ignore it?
Care if:
- Your sign-in system is an OAuth 2.0 authorization server without OIDC support, and you wanted to put it in front of agents or tool gateways.
- You run an agent that calls a third-party API whose authorization server publishes only RFC 8414 metadata.
- You maintained a workaround, such as a proxy that fakes an OIDC document, or manual endpoint configuration.
Ignore it if:
- Your provider already publishes OIDC discovery and your AgentCore setup works.
- You do not use AgentCore Identity.
What does it cost, and what are the limits?
As of November 2026, the announcement says the feature is available at no additional cost, in all AWS Regions where Amazon Bedrock AgentCore is available. The page does not list those Regions, so check the AgentCore documentation for yours.
The custom provider guide says a discovery URL is the recommended option for a custom outbound provider. It also keeps a manual option, where you enter the issuer, authorization endpoint and token endpoint yourself, for providers with no discovery document.
How can a team try it safely this week?
- Pick one non-production provider. Use a test tenant of your authorization server, not the one your customers use.
- Fetch the metadata yourself first. Open the discovery URL in a browser or with
curl. Confirm the issuer in the JSON matches what your tokens carry iniss. - Configure the inbound authorizer. The guide says at least one of allowed audiences, allowed clients, allowed scopes or required custom claims is required, and that if you set more than one, all are verified. Start with audience and client.
- Test the failure cases. Send a token with the wrong audience, an expired token and a token from another client. Each should be rejected; confirm that in your own test.
- Then try the outbound side. Create a custom credential provider with the discovery URL and a test client. The guide lets you store the client secret in AWS Secrets Manager, so you do not paste it into a console.
What should we check before adopting it?
When we review an agent platform's identity setup, we start with these questions:
- Who owns the issuer? A discovery URL tells AgentCore which keys to trust. Treat changes to it like changes to a trust boundary, with review.
- Is the audience narrow? An audience claim says which API a token is meant for. A token valid at many APIs defeats the point of the check.
- Are scopes and claims doing real work? Authenticating is not authorizing. Decide which agent may call which tool.
- What does the audit trail show? You want to tell a user action from an agent action. If you are on Cognito, we covered its token exchange for that purpose.
- Is the metadata endpoint reliable? The authorizer fetches keys from it, so its availability and its rotation behaviour are part of your agent's uptime.
If you want a senior second opinion on how agents authenticate to your systems, see our services.
General information, not security or legal advice. Features, Regions and limits change: confirm them in the AWS documentation, and have a security professional review your identity design.