Microsoft Entra Agent ID key concepts
In brief
The article now explains single- and multi-instance blueprints, token relationships, required identifiers, and blueprint principal creation. It notes that Microsoft Graph does not create a multi-instance blueprint principal automatically.
What Entra admins need to know
Use the correct tenant, client, and object IDs. When creating a multi-instance blueprint through Microsoft Graph, create its blueprint principal separately.
This editorial summary was generated by AI from the documentation changes. Verify important details in the full Microsoft Learn article.
Documentation change
The comparison below shows only the changed extract. Use the full-page view for complete context.
Microsoft Entra Agent ID key concepts
The MicrosoftAn agent identity platform provides specializedlets an AI agent authenticate to APIs and other resources. Its agent identity constructs designed specifically for AI agents operating in enterprise environments. Theseblueprint holds authentication settings and credentials, while the agent identity constructs enable secure authenticationreceives access to resources. For example, an agent that reads a protected API needs a blueprint, an agent identity, and authorization patternspermission to call that differ from traditional user and application identities, addressing the unique requirements of autonomous AI systems.API.
This article explains those objects, the core concepts that formpeople who manage them, and the foundation ofidentifiers you record during setup. You don't need to create every object described here: follow Choose an agent integration path for your scenario.
An agent identity management:blueprint defines the configuration for one or more agent identities, agent identity blueprints,identities. In multi-instance mode, it provides a template for separate child identities. In single-instance mode, the blueprint and their supporting components. Understanding these concepts is essential for developers who need to implement secure, scalable authentication patterns for AI agents.
The agent identity architecture follows a hierarchical model where agent identity blueprints serve as templates for creating multiple agent instances, eachits one Agent ID share an OAuth client ID, with distinct identities and capabilities. This approach enables centralized management while providing the flexibility needed for diverse AI agent deployment scenarios.no blueprint principal.
Core identity concepts
Agent identity
An agent identity is the primary identity an AI agent uses to authenticate to systems and access resources. Unlike user accounts, agent identitiesCredentials, when needed, are configured on its blueprint rather than on the identity. Multi-instance Agent IDs authenticate through blueprint impersonation. Single-instance Agent IDs request tokens directly with the shared client ID; public clients don't have credentials of their own. They authenticate using tokens issued by their agent identity blueprint.send a client credential. For more information, see Agent identities.
Agent identity blueprint
An agent identity blueprint is an object in Microsoft Entra ID that serves as the templatedefines authentication settings and authentication foundationother configuration for one or more agent identities. The blueprint holds credentialsIts instance mode determines the identity relationship and uses them to acquire tokens on behalf of all agent identities created from it.token acquisition pattern. Policies and settings applied to a blueprint, such as Conditional Access, take effect for all its agent identities. For more information, see Agent identity blueprints.
Agent identity blueprint principal
When aA multi-instance blueprint is added to a tenant, Microsoft Entra createsneeds a corresponding principal object. An agent identity blueprint principal isin the tenant. Managed creation experiences can create both objects, but creating the blueprint through Microsoft EntraGraph doesn't create its principal. Follow Create an agent blueprint principal when you provision it yourself. This object that records athe blueprint's presence in athe tenant and enables it to acquire tokens and appear in audit logs. For more information, see Agent identity blueprint principals.A single-instance blueprint doesn't have a blueprint principal.
Traditional service principal (not recommended for AI agents)
The Microsoft Entra ID Auth SDK (sidecar) is a containerized web service that handles token acquisition, validation, and secure downstream API calls for agents registered in the Microsoft identity platform. It runs as a companion container alongside your application, allowing you to offload identity logic to a dedicated service. For more information, see Microsoft Entra ID Auth SDK (sidecar).
Record the identifiers for your scenario
Record these values as you complete setup. Use only the rows your tutorial needs; you don't need to create every object in these tables. An application (client) ID identifies an OAuth client or resource. An object ID identifies one directory object in one tenant. They aren't interchangeable, even when both are GUIDs.
Tenant and management client
The management client provisions and configures identities in the selected tenant. Its client ID isn't the agent's runtime client ID.
| Agent ID identifier | Identifier source | Required operation or field |
|---|---|---|
| Tenant ID | Your directory's Tenant ID. | The authority and the tenant targeted by management requests. |
| Management client ID | The separate management app's Application (client) ID. | Connect-MgGraph -ClientId. This isn't the agent's runtime client ID. |
Blueprint and agent identities
Use object IDs for directory operations and client IDs for OAuth requests. Single-instance provisioning uses the blueprint object ID; fleet provisioning uses its client ID.
| Agent ID identifier | Identifier source | Required operation or field |
|---|---|---|
| Blueprint object ID | id in the blueprint creation response. | Microsoft Graph /applications/{id} updates, including credentials, redirect URIs, and scopes. Single-instance provisioning also uses this value for agentIdentityBlueprintId. |
| Blueprint client ID | appId in the blueprint creation response. | Blueprint authentication and its api:// identifier URI. Fleet provisioning uses this value for agentIdentityBlueprintId. |
| Blueprint principal object ID | id in the blueprint principal creation response. Fleet scenarios only. | Operations on the blueprint's tenant service principal, including configured permission inheritance. Don't create this object for a single-instance agent. |
| Agent object ID | id in the agent identity creation response. | Microsoft Graph /servicePrincipals/{id}, assignments, and oauth2PermissionGrants.clientId. Despite that grant field's name, it takes the agent's service principal object ID, not its OAuth client ID. |
| Agent client ID | appId on the agent identity. | Runtime OAuth requests. A cloud or local single-instance agent shares its blueprint's client ID; a fleet identity has a separate client ID. |
Resources and permissions
The resource client ID identifies the API. Permission grants use its service principal object ID; scope and role GUIDs identify the permissions within that resource.
| Agent ID identifier | Identifier source | Required operation or field |
|---|---|---|
| Resource client ID | The downstream API application's appId. | Token audience, scope URI, and requiredResourceAccess.resourceAppId. |
| Resource service principal object ID | id of that API's service principal in the grant tenant. | resourceId in permission grants. Don't substitute the resource client ID. |
| Delegated scope ID and value | The API's oauth2PermissionScopes[].id and .value. | The GUID goes in permission declarations; the value, such as read_data, goes in a scope request or delegated grant. |
| App role ID | The resource's appRoles[].id. | Application-permission declarations and app role assignments. |
Human users and dedicated agent users
Both values are directory object IDs, not client IDs. A dedicated agent-user account doesn't replace the human user in an OBO request.
| Agent ID identifier | Identifier source | Required operation or field |
|---|---|---|
| Human user object ID | The signed-in user's directory id. | User-specific consent and ownership or sponsorship when the procedure requires a user. |
| Agent-user object ID | id returned when you create the optional dedicated agent user. | Dedicated-agent-user operations, not human-user OBO requests. |
Keep credentials separate from this worksheet. Never record access tokens, client secrets, private keys, or PFX passwords in source control or shared troubleshooting notes.
Next steps
- Choose an integration path for your agent.
- Create an identity for a cloud or local agent.
- Set up a blueprint for an agent fleet.
Related content
@@ -3,17 +3,18 @@ title: Fundamental concepts in Microsoft Entra Agent ID titleSuffix: Microsoft Entra Agent ID description: Discover the role of agent identities in AI authentication. Understand their unique identifiers, token usage, and how they enable secure access to systems. ms.reviewer: dastrock-ms.date: 04/03/2026+ms.date: 11/06/2026 ms.topic: concept-article+ai-usage: ai-assisted #customer-intent: As a developer, I want to understand the core concepts of agent identities and blueprints in Microsoft Entra ID so that I can implement secure authentication patterns for AI agents in my applications. --- # Microsoft Entra Agent ID key concepts -The Microsoft agent identity platform provides specialized identity constructs designed specifically for AI agents operating in enterprise environments. These identity constructs enable secure authentication and authorization patterns that differ from traditional user and application identities, addressing the unique requirements of autonomous AI systems.+An agent identity lets an AI agent authenticate to APIs and other resources. Its agent identity blueprint holds authentication settings and credentials, while the agent identity receives access to resources. For example, an agent that reads a protected API needs a blueprint, an agent identity, and permission to call that API. -This article explains the core concepts that form the foundation of agent identity management: agent identities, agent identity blueprints, and their supporting components. Understanding these concepts is essential for developers who need to implement secure, scalable authentication patterns for AI agents.+This article explains those objects, the people who manage them, and the identifiers you record during setup. You don't need to create every object described here: follow [Choose an agent integration path](choose-agent-integration-path.md) for your scenario. -The agent identity architecture follows a hierarchical model where agent identity blueprints serve as templates for creating multiple agent instances, each with distinct identities and capabilities. This approach enables centralized management while providing the flexibility needed for diverse AI agent deployment scenarios.+An agent identity blueprint defines the configuration for one or more agent identities. In multi-instance mode, it provides a template for separate child identities. In single-instance mode, the blueprint and its one Agent ID share an OAuth client ID, with no blueprint principal. ## Core identity concepts @@ -21,15 +22,15 @@ The following concepts form the foundation of Microsoft Entra Agent ID and the M ### Agent identity -An agent identity is the primary identity an AI agent uses to authenticate to systems and access resources. Unlike user accounts, agent identities don't have credentials of their own. They authenticate using tokens issued by their agent identity blueprint. For more information, see [Agent identities](agent-identities.md).+An agent identity is the primary identity an AI agent uses to authenticate to systems and access resources. Credentials, when needed, are configured on its blueprint rather than on the identity. Multi-instance Agent IDs authenticate through blueprint impersonation. Single-instance Agent IDs request tokens directly with the shared client ID; public clients don't send a client credential. For more information, see [Agent identities](agent-identities.md). ### Agent identity blueprint -An agent identity blueprint is an object in Microsoft Entra ID that serves as the template and authentication foundation for one or more agent identities. The blueprint holds credentials and uses them to acquire tokens on behalf of all agent identities created from it. Policies and settings applied to a blueprint, such as Conditional Access, take effect for all its agent identities. For more information, see [Agent identity blueprints](agent-blueprint.md).+An agent identity blueprint is an object in Microsoft Entra ID that defines authentication settings and other configuration for one or more agent identities. Its [instance mode](agent-blueprint.md#instance-modes-preview) determines the identity relationship and token acquisition pattern. Policies and settings applied to a blueprint, such as Conditional Access, take effect for its agent identities. For more information, see [Agent identity blueprints](agent-blueprint.md). ### Agent identity blueprint principal -When a blueprint is added to a tenant, Microsoft Entra creates a corresponding principal object. An agent identity blueprint principal is the Microsoft Entra object that records a blueprint's presence in a tenant and enables it to acquire tokens and appear in audit logs. For more information, see [Agent identity blueprint principals](agent-blueprint.md#agent-identity-blueprint-principals).+A multi-instance blueprint needs a corresponding blueprint principal in the tenant. Managed creation experiences can create both objects, but creating the blueprint through Microsoft Graph doesn't create its principal. Follow [Create an agent blueprint principal](create-blueprint.md#create-an-agent-blueprint-principal) when you provision it yourself. This object records the blueprint's presence in the tenant and enables it to acquire tokens and appear in audit logs. A single-instance blueprint doesn't have a blueprint principal. ### Traditional service principal (not recommended for AI agents) @@ -59,6 +60,59 @@ For more information, see [Administrative relationships for agent identities (Ow The Microsoft Entra ID Auth SDK (sidecar) is a containerized web service that handles token acquisition, validation, and secure downstream API calls for agents registered in the Microsoft identity platform. It runs as a companion container alongside your application, allowing you to offload identity logic to a dedicated service. For more information, see [Microsoft Entra ID Auth SDK (sidecar)](/entra/msidweb/agent-id-sdk/overview). +## Record the identifiers for your scenario++Record these values as you complete setup. Use only the rows your tutorial needs; you don't need to create every object in these tables. An application (client) ID identifies an OAuth client or resource. An object ID identifies one directory object in one tenant. They aren't interchangeable, even when both are GUIDs.++### Tenant and management client++The management client provisions and configures identities in the selected tenant. Its client ID isn't the agent's runtime client ID.++| Agent ID identifier | Identifier source | Required operation or field |+| --- | --- | --- |+| Tenant ID | Your directory's **Tenant ID**. | The authority and the tenant targeted by management requests. |+| Management client ID | The separate management app's **Application (client) ID**. | `Connect-MgGraph -ClientId`. This isn't the agent's runtime client ID. |++### Blueprint and agent identities++Use object IDs for directory operations and client IDs for OAuth requests. Single-instance provisioning uses the blueprint object ID; fleet provisioning uses its client ID.++| Agent ID identifier | Identifier source | Required operation or field |+| --- | --- | --- |+| Blueprint object ID | `id` in the blueprint creation response. | Microsoft Graph `/applications/{id}` updates, including credentials, redirect URIs, and scopes. Single-instance provisioning also uses this value for `agentIdentityBlueprintId`. |+| Blueprint client ID | `appId` in the blueprint creation response. | Blueprint authentication and its `api://` identifier URI. Fleet provisioning uses this value for `agentIdentityBlueprintId`. |+| Blueprint principal object ID | `id` in the blueprint principal creation response. Fleet scenarios only. | Operations on the blueprint's tenant service principal, including configured permission inheritance. Don't create this object for a single-instance agent. |+| Agent object ID | `id` in the agent identity creation response. | Microsoft Graph `/servicePrincipals/{id}`, assignments, and `oauth2PermissionGrants.clientId`. Despite that grant field's name, it takes the agent's service principal object ID, not its OAuth client ID. |+| Agent client ID | `appId` on the agent identity. | Runtime OAuth requests. A cloud or local single-instance agent shares its blueprint's client ID; a fleet identity has a separate client ID. |++### Resources and permissions++The resource client ID identifies the API. Permission grants use its service principal object ID; scope and role GUIDs identify the permissions within that resource.++| Agent ID identifier | Identifier source | Required operation or field |+| --- | --- | --- |+| Resource client ID | The downstream API application's `appId`. | Token audience, scope URI, and `requiredResourceAccess.resourceAppId`. |+| Resource service principal object ID | `id` of that API's service principal in the grant tenant. | `resourceId` in permission grants. Don't substitute the resource client ID. |+| Delegated scope ID and value | The API's `oauth2PermissionScopes[].id` and `.value`. | The GUID goes in permission declarations; the value, such as `read_data`, goes in a scope request or delegated grant. |+| App role ID | The resource's `appRoles[].id`. | Application-permission declarations and app role assignments. |++### Human users and dedicated agent users++Both values are directory object IDs, not client IDs. A dedicated agent-user account doesn't replace the human user in an OBO request.++| Agent ID identifier | Identifier source | Required operation or field |+| --- | --- | --- |+| Human user object ID | The signed-in user's directory `id`. | User-specific consent and ownership or sponsorship when the procedure requires a user. |+| Agent-user object ID | `id` returned when you create the optional dedicated agent user. | Dedicated-agent-user operations, not human-user OBO requests. |++Keep credentials separate from this worksheet. Never record access tokens, client secrets, private keys, or PFX passwords in source control or shared troubleshooting notes.++## Next steps++- [Choose an integration path for your agent](choose-agent-integration-path.md).+- [Create an identity for a cloud or local agent](create-single-instance-agent.md).+- [Set up a blueprint for an agent fleet](set-up-blueprint-agent-fleet.md).+ ## Related content - [What is an agent ID?](what-are-agent-identities.md) 