entra-agent-id

作者: microsoft

使用 Microsoft Graph beta API 建立和管理支援 OAuth2 的 AI 代理身分識別。

npx skills add https://github.com/microsoft/agent-skills --skill entra-agent-id

Microsoft Entra Agent ID

Create and manage OAuth2-capable identities for AI agents using Microsoft Graph beta API.

Preview API — All Agent Identity endpoints are under /beta only. Not available in /v1.0.

Before You Start

Search microsoft-docs MCP for the latest Agent ID documentation:

  • Query: "Microsoft Entra agent identity setup"
  • Verify: API parameters match current preview behavior

Conceptual Model

Agent Identity Blueprint (application)        ← one per agent type/project
  └── BlueprintPrincipal (service principal)   ← MUST be created explicitly
        ├── Agent Identity (SP): agent-1       ← one per agent instance
        ├── Agent Identity (SP): agent-2
        └── Agent Identity (SP): agent-3

Prerequisites

PowerShell (recommended for interactive setup)

# Requires PowerShell 7+
Install-Module Microsoft.Graph.Beta.Applications -Scope CurrentUser -Force

Python (for programmatic provisioning)

pip install azure-identity requests

Required Entra Roles

One of: Agent Identity Developer, Agent Identity Administrator, or Application Administrator.

Environment Variables

AZURE_TENANT_ID=<your-tenant-id>
AZURE_CLIENT_ID=<app-registration-client-id>
AZURE_CLIENT_SECRET=<app-registration-secret>

Authentication

⚠️ DefaultAzureCredential is NOT supported. Azure CLI tokens contain Directory.AccessAsUser.All, which Agent Identity APIs explicitly reject (403). You MUST use a dedicated app registration with client_credentials flow or connect via Connect-MgGraph with explicit delegated scopes.

PowerShell (delegated permissions)

Connect-MgGraph -Scopes @(
    "AgentIdentityBlueprint.Create",
    "AgentIdentityBlueprint.ReadWrite.All",
    "AgentIdentityBlueprintPrincipal.Create",
    "User.Read"
)
Set-MgRequestContext -ApiVersion beta

$currentUser = (Get-MgContext).Account
$userId = (Get-MgUser -UserId $currentUser).Id

Python (application permissions)

import os
import requests
from azure.identity import ClientSecretCredential

credential = ClientSecretCredential(
    tenant_id=os.environ["AZURE_TENANT_ID"],
    client_id=os.environ["AZURE_CLIENT_ID"],
    client_secret=os.environ["AZURE_CLIENT_SECRET"],
)
token = credential.get_token("https://graph.microsoft.com/.default")

GRAPH = "https://graph.microsoft.com/beta"
headers = {
    "Authorization": f"Bearer {token.token}",
    "Content-Type": "application/json",
    "OData-Version": "4.0",  # Required for all Agent Identity API calls
}

Core Workflow

Step 1: Create Agent Identity Blueprint

Sponsors are required and must be User objects — ServicePrincipals and Groups are rejected.

import subprocess

# Get sponsor user ID (client_credentials has no user context, so use az CLI)
result = subprocess.run(
    ["az", "ad", "signed-in-user", "show", "--query", "id", "-o", "tsv"],
    capture_output=True, text=True, check=True,
)
user_id = result.stdout.strip()

blueprint_body = {
    "@odata.type": "Microsoft.Graph.AgentIdentityBlueprint",
    "displayName": "My Agent Blueprint",
    "[email protected]": [
        f"https://graph.microsoft.com/beta/users/{user_id}"
    ],
}
resp = requests.post(f"{GRAPH}/applications", headers=headers, json=blueprint_body)
resp.raise_for_status()

blueprint = resp.json()
app_id = blueprint["appId"]
blueprint_obj_id = blueprint["id"]

Step 2: Create BlueprintPrincipal

This step is mandatory. Creating a Blueprint does NOT auto-create its service principal. Without this, Agent Identity creation fails with: 400: The Agent Blueprint Principal for the Agent Blueprint does not exist.

sp_body = {
    "@odata.type": "Microsoft.Graph.AgentIdentityBlueprintPrincipal",
    "appId": app_id,
}
resp = requests.post(f"{GRAPH}/servicePrincipals", headers=headers, json=sp_body)
resp.raise_for_status()

If implementing idempotent scripts, check for and create the BlueprintPrincipal even when the Blueprint already exists (a previous run may have created the Blueprint but crashed before creating the SP).

Step 3: Create Agent Identities

agent_body = {
    "@odata.type": "Microsoft.Graph.AgentIdentity",
    "displayName": "my-agent-instance-1",
    "agentIdentityBlueprintId": app_id,
    "[email protected]": [
        f"https://graph.microsoft.com/beta/users/{user_id}"
    ],
}
resp = requests.post(f"{GRAPH}/servicePrincipals", headers=headers, json=agent_body)
resp.raise_for_status()
agent = resp.json()

API Reference

OperationMethodEndpointOData Type
Create BlueprintPOST/applicationsMicrosoft.Graph.AgentIdentityBlueprint
Create BlueprintPrincipalPOST/servicePrincipalsMicrosoft.Graph.AgentIdentityBlueprintPrincipal
Create Agent IdentityPOST/servicePrincipalsMicrosoft.Graph.AgentIdentity
List Agent IdentitiesGET/servicePrincipals?$filter=...
Delete Agent IdentityDELETE/servicePrincipals/{id}
Delete BlueprintDELETE/applications/{id}

All endpoints use base URL: https://graph.microsoft.com/beta

Required Permissions

PermissionPurpose
Application.ReadWrite.AllBlueprint CRUD (application objects)
AgentIdentityBlueprint.CreateCreate new Blueprints
AgentIdentityBlueprint.ReadWrite.AllRead/update Blueprints
AgentIdentityBlueprintPrincipal.CreateCreate BlueprintPrincipals
AgentIdentity.Create.AllCreate Agent Identities
AgentIdentity.ReadWrite.AllRead/update Agent Identities

There are 18 Agent Identity-specific Graph application permissions. Discover all:

az ad sp show --id 00000003-0000-0000-c000-000000000000 \
  --query "appRoles[?contains(value, 'AgentIdentity')].{id:id, value:value}" -o json

Grant admin consent (required for application permissions):

az ad app permission admin-consent --id <client-id>

Admin consent may fail with 404 if the service principal hasn't replicated. Retry with 10–40s backoff.

Cleanup

# Delete Agent Identity
requests.delete(f"{GRAPH}/servicePrincipals/{agent['id']}", headers=headers)

# Delete BlueprintPrincipal (get SP ID first)
sps = requests.get(
    f"{GRAPH}/servicePrincipals?$filter=appId eq '{app_id}'",
    headers=headers,
).json()
for sp in sps.get("value", []):
    requests.delete(f"{GRAPH}/servicePrincipals/{sp['id']}", headers=headers)

# Delete Blueprint
requests.delete(f"{GRAPH}/applications/{blueprint_obj_id}", headers=headers)

Best Practices

  1. Always create BlueprintPrincipal after Blueprint — not auto-created; implement idempotent checks on both
  2. Use User objects as sponsors — ServicePrincipals and Groups are rejected
  3. Handle permission propagation delays — after admin consent, wait 30–120s; retry with backoff on 403
  4. Include OData-Version: 4.0 header on every Graph request
  5. Use Workload Identity Federation for production auth — for local dev, use a client secret on the Blueprint (see references/oauth2-token-flow.md)
  6. Set identifierUris on Blueprint before using OAuth2 scoping (api://{app-id})
  7. Never use Azure CLI tokens for API calls — they contain Directory.AccessAsUser.All which is hard-rejected
  8. Check for existing resources before creating — implement idempotent provisioning

References

FileContents
references/oauth2-token-flow.mdProduction (Managed Identity + WIF) and local dev (client secret) token flows
references/known-limitations.md29 known issues organized by category (from official preview known-issues page)
references/sdk-sidecar.mdMicrosoft Entra SDK for AgentID — endpoints, 3P agent patterns, Docker/K8s deployment, security

External Links

ResourceURL
Official Setup Guidehttps://learn.microsoft.com/en-us/entra/agent-id/identity-platform/agent-id-setup-instructions
AI-Guided Setuphttps://learn.microsoft.com/en-us/entra/agent-id/identity-platform/agent-id-ai-guided-setup
Microsoft Entra SDK for AgentID — Overviewhttps://learn.microsoft.com/en-us/entra/msidweb/agent-id-sdk/overview
Microsoft Entra SDK for AgentID — Endpointshttps://learn.microsoft.com/en-us/entra/msidweb/agent-id-sdk/endpoints

來自 microsoft 的更多技能

oss-growth
microsoft
開源增長駭客角色
official
microsoft-foundry
microsoft
端到端部署、評估與管理 Foundry 代理:Docker 建置、ACR 推送、託管/提示代理建立、容器啟動、批次評估、持續評估、提示最佳化工作流程、agent.yaml、從追蹤資料集整理。用途:將代理部署至 Foundry、託管代理、建立代理、調用代理、評估代理、執行批次評估、持續評估、持續監控、持續評估狀態、最佳化提示、改善提示、提示最佳化器、最佳化代理指令、改善代理...
officialdevelopmentdevops
azure-ai
microsoft
用於 Azure AI:搜尋、語音、OpenAI、文件智慧。協助搜尋、向量/混合搜尋、語音轉文字、文字轉語音、轉錄、OCR。適用情境:AI 搜尋、查詢搜尋、向量搜尋、混合搜尋、語意搜尋、語音轉文字、文字轉語音、轉錄、OCR、將文字轉換為語音。
officialdevelopmentapi
azure-deploy
microsoft
對已準備好的應用程式執行 Azure 部署,這些應用程式需具備現有的 .azure/deployment-plan.md 與基礎架構檔案。當使用者要求建立新應用程式時,請勿使用此技能——應改用 azure-prepare。此技能會執行 azd up、azd deploy、terraform apply 及 az deployment 命令,並內建錯誤復原機制。需具備來自 azure-prepare 的 .azure/deployment-plan.md,以及來自 azure-validate 的驗證狀態。適用時機:「執行 azd up」、「執行 azd deploy」、「執行部署」……
officialdevopsaws
azure-storage
microsoft
Azure Storage Services 包括 Blob 儲存體、檔案共用、佇列儲存體、表格儲存體和 Data Lake。回答關於儲存存取層(熱、冷、凍結、封存)、各層使用時機及層級比較的問題。提供物件儲存、SMB 檔案共用、非同步訊息、NoSQL 鍵值及大數據分析。包含生命週期管理。用於:blob 儲存體、檔案共用、佇列儲存體、表格儲存體、data lake、上傳檔案、下載 blob、儲存帳戶、存取層...
officialdevelopmentdatabase
azure-diagnostics
microsoft
在 Azure 上使用 AppLens、Azure Monitor、資源健康狀態和安全分類來偵錯 Azure 生產問題。適用時機:偵錯生產問題、疑難排解應用程式服務、應用程式服務高 CPU、應用程式服務部署失敗、疑難排解容器應用程式、疑難排解函數、疑難排解 AKS、kubectl 無法連線、kube-system/CoreDNS 失敗、Pod 擱置、CrashLoop、節點未就緒、升級失敗、分析記錄、KQL、深入解析、映像提取失敗、冷啟動問題、健康狀態探查失敗...
officialdevopsdevelopment
azure-prepare
microsoft
準備 Azure 應用程式以進行部署(基礎架構 Bicep/Terraform、azure.yaml、Dockerfile)。用於建立/現代化或建立+部署;不適用於跨雲端遷移(請使用 azure-cloud-migrate)。請勿用於:copilot-sdk 應用程式(請使用 azure-hosted-copilot-sdk)。適用時機:「建立應用程式」、「建置 Web 應用程式」、「建立 API」、「建立無伺服器 HTTP API」、「建立前端」、「建立後端」、「建置服務」、「現代化應用程式」、「更新應用程式」、「新增驗證」、「新增快取」、「託管於 Azure」、「建立並...」
officialdevelopmentdevops
azure-validate
microsoft
部署前驗證 Azure 就緒狀態。對設定、基礎架構(Bicep 或 Terraform)、RBAC 角色指派、受控身分權限及先決條件進行深度檢查,再進行部署。適用時機:驗證我的應用程式、檢查部署就緒狀態、執行預檢檢查、驗證設定、確認是否可部署、驗證 azure.yaml、驗證 Bicep、部署前測試、疑難排解部署錯誤、驗證 Azure Functions、驗證函式應用程式、驗證無伺服器...
officialdevopstesting