Skip to content
IAM Insights

Explainers · Glossary

Resource Indicators for OAuth 2.0

Resource Indicators · RFC 8707 · RFC8707

By Boston Identity · Explained · 6 min read

The short answer

An OAuth extension that lets an app tell the token issuer which service it will use a token at, so the issuer can make a token that works only there and not at every service it protects.

What it is
A protocol
Defined by
IETF, RFC 8707 (Standards Track)
Who it's for
Teams connecting agents and apps to APIs
In our briefings
34By month of the source: 2025-12: 0, 2026-01: 1, 2026-02: 0, 2026-03: 0, 2026-04: 1, 2026-05: 2, 2026-06: 2, 2026-07: 4, 2026-08: 1, 2026-09: 7, 2026-10: 3, 2026-11: 00by month, last 12
01

What is Resource Indicators for OAuth 2.0?

Resource indicators are one request parameter, named resource, added to OAuth 2.0, the open standard apps use to get access tokens for APIs on a person's behalf.§2 With it, a client (the app asking for access) tells the authorization server (the service that issues tokens) which protected resource it means to use the token at.Abstract The value must be an absolute URI, typically a web address such as https://api.example.com/app/, though it can also be a more abstract identifier.§2

The problem it solves is a token that works in too many places. Most OAuth access tokens are bearer tokens: anyone holding one can use it.§1 Scope, the older OAuth field, usually says what kind of access is wanted, such as "calendar" or "admin:org", and not where the token will be redeemed.§1 Once the authorization server knows the destination, it can restrict the token's audience, the service the token is for, so it fails at other resources.§1 The authors note that several deployments had already built their own parameters for this, and the RFC aims to offer one shared alternative.§1

Not the same as scope, which says what the client may do. The resource parameter says where it may do it, and a request can carry both.§2.2

02

How does Resource Indicators for OAuth 2.0 work?

Three kinds of actor take part: a client, the authorization server that issues tokens, and the protected resources, here a calendar API and a contacts API taken from the RFC's own examples.§2.1

Figure steps
  1. The client names the resources it wants
  2. The authorization server checks the request
  3. The client asks for a token for one resource
  4. The authorization server mints an audience-restricted token
  5. The client refreshes for the next resource
Fig. 1

The client names both APIs up front, then takes a calendar-only token and later trades its refresh token for a contacts-only one, so neither token works at the other service.

Drawn from RFC 8707, section 2, Figures 2 to 6

Swipe for the full figure →

  1. The client names the resources it wants

    In its authorization request the client may add one resource parameter per service, for example the calendar and the contacts APIs together.S1 Each value must be an absolute URI with no fragment, and should not carry a query string unless the application needs one.§2 The client should use the most specific address it can, usually the base address of the API.§2

  2. The authorization server checks the request

    If the server cannot parse a value or will not accept it, it should reject the request with the error invalid_target.§2.1 If the client leaves the parameter out, the server may go ahead with no resource or a default one, or it may require the parameter and fail the request.§2.1 In the code flow, the named resources apply to the whole grant the authorization code stands for.§2.1

  3. The client asks for a token for one resource

    At the token endpoint the client sends the resource parameter again, now naming where it will use this particular token, such as the calendar alone.S1 This works for every grant type.§2.2 Which values the server accepts is up to its local policy, and it may limit them to what the resource owner originally granted.§2.2 When an authorization code is exchanged for a token covering only some of the granted resources, any refresh token returned stays bound to the full original grant.§2.2

  4. The authorization server mints an audience-restricted token

    The server should restrict the token's audience to the named resource, for example in the aud claim of a JSON Web Token (JWT).§2 It may use the exact resource value as the audience, or map it to a broader identifier.§2 Where it can, it should also downscope, giving the token only the scope that service needs, and it reports the effective scope when that differs from the request.S1

  5. The client refreshes for the next resource

    Later the client sends its refresh token with a different resource value, the contacts API, and gets a separate token for that service with the contacts scope.S1

03

Why does Resource Indicators for OAuth 2.0 matter now?

The Model Context Protocol (MCP), which connects AI agents to tools, made this parameter mandatory in its 2026-07-28 authorization specification.1 MCP clients must send a resource parameter naming the server's canonical address on every authorization and token request, and MCP servers must validate the audience of every token.1 We read that as ruling out the token-passthrough gateway: under the specification, a proxy that hands a person's token on to the tools behind it is out of bounds.1

At least one vendor has it in beta. Cisco Duo added resource indicators to a beta MCP integration, so an agent can name the resource it wants and get a token valid only there.2 Duo's Medfisch says this keeps a token for one service from being replayed at another.2

The parameter does not settle everything for agents. One arXiv author, Liu, argues that OAuth does not say how a provider's canonical resource should override a resource string the caller supplies.3 That is one paper with an in-memory prototype, so we hold it at low confidence.3

04

Where does it fit?

Compared withHow Resource Indicators for OAuth 2.0 differs
JWT Profile for OAuth 2.0 Access TokensResource indicators shape the client's request to the issuer. The audience they produce ends up inside the issued token, in the aud claim when the token is a JWT,§2 and the layout of that token is a separate question from this RFC.
OAuth Token IntrospectionResource indicators set a token's audience when it is issued. Introspection is a separate exchange, and its response carries that same audience in a top-level aud member.§2
05

What goes wrong?

  • A token with several audiences is valid at each of them, so any one of those services can use it to reach the others. The RFC encourages a single resource per request and says multi-audience tokens need a high degree of trust between the parties.§3
  • On multi-tenant or user-content hosts, a resource value that leaves out the tenant's part of the address lets one tenant's token work against another tenant's data.§3
  • An abstract identifier instead of a real address shifts work to the client, which must check out of band that the endpoint it sends tokens to is the intended audience.§3 A network address lets the client check that directly, which lowers the risk of a malicious endpoint collecting tokens meant for another service.§3
  • The parameter is optional in the RFC, and audience restriction is a should, not a must.§2 Protection exists only where the issuer restricts the audience and the resource checks it; MCP narrows that gap: its clients must name the resource and its servers must validate each token's audience.1
  • Naming several resources and several scopes in one request asks for a token with every scope usable at every resource, which the RFC calls a cartesian product.§2.2 The server should downscope each token to what its service needs.§2.2
  • The resource parameter may let the authorization server track a user's behaviour in finer detail than before.§4
06

What should a security leader ask?

  1. 01

    Does our authorization server reject token requests that omit the resource parameter, or does it quietly issue a token with a default audience?

  2. 02

    Which of our APIs check the aud claim on every token, and which accept any valid token from our issuer?

  3. 03

    Do any of our gateways or MCP proxies pass a user's token on to the services behind them?

  4. 04

    Where we host tenants on one domain, does each resource value include the tenant's part of the address?

  5. 05

    Which of our clients request tokens for more than one resource at a time, and do those services trust each other enough for that?

07

Sources

›Evidence: 3 claims from 3 briefings
  1. 1BI-FIND-2026-0191 · MCP's authorization spec makes the agent a plain OAuth client and leaves delegation to an extensionThe specification binds every token to one server and forbids passing it on. Clients MUST send a resource parameter naming the server's canonical URI on every authorization and token request, and MUST check the issuer on every authorization response, the two controls against mix-up attacks. Servers MUST validate the audience of every token and 'MUST NOT accept or transit any other tokens'. Tokens travel only in the Authorization header, refresh tokens stay confidential and are never assumed, and offline access should not be advertised. That is the same anti-pattern the six-vendor agent draft names, a tool forwarding the token it received, and it is the rule a token-passthrough gateway breaks: a proxy that hands a person's token to the tools behind it is out of specification, whatever the vendor calls it.
  2. 2BI-FIND-2026-1194 · Cisco Duo's MCP beta lets agents register themselves; Duo calls client ID metadata documents the futureCisco Duo added two OAuth standards to a new beta integration for the Model Context Protocol (MCP) and to its OAuth 2.1/OIDC integration, also in beta, to match the MCP specification released on November 25. Dynamic client registration (RFC 7591) lets agents register themselves as OAuth clients, and Duo says admins keep full control over the access level of clients registered that way. Resource indicators (RFC 8707) let an agent name the resource it wants so Duo issues a token valid only there, which Medfisch says ensures a token for one service "cannot be replayed or accepted by another." Client ID metadata documents are not in the beta: Medfisch calls them "the future" and says Duo is building support, with no date.
  3. 3BI-FIND-2026-1270 · In one author's MCP benchmark, a confirmation prompt on every call still let 6 of 10 attacks throughLiu wants this execution layer to sit next to OAuth. In Liu's reading, OAuth does not specify how a local agent runtime should bind a file handle to a principal, how a provider's canonical resource should override a resource string the caller supplies, or how a denied data-pipe attempt should be recorded as audit evidence, and "a realistic deployment would need both layers." The paper answers those gaps with HCP's own rules and proposes no change to OAuth. Its prototype keeps grants and audit records in memory, with no provider attestation.

›In our briefings

The claim each of the newest briefings on Resource Indicators for OAuth 2.0 would have you leave with.

  1. The IETF datatracker shows that the IESG approved OAuth Identity and Authorization Chaining Across Domains (draft-ietf-oauth-identity-chaining) as a Proposed Standard on 22 June 2026, and that it has been in the RFC Editor's queue since that day. Revision 17, the text we read, was posted on 19 July 2026 during that editing. Its authors are Arndt Schwenkschuster and Pieter Kasselman of Defakto Security, Kelley Burgin of MITRE, Mike Jenkins of NSA-CCSS, Brian Campbell of Ping Identity and Aaron Parecki of Okta. Campbell co-wrote RFC 8693 and RFC 7523, the two RFCs the draft combines, and Parecki co-wrote RFC 9728, one of the ways it names for finding the other domain's authorization server. The working group's write-up for the IESG lists implementations, among them Keycloak 26.5 and Okta's. Two other drafts we have read point to it: the transaction-tokens draft leaves crossing domains to it, and the AI agent authentication draft says cross-domain access SHOULD use identity chaining.

    OAuth's draft for crossing trust domains is approved for publication and never says agentBI-FIND-2026-0184

  2. Authors from Anthropic, OpenAI, Okta, Defakto Security and Ping Identity propose that an authorization server register an agent platform's Workload Identity Provider once, after which it does not reject an agent just because it has never seen it. The provider signs a JWT naming one agent by an opaque identifier that is never reassigned, and the agent presents it through the existing RFC 7523 JWT-bearer grant. The server checks the grant against the one registration it matches and that registration's keys, grants access under its own policy, and can cap how many new agents a registration brings. A server "MUST NOT require a client registration per Agent". The authors' reason is what happens today: every agent of an installation ends up sharing one credential, "at the cost of any attribution of an individual agent's actions".

    Register a platform's identity provider once, not each agent: trust rests on its issuer, with no revocationBI-FIND-2026-0521

  3. WorkOS's checklist limits a delegated agent to the intersection of what the agent is allowed to do and what the user it acts for is "currently allowed to do," and says to check both on every tool call, not once per session. Its example code reads the agent's ceiling from the token and makes a live check for the user, and its test is that revoking a user's access to a project mid-task makes the agent's next call on that project fail. In the post's own example, an agent whose ceiling includes projects:delete still cannot delete a project the user cannot.

    WorkOS's agent checklist limits a delegated agent to what it and its user may both do, checked on every callBI-FIND-2026-1997

›Cited in 34 briefings

Every briefing that discusses this term, newest first.

  1. OAuth's draft for crossing trust domains is approved for publication and never says agent

    Draft · · IETF OAuth Working Group · Arndt Schwenkschuster et al. · Jul 19, 2026

  2. Register a platform's identity provider once, not each agent: trust rests on its issuer, with no revocation

    Draft · · IETF Datatracker · Paul Carleton et al. · Sep 22, 2026

  3. WorkOS's agent checklist limits a delegated agent to what it and its user may both do, checked on every call

    Article · · WorkOS Blog · Maria Paktiti · Sep 30, 2026

  4. Cross-App Access draft warns that one stolen sign-in token could reach every app the identity provider permits

    Draft · · IETF Datatracker · Aaron Parecki et al. · May 21, 2026

  5. IETF workload practices draft describes single-audience JWTs in place of stored secrets and discourages token exchange

    Draft · · IETF Datatracker · Arndt Schwenkschuster et al. · Sep 22, 2026

  6. Niyikiza's agent tokens narrow to a tool's arguments offline; revision -02 breaks -01 implementations

    Draft · · IETF Datatracker · Niki Aimable Niyikiza · Jun 15, 2026

  7. MCP's November 2025 authorization spec puts metadata documents ahead of dynamic client registration

    Article · · Model Context Protocol · Nov 25, 2025

  8. One global registry for AI agent identities, and no one allowed to switch an identity off

    Draft · · IETF Datatracker · Christopher Drake, 1id.com · Oct 27, 2026

  9. A cross-domain transaction-token grant now names AI agents as a use and keeps the call chain at home

    Draft · · IETF Datatracker · George Fletcher et al. · Sep 10, 2026

  10. An OAuth agent profile has the token server block any widening of scope, and leaves revoking the chain optional

    Draft · · IETF Datatracker · Sanjeev Kumar, Orchestrum Technologies LLP · Aug 30, 2026

  11. In Huawei's reworked OAuth draft, evidence signed by the token's subject can narrow access but never widen it

    Draft · · IETF Datatracker · Cheng-Kang Chu et al. · Sep 1, 2026

  12. TPX replaces an app's embedded AI provider key with a capped, metered grant the user approves at their provider

    Article · · tokenpony.dev · Alex Polvi · Jul 20, 2026

  13. PayPal moved MCP tool permission checks into the server, and prompt-injected agents stopped getting through

    Paper · · arXiv · Huan Li et al. · Sep 18, 2026

  14. A phishing agent hosted on Microsoft's own domain, and why giving agents identities does not stop it

    Article · · WorkOS Blog · Maria Paktiti · Sep 29, 2026

  15. A draft token hides some claims, works with old servers unchanged, and leaves agent delegation claims visible

    Draft · · IETF Datatracker · Su-hyeon Forten Lee · Oct 3, 2026

  16. Three MCP auth flaws in a month, and the spec had already written the rule the SDKs skipped

    Article · · WorkOS Blog · Maria Paktiti · Oct 2, 2026

  17. MCP's authorization spec makes the agent a plain OAuth client and leaves delegation to an extension

    Article · · Model Context Protocol project · Jul 28, 2026

  18. The browser OAuth draft favors a backend that keeps tokens out of JavaScript

    Draft · · IETF Datatracker · Aaron Parecki et al. · Jul 6, 2026

  19. In one author's MCP benchmark, a confirmation prompt on every call still let 6 of 10 attacks through

    Paper · · arXiv · Ting Liu, SymbolicLight · Jun 27, 2026

  20. Atlassian's first draft for offline delegation banned bearer handles; revision 05 just recommends sender binding

    Draft · · IETF Datatracker · Larry Zhu et al. · May 22, 2026

  21. A 'decentralized' agent identity draft blocks risky calls when its Registry is down, and did:aip is taken

    Draft · · IETF Datatracker · Paras Singla · Apr 17, 2026

  22. Cisco Duo's MCP beta lets agents register themselves; Duo calls client ID metadata documents the future

    Article · · Cisco Duo · Colin Medfisch · Jan 7, 2026

  23. Huawei-led OAuth draft lets clients chain narrower tokens offline but leaves revocation delivery out of scope

    Draft · · IETF Datatracker · Li Ruochen et al. · Oct 20, 2025

  24. OpenID's credential issuance standard is OAuth for wallets, and assumes a trusted wallet is an app on a phone

    Article · · OpenID Foundation · Torsten Lodderstedt et al. · Sep 16, 2025

  25. The July 2025 Cross App Access draft lets the IdP decide app-to-app access and treats agents as plain clients

    Draft · · IETF Datatracker · Aaron Parecki et al. · Jul 2, 2025

  26. MCP's June 2025 authorization spec: a server accepts only tokens issued for it and passes none upstream

    Article · · Model Context Protocol project · Jun 18, 2025

  27. RFC 9728 tells clients where an API's tokens come from and leaves most of the trust decision to them

    Article · · RFC Editor · Michael B. Jones et al. · Apr 2025

  28. RFC 9700 bans OAuth's password grant and loose redirect matching; binding access tokens stays a SHOULD

    Article · · IETF (RFC Editor) · Torsten Lodderstedt et al. · Jan 2025

  29. In GNAP (RFC 9635), a client is known by its key, and tokens stay key-bound unless the server flags them bearer

    Article · · RFC Editor · Justin Richer et al. · Oct 1, 2024

  30. DPoP token binding depends on checks at the resource server

    Article · · RFC Editor · Daniel Fett et al. · Sep 2023

  31. Rich Authorization Requests need API-specific permission rules

    Article · · RFC Editor · Torsten Lodderstedt et al. · May 1, 2023

  32. RFC 9068 makes resource servers check a JWT access token's type and audience, and warns a client could pose as a user

    Article · · RFC Editor · Vittorio Bertocci, Okta · Oct 1, 2021

  33. OAuth's resource indicators RFC says servers should bind tokens to a resource, but requires it of neither side

    Article · · RFC Editor · Brian Campbell et al. · Feb 1, 2020

  34. The token exchange RFC that agent designs cite says its record of earlier actors is not for access decisions

    Article · · IETF Datatracker · Michael B. Jones et al. · Jan 1, 2020

Related terms

What appears alongside Resource Indicators for OAuth 2.0 in the same briefings.

JWT Profile for OAuth 2.0 Access TokensOAuth 2.0 Protected Resource MetadataOAuth 2.0 Authorization Server MetadataPKCEDynamic Client RegistrationJWT Bearer GrantJSON Web Token Best Current PracticesOAuth Token Introspection

Was this explainer useful?

Share this explainer

Pass it on. Every claim has a permanent number, so what you share today still points at the same words next year.