# What is SCIM?

> SCIM, the System for Cross-domain Identity Management, is a standard web API that lets one system create, read, change and delete user and group accounts in another, using one shared schema instead of a custom connector per application.

Explainer by Boston Identity, 2026-10-30. 6 min read. https://insights.bostonidentity.com/glossary/scim

- What it is: protocol
- Defined by: IETF, RFC 7643 and RFC 7644 (2015)
- Who it's for: Teams that add and remove accounts in SaaS apps

## What is SCIM?

The System for Cross-domain Identity Management (SCIM) is how an enterprise tells a cloud application that a person has joined, moved or left [S3 §2.4]. It runs over HTTP, the web's request protocol, and carries each account as a JSON (JavaScript Object Notation) document [S1 §3.1]. RFC 7643 defines the shared schema: a User, a Group, and an enterprise extension with fields such as department and manager [S2 §4]. RFC 7644 defines the protocol for creating, reading, changing and deleting those records [S1 §3.2].

Before SCIM, the authors write, many cloud providers built their own user-management protocols, so a company that bought from several of them paid for the same integration work again and again [S2 §1]. SCIM seeks to simplify that with a common user schema and one protocol for exchanging it [S2 §1]. In its terms, the client is the system that sends the changes, and the service provider is the application that keeps the accounts [S2 §1.2].

RFC 7642, the use-case document, sets SCIM apart from SAML web single sign-on: SCIM provisions and deprovisions accounts in a separate context from authentication [S3 §1].

Think of it like this: It works like HR sending one standard form to the front desk of every building an employee uses, with an update when they move and a last form when they leave. Each building still decides which doors the badge opens, and SCIM plays no part when someone swipes it.

Not the same as: Not single sign-on. A SAML login proves who someone is at the moment they sign in, while SCIM creates, changes and removes the account itself in a separate exchange [S3 §1].

## How does SCIM work?

Two parties take part. The client sends the changes, often from inside the enterprise; the service provider is the application, often run by another company, that keeps the account [S2 §1.2]. We follow one employee's account from the first day to the last.

[Figure: One account's life in SCIM, from discovery to delete, as requests between the client and the app.]

1. **The client asks what the app supports.** Before it sends any account, the client can read three discovery endpoints: /ServiceProviderConfig, /ResourceTypes and /Schemas [S1 §3.2]. The configuration says whether the app supports features such as PATCH and bulk requests [S2 §5]. The schema lists each attribute and whether it is required, because a provider may support only part of the standard schema [S1 §3.1].
2. **The client creates the account.** When someone joins, the client sends an HTTP POST to /Users with a User record, for example a userName and the client's own externalId [S1 §3.3]. RFC 7642 describes this as the enterprise pushing an account creation request to the cloud provider [S3 §2.4.1]. If the app already holds a User with that userName, it must refuse with 409 Conflict [S1 §3.3].
3. **The app assigns its own id.** On success the app must answer 201 Created and give the record's address in a Location header, and its response should carry the app's copy of the record [S1 §3.3]. The app issues the id, which must be stable and never reassigned, and the client must not set it [S2 §3.1]. The app is free to alter or ignore parts of what was sent, so that copy can help the client compare the two views [S1 §3.3].
4. **The client looks the account up.** Later the client can read the account at /Users/{id} [S1 §3.4.1]. Where the app supports filtering, which is optional, it can also search with a filter such as userName eq "bjensen" [S1 §3.4.2.2]. The externalId lets the client find a record by its own identifier without keeping a mapping table [S2 §3.1].
5. **The client changes the account.** When the person moves to another department, the client pushes an update [S3 §2.4.2]. PUT replaces the record's attributes, and every implementation must support it; PATCH sends only the changes, such as adding a user to a Group's members, and implementations should support it [S1 §3.5]. Setting active to false typically means the account is suspended, but each app defines what the flag means [S2 §4.1.1].
6. **The client deletes the account.** When the person leaves, the client sends DELETE to /Users/{id}, and the app answers 204 No Content [S1 §3.6]. The app may keep the record internally, but it must return 404 Not Found for every later request about it and leave it out of search results [S1 §3.6]. RFC 7642's leaver example suspends the account first and removes it a week later [S3 §2.4.3].

## Why does SCIM matter now?

SCIM dates from 2015, and AI agents have put it back on the agenda. In February, the National Cybersecurity Center of Excellence at the National Institute of Standards and Technology (NIST) listed SCIM among the existing standards it is considering for agent identity [BI-FIND-2026-0686](https://insights.bostonidentity.com/ref/BI-FIND-2026-0686). Tobin South, who co-chairs the OpenID Foundation's AI identity group, holds that web login, SAML, SCIM and the identity provider's handling of agents each need a modest upgrade rather than a new identity system [BI-FIND-2025-0088](https://insights.bostonidentity.com/ref/BI-FIND-2025-0088).

A joint draft by authors from Microsoft, Okta and Nextident makes an AI agent its own SCIM resource at /Agents, with an active flag and an optional list of owners [BI-FIND-2026-0407](https://insights.bostonidentity.com/ref/BI-FIND-2026-0407) [BI-FIND-2026-0409](https://insights.bostonidentity.com/ref/BI-FIND-2026-0409). No SCIM agent draft is yet a working-group document [BI-FIND-2026-0411](https://insights.bostonidentity.com/ref/BI-FIND-2026-0411). The joint draft gives the agent no credential, so removing an agent over SCIM says nothing to the tokens, sessions or keys it holds [BI-FIND-2026-0410](https://insights.bostonidentity.com/ref/BI-FIND-2026-0410).

The base protocol has grown too: RFC 9865 updated RFC 7644 with cursor-based paging [BI-FIND-2025-0539](https://insights.bostonidentity.com/ref/BI-FIND-2025-0539), and RFC 9967 added SCIM events, which report a change and let each receiver choose its own follow-up [BI-FIND-2026-1697](https://insights.bostonidentity.com/ref/BI-FIND-2026-1697).

## Where does it fit?

- Compared with SAML: SAML carries signed statements about a user to an application at sign-in. SCIM manages the account record itself, in a separate exchange from authentication [S3 §1].
- Compared with LDAP: LDAP reads and manages entries in a directory. SCIM's authors drew on LDAP directories among other sources, and they note that older wire protocols do not easily cross firewalls, which is one reason SCIM runs over HTTP [S2 §1].
- Compared with SCIM AI Agent Resource Extension: This individual draft adds an Agent resource type to SCIM at /Agents and keeps the ordinary SCIM lifecycle for it [BI-FIND-2026-0407](https://insights.bostonidentity.com/ref/BI-FIND-2026-0407) [BI-FIND-2026-0409](https://insights.bostonidentity.com/ref/BI-FIND-2026-0409). It carries no credential the agent could use [BI-FIND-2026-0410](https://insights.bostonidentity.com/ref/BI-FIND-2026-0410).

## What goes wrong?

- RFC 7644 says it "does not itself define a SCIM-specific scheme for authentication and authorization", and the provider must map each client to an access control policy [S1 §2]. It depends on TLS (Transport Layer Security) or standard HTTP authentication schemes, and tokens issued on weak or no authentication should not be used, apart from cases such as one-time anonymous registration [S1 §2].
- Filtering, PATCH and bulk requests are optional or only recommended, so a vendor's claim to support SCIM says little until the client reads what the app advertises [BI-FIND-2015-0017](https://insights.bostonidentity.com/ref/BI-FIND-2015-0017).
- Each app decides what a false active flag means [S2 §4.1.1] and what access a Group membership grants [S2 §4.2], so we treat a successful deactivation or group change as something to test against the app's real behaviour [BI-FIND-2015-0038](https://insights.bostonidentity.com/ref/BI-FIND-2015-0038).
- The delete rules say nothing about the removed user's sessions or tokens, and the protocol defines no query for deletions, so a client that missed one has to probe each id it knows or compare full listings [BI-FIND-2015-0018](https://insights.bostonidentity.com/ref/BI-FIND-2015-0018).
- A bulk request carries on past failures, and the RFC's own example returns 200 OK for a request in which every operation failed, so an offboarding job has to check each operation's status [BI-FIND-2015-0016](https://insights.bostonidentity.com/ref/BI-FIND-2015-0016).

## What should a security leader ask?

1. Which of our applications receive accounts over SCIM, and which still depend on someone removing accounts by hand?
2. For each SCIM application, what does setting active to false actually do: block login, end open sessions, or revoke tokens?
3. When a leaver's account is deleted over SCIM, how do we confirm that each application removed access, including sessions and tokens?
4. What does each vendor's /ServiceProviderConfig say about PATCH, filtering and bulk support?
5. If we register AI agents over SCIM, which system issues and revokes the credentials those agents use?

## Sources

- [S1] [System for Cross-domain Identity Management: Protocol (RFC 7644)](https://www.rfc-editor.org/rfc/rfc7644) (IETF (RFC Editor), Proposed Standard, September 2015; updated by RFC 9865)
- [S2] [System for Cross-domain Identity Management: Core Schema (RFC 7643)](https://www.rfc-editor.org/rfc/rfc7643) (IETF (RFC Editor), Proposed Standard, September 2015)
- [S3] [System for Cross-domain Identity Management: Definitions, Overview, Concepts, and Requirements (RFC 7642)](https://www.rfc-editor.org/rfc/rfc7642) (IETF (RFC Editor), Informational, September 2015)

## Evidence

- [BI-FIND-2026-0686](https://insights.bostonidentity.com/ref/BI-FIND-2026-0686): NIST's National Cybersecurity Center of Excellence framed agent identity in February as a question of applying what exists, not inventing something new. Its concept paper lists OAuth, OpenID Connect, SPIFFE, SCIM, the NGAC access-control standard and MCP as candidates it is "currently considering", and names none of the agent-specific IETF work, such as ID-JAG, the WIMSE drafts or token exchange for agents. It keeps human delegation and logging in scope and leaves agents from outside the enterprise for a later effort. The next steps are a possible project description and a lab-built practice guide, with no date; the comment period closed on 2 April. (from "NIST's agent identity work starts from OAuth, SPIFFE and SCIM and defers agents from outside the enterprise")
- [BI-FIND-2025-0088](https://insights.bostonidentity.com/ref/BI-FIND-2025-0088): The co-chair of the OpenID Foundation's AI identity group does not think agents need a new identity system. Tobin South's position is that web login, SAML, SCIM and the identity provider's handling of agents each need a modest upgrade, that no single fix exists, and that anyone who says otherwise is selling a product. He points out that identity providers already let a company declare an agent, tie it to a person and switch it off. The IETF's agent-authentication draft opens with the same complaint, that the field keeps reinventing mechanisms it already has. (from "An OpenID Foundation co-chair says AI agents need modest upgrades to today's identity systems, not a new one")
- [BI-FIND-2026-0407](https://insights.bostonidentity.com/ref/BI-FIND-2026-0407): The competing SCIM proposals for AI agents have collapsed into one draft, which the working group has not yet taken up. WorkOS frames the October 2025 Okta draft by Abbey and Cohen as modelling an agent as a property of an 'agentic application', against Microsoft's Wahl modelling it as its own resource. The Okta draft as filed already defines Agent as its own SCIM resource type, related to an agentic application as owned, authorized or guest, and the SCIM working group's IETF 125 slides list both drafts as representing agents as a resource type. By June 2026 there is a single informational draft, draft-wzdk-scim-agent-resource-00, by Mark Wahl and Pamela Dingle of Microsoft, Danny Zollner of Okta and Ismael Kazzouzi of Nextident, keeping the agent as its own resource with the ordinary SCIM lifecycle. WorkOS calls the joint authorship 'a real signal of forming consensus'. We checked the datatracker: it is an individual submission, not a SCIM working-group document, it replaces neither earlier draft, both of which have expired, and it expires on 7 December 2026. The post's label of the Okta draft as standards track does not match the datatracker, which records no intended status for it. (from "Three companies merged SCIM's two agent drafts, but no working group has taken it up and it expires in December")
- [BI-FIND-2026-0409](https://insights.bostonidentity.com/ref/BI-FIND-2026-0409): The draft makes an AI agent its own SCIM resource type, Agent, at /Agents (URN urn:ietf:params:scim:schemas:core:2.0:Agent) with three required attributes, active, displayName and an agentUserName unique across the provisioning domain, an optional description, and an optional multi-valued owners attribute whose immutable value points at a User, a Group or another Agent. It also amends the RFC 7643 Group members attribute so an agent can join groups much as a user does; roles and entitlements come up only in its introduction. Identity providers and SaaS targets that already speak SCIM would need one new resource type and one changed attribute to carry agents. (from "A joint SCIM draft gives an agent a username, an owner and an on-off switch, and leaves its credentials alone")
- [BI-FIND-2026-0411](https://insights.bostonidentity.com/ref/BI-FIND-2026-0411): There is still no working-group document. The datatracker holds five individual SCIM agent drafts: three expired (Wahl's schema and both Abbey and Cohen filings) and two active: this one, which expires on 7 December 2026, and a separate SCIM Agent Governance Extension filed by Saurabh Kushwaha on 27 July 2026. The SCIM working group heard progress slides at IETF 124 and IETF 125 and has adopted none of them; its header says Informational while the datatracker records no intended status at all. Three vendors on one author list shows the vendors agree; adoption is the working group's call. (from "A joint SCIM draft gives an agent a username, an owner and an on-off switch, and leaves its credentials alone")
- [BI-FIND-2026-0410](https://insights.bostonidentity.com/ref/BI-FIND-2026-0410): The draft provisions an agent's existence and nothing it can use to act. There is no credential, key, client identifier or authentication attribute anywhere in the schema; the abstract says the identity lets the agent "subsequently be authenticated and authorized" and leaves how to other specifications. The only agent-specific lifecycle control is the active flag, whose meaning the service defines, plus the generic SCIM update and delete. So deprovisioning an agent over SCIM tells the target the agent should no longer exist and says nothing to any token, session, key or scheduled job that agent holds. The security and privacy sections are copied from the base SCIM RFCs and add no agent-specific threat. (from "A joint SCIM draft gives an agent a username, an owner and an on-off switch, and leaves its credentials alone")
- [BI-FIND-2025-0539](https://insights.bostonidentity.com/ref/BI-FIND-2025-0539): RFC 9865 formally updates RFC 7644, SCIM's protocol specification, and under it a client can be handed cursors it never asked for. A provider MAY make cursors its only method by returning a nextCursor even when the request carried no cursor parameter. A provider that supports both methods MUST choose a default for requests with no pagination parameter, and the RFC's own ServiceProviderConfig example sets that default to cursor. Only a lowercase should, not a requirement keyword, advises providers adding cursors to existing index-based paging to keep index as the default. Clients MUST NOT read a missing ServiceProviderConfig value as meaning there is no limit on page size or cursor lifetime, or no default method. (from "RFC 9865 publishes SCIM cursor paging with draft -11's rules intact: a cursor still grants no access")
- [BI-FIND-2026-1697](https://insights.bostonidentity.com/ref/BI-FIND-2026-1697): Under RFC 9967 a SCIM event reports a change, and each receiver can choose its own follow-up; the authors say receivers do this "instead of interpreting SETs as commands". The deactivate event says a resource such as a User "has been deactivated and disabled", but its exact meaning SHOULD be agreed between the event publisher and the receiver. The RFC adds only that this typically means the subject may no longer have an active security session. The activate event "does not necessarily reflect any particular state change" either. (from "RFC 9967 lets SCIM report a deactivated account and leaves what that means to each sender and receiver")
- [BI-FIND-2015-0017](https://insights.bostonidentity.com/ref/BI-FIND-2015-0017): Much of what a provisioning client relies on is optional in RFC 7644. Filtering and sorting are OPTIONAL. Implementers MUST support PUT but only SHOULD support PATCH, which the RFC also calls "an OPTIONAL server function." Bulk is optional, ETags are a MAY, and multi-tenancy is OPTIONAL with no scheme defined. The RFC points clients to three discovery endpoints, /ServiceProviderConfig, /Schemas and /ResourceTypes, and says clients SHOULD NOT expect a resource back with exactly the schema and values they sent. Paging is by index and keeps no state, so clients MUST be prepared for inconsistent results; the RFC Editor's page now lists RFC 9865, on cursor-based pagination, as an update. We read this as meaning that a vendor's claim to support SCIM says little until the client reads what the provider advertises. (from "RFC 7644 leaves SCIM's access rules to each deployment, and a bulk request can succeed in part")
- [BI-FIND-2015-0038](https://insights.bostonidentity.com/ref/BI-FIND-2015-0038): RFC 7643 leaves the definitive meaning of a User’s active flag to the service provider. It gives login enabled and account suspended as typical examples for true and false. The Group schema likewise leaves membership semantics and any resulting authorization to the provider, explicitly outside this specification. We therefore read successful provisioning of these attributes as something to test against each application’s access behavior, rather than a complete specification of that behavior. (from "SCIM leaves the meaning of active status and group access to each provider")
- [BI-FIND-2015-0018](https://insights.bostonidentity.com/ref/BI-FIND-2015-0018): A deleted account drops out of view under RFC 7644. A provider MAY keep a deleted resource rather than erase it, but MUST answer 404 for every operation on it and MUST leave it out of future query results. It SHOULD NOT let the old record block a new user created with the same userName. A client can filter on meta.lastModified to find changed resources where the provider supports filtering, but the RFC defines no query for deletions. We read that as leaving a client two ways to learn of a deletion it missed: probe each id it knows and get a 404, or compare full listings. Our briefing on the 2023 SCIM delta-query draft found its authors treating a removal that never syncs as an access problem. Section 3.6, on deleting, says nothing about the deleted user's sessions or tokens. (from "RFC 7644 leaves SCIM's access rules to each deployment, and a bulk request can succeed in part")
- [BI-FIND-2015-0016](https://insights.bostonidentity.com/ref/BI-FIND-2015-0016): Under RFC 7644 a PATCH is all or nothing and a bulk request is not. A PATCH "SHALL be treated as atomic": if one operation fails, the original resource MUST be restored. In a bulk request the service provider MUST carry on past partial failures. The client can override that with failOnErrors, the number of errors after which the provider stops and fails the remaining operations; the response then carries the error and any operations that succeeded before it. The RFC says a bulk job processed successfully gets HTTP 200 OK, and its own example of a request in which every operation failed still returns 200 OK, so the top-level status does not show whether the operations worked. The provider MAY also change the order of operations, as long as the result is the same, and it must try to resolve circular references between them but may stop with a 409. We read this as a risk for offboarding jobs sent in bulk: a request that removes group memberships and then deletes the account can apply in part, so the client has to check the status of every operation, not the 200 alone. (from "RFC 7644 leaves SCIM's access rules to each deployment, and a bulk request can succeed in part")

---

© 2026 Boston Identity LLC. All rights reserved. Published by Boston Identity LLC.
Quote up to 100 words with the credit "Boston Identity, IAM Insights" and a link to https://insights.bostonidentity.com/glossary/scim. Claim numbers may be cited freely. Training, fine-tuning and text-and-data-mining use of this page is permitted.
Terms: https://insights.bostonidentity.com/terms · How to cite: https://insights.bostonidentity.com/cite
