Agent Platform Agent Registry

google/skills/skills/cloud/agent-platform-agent-registry

作者 google5fb62c2cf7bc7eda841109c3470d132e87055e23無授權條款24K 個星標收錄於 2026年11月20日更新於 2026年11月20日儲存庫今天更新

Manages Google Cloud Agent Registry (agentregistry.googleapis.com) resources using gcloud CLI and Kubernetes manifests, including agents, MCP servers, endpoints, services, bindings, AI applications, skills and skill revisions, resource-level IAM policies, Cloud Run and GKE automatic workload registration, Cloud Resource Manager management boundaries, 5-endpoint Quick Add for Agent Gateway, and Python ADK integration. Use when registering, discovering, searching, filtering, or governing agents, MCP servers, endpoints, skills, AI applications, or management boundaries in Google Cloud Agent Registry. Don't use for the preview AI Platform Skill Registry Python scripts (use agent-platform-skill-registry), deploying raw Vertex AI Agent Engine code (use agent-platform-deploy), or runtime trace debugging (use agent-platform-troubleshooting).

AI 產生的概覽

透過 gcloud、Kubernetes 清單與 Python ADK 管理 Google Cloud Agent Registry 資源,包含代理、MCP 伺服器、技能與管理邊界。

功能
引導代理使用 gcloud agent-registry 指令註冊、列出、檢視、更新與刪除 Agent Registry 資源,涵蓋代理、MCP 伺服器、端點、服務、繫結、技能及技能修訂版本。也涵蓋 Cloud Run 與 GKE 工作負載的自動註冊、供 Agent Gateway 使用的複合 Quick Add 服務、資源層級 IAM 政策,以及 Cloud Resource Manager 管理邊界。附有 GKE 註解指令碼與 Quick Add 輔助指令碼,並在執行變更類指令前先請求確認。
適用情境
適用於在 Google Cloud Agent Registry 中註冊、探索、搜尋、篩選或治理代理、MCP 伺服器、端點、技能、AI 應用程式或管理邊界。不適用於預覽版 Skill Registry 的 Python 指令碼、直接部署 Vertex AI Agent Engine 程式碼,或執行階段追蹤除錯。
執行需求
需要 Google Cloud SDK gcloud 560.0.0 以上版本(Cloud Run 自動註冊需 beta 2026.03.09+)、已驗證的 gcloud 工作階段或應用程式預設憑證、已啟用帳單的專案,以及 agentregistry.admin、editor、viewer 或 user 等 IAM 角色。需要可存取 Google Cloud API 的網路;ADK 整合可選用 Python ADK(google-adk)。附帶兩個可執行指令碼:annotate_gke.py 與 quick-add-gcp-service.sh。
<!-- disableFinding(LINE_OVER_80) -->

Google Cloud Agent Registry

Interact with Google Cloud Agent Registry using gcloud agent-registry (agents, mcp-servers, services, endpoints, bindings, skills, findings, publishers, ai-applications, operations), gcloud beta run (--functional-type and --identity-type for automatic registration of Cloud Run services and jobs), Kubernetes manifests and scripts/annotate_gke.py [blocked] (GKE automatic registration), gcloud resource-manager (boundary-configs, capability-configs, boundaries, and tags bindings for management boundaries), and Python ADK (google-adk).

Usage Attribution

MUST READ: Before executing any gcloud, REST API, or MCP command, you MUST read and follow the instructions in Usage Attribution [blocked].

Workflow

  1. Resolve session context — at the start of each session, silently run:
    bash
    gcloud config get-value project 2>/dev/nullgcloud config get-value compute/region 2>/dev/null
    Store the results as SESSION_PROJECT and SESSION_LOCATION and use them as defaults.
  2. Parse the user's request to identify the target resource and action.
  3. Construct the gcloud command or Kubernetes manifest using session defaults. See gcloud Command Reference [blocked] for full flag synopses, command summary tables, and 60+ natural-language-to-command examples.
  4. Documentation Fallback — if a user asks about a feature, flag, schema, or error message not covered in this skill, query the developerknowledge MCP server or public .md.txt documentation (see Documentation Fallback).
  5. Approval:
    • Read-only commands (list, describe, search, get-iam-policy): Execute immediately without asking for approval.
    • Mutating commands (create, delete, update, deploy, set-iam-policy, add-iam-policy-binding, remove-iam-policy-binding): Display the exact command and ask: "Ready to run this command? (yes/no)"
    • Multi-task requests: Always output the commands for every requested task in the same response before asking for approval.
  6. Execute and display the output.

Prerequisites, Auth, Billing, Region & IAM Permissions

  • Cloud SDK: Requires gcloud 560.0.0+ (beta 2026.03.09+ for Cloud Run automatic registration; ensure gcloud is up to date for GA gcloud agent-registry skills). Do not install or update tools or packages on the host without explicit user approval; if gcloud is missing, share https://docs.cloud.google.com/sdk/docs/install-sdk.md.txt and fall back to .md.txt documentation.
  • Billing & Pricing: The target project must be linked to an active Google Cloud Billing account. For Cloud Run, GKE, or Vertex AI pricing, direct users to https://cloud.google.com/pricing.
  • Region & Location Availability: Configure a default region (gcloud config set compute/region us-central1) and verify supported regions at https://docs.cloud.google.com/agent-registry/locations.md.txt and https://cloud.google.com/about/locations (note: Skills only support global, us, or eu).
  • Authentication: Check active credentials first with gcloud auth list. Because gcloud auth login and gcloud auth application-default login launch interactive browser flows and stop responding in non-interactive shells, never execute them directly — if unauthenticated, ask the user to run gcloud auth login (for CLI sessions) or gcloud auth application-default login (for local Python ADK Application Default Credentials) in their own terminal, or configure GOOGLE_APPLICATION_CREDENTIALS.
bash
# Check active authentication status first (read-only)gcloud auth list
# If unauthenticated, ask the user to run these interactively in their terminal:# gcloud auth login# gcloud auth application-default logingcloud config set project PROJECT_IDgcloud config set compute/region us-central1gcloud services enable agentregistry.googleapis.com
RoleAccess Level
roles/agentregistry.adminFull administrative access across all Agent Registry resources, including bindings, skill scan findings, and resource-level IAM policies (aiApplications.{getIamPolicy,setIamPolicy} and skills.{getIamPolicy,setIamPolicy})
roles/agentregistry.editorRegister and update agents, MCP servers, endpoints, services, skills, skill revisions, and AI Applications (excludes bindings, scan findings, and IAM policy mutations)
roles/agentregistry.viewerRead-only access (get, list, search) to agents, MCP servers, endpoints, services, bindings, skills, skill revisions, publishers, and AI Applications
roles/agentregistry.userAgent Registry User — least-privilege GA role granting read-only viewer access across Agent Registry plus permissions to create, update, and delete standalone skills, skill revisions, and scanning settings without edit access to agents, MCP servers, or endpoints
roles/run.admin + roles/apphub.admin + roles/iam.serviceAccountUserDeploy or update Cloud Run services and jobs with --functional-type and --identity-type
roles/resourcemanager.editorCreate and update BoundaryConfig, CapabilityConfig, and Boundary on the parent folder or organization
roles/resourcemanager.viewerRead BoundaryConfig, CapabilityConfig, and Boundary resources
roles/resourcemanager.tagUserBind or unbind tag values (tags bindings create/delete) on projects
roles/owner or roles/editorRequired on a candidate Bring-Your-Own project to associate it as a Management Project

Resource Types & Registration

  • Manual Registration (services create): gcloud agent-registry services create creates a writable Service resource to manually register external or cross-project Agent, McpServer, and Endpoint resources in Agent Registry, and automatically projects a corresponding read-only Agent, McpServer, or Endpoint catalog entry (agentregistry-{ID}, returned in registryResource on services describe) used for runtime discovery and IAP/Agent Gateway policies.

1. MCP Servers

If the user has not provided the MCP tool spec JSON yet, ask: "Please paste the contents of your MCP server spec JSON file." and preview the command template using 'PASTE_MCP_SPEC_JSON_HERE':

bash
# Register an MCP Server (mutating — ask for approval first)gcloud agent-registry services create my-mcp-service \  --location=us-central1 \  --display-name="My MCP Server" \  --description="Connects to an external MCP server" \  --mcp-server-spec-type=tool-spec \  --mcp-server-spec-content='PASTE_MCP_SPEC_JSON_HERE' \  --interfaces='[{"protocolBinding": "jsonrpc", "url": "https://mcp.example.com/v1"}]'
# List Regional or Global MCP Servers (read-only — execute immediately)gcloud agent-registry mcp-servers list --location=us-central1gcloud agent-registry mcp-servers list --location=global

2. Agents

For A2A agents, if the user has not pasted agent_card.json yet, ask: "Please paste the contents of your agent_card.json file." and preview the command template using 'PASTE_AGENT_CARD_JSON_HERE':

bash
# Register a Standard Agent (no-spec)gcloud agent-registry services create my-custom-agent \  --location=us-central1 \  --display-name="My Custom Agent" \  --description="Custom HTTP JSON agent" \  --agent-spec-type=no-spec \  --interfaces='[{"protocolBinding": "http-json", "url": "https://agent.example.com/v1"}]'
# Register an A2A Agent (a2a-agent-card)gcloud agent-registry services create my-a2a-agent \  --location=us-central1 \  --display-name="My A2A Agent" \  --description="Sample A2A Agent" \  --agent-spec-type=a2a-agent-card \  --agent-spec-content='PASTE_AGENT_CARD_JSON_HERE'
# List Regional or Global Agentsgcloud agent-registry agents list --location=us-central1gcloud agent-registry agents list --location=global

3. Endpoints, Services & Quick Add GCP Service (Composite Agent Gateway Registration)

bash
# Register a single Endpoint, list/describe/update/delete Services, and inspect Operationsgcloud agent-registry services create gemini-models --location=us-central1 \  --display-name="Vertex AI Model Garden" --description="List of all models in Vertex AI Model Garden" \  --endpoint-spec-type=no-spec \  --interfaces='[{"protocolBinding": "jsonrpc", "url": "https://us-central1-aiplatform.googleapis.com/v1beta1/publishers/*/models"}]'gcloud agent-registry endpoints list --location=us-central1gcloud agent-registry services list --location=us-central1gcloud agent-registry services describe gemini-models --location=us-central1gcloud agent-registry services update gemini-models --display-name="Model Garden on Vertex AI" --location=us-central1gcloud agent-registry services delete gemini-models --location=us-central1gcloud agent-registry operations list --location=us-central1gcloud agent-registry operations describe OPERATION_ID --location=us-central1
Quick Add GCP Service (All 5 Endpoint URL Variants)

When an agent runs behind a governed Agent Gateway (default-deny egress), Google Cloud SDKs dynamically resolve endpoints across up to 5 hostname variants:

  1. Base: https://${SERVICE_ID}.googleapis.com
  2. Base mTLS: https://${SERVICE_ID}.mtls.googleapis.com
  3. Locational: https://${LOCATION}-${SERVICE_ID}.googleapis.com (omit when --location=global)
  4. Locational mTLS: https://${LOCATION}-${SERVICE_ID}.mtls.googleapis.com (omit when --location=global)
  5. Regional REP: https://${SERVICE_ID}.${LOCATION}.rep.googleapis.com (omit when --location=global)

Register a single composite Service containing all applicable variants using scripts/quick-add-gcp-service.sh [blocked] or gcloud agent-registry services create (always previewing and asking for approval first):

  • Service ID Length (4–63 chars): Service IDs must match ^[a-z][a-z0-9-]{2,61}[a-z0-9]$. For 3-letter APIs (iam, kms, run, dns), suffix the Service ID (e.g., ${SERVICE_ID}-api) while keeping ${SERVICE_ID} in the endpoint URLs.
  • Global Location Guardrail: When --location=global, include only Base and Base mTLS so invalid regional DNS names are not registered.
  • IAP Egress Policy Target: Bind roles/iap.egressor for the Agent Identity to the projected Endpoint catalog ID (endpoints/agentregistry-{UUID} from registryResource on services describe), not the user-chosen Service ID.
bash
# Option A: Bundled Quick Add helper script (handles 3-letter APIs, --location=global, and --dry-run)./scripts/quick-add-gcp-service.sh --service=aiplatform --location=us-central1 --project=PROJECT_ID
# Option B: Single composite gcloud command (all 5 variants for a regional location)SERVICE_ID="aiplatform" && SERVICE_NAME="aiplatform" && LOCATION="us-central1" && PROJECT_ID="PROJECT_ID"gcloud agent-registry services create "${SERVICE_NAME}" --project="${PROJECT_ID}" --location="${LOCATION}" \  --display-name="${SERVICE_ID}" --description="Google Cloud ${SERVICE_ID} API endpoints" --endpoint-spec-type=no-spec \  --interfaces="protocolBinding=jsonrpc,url=https://${SERVICE_ID}.googleapis.com" \  --interfaces="protocolBinding=jsonrpc,url=https://${SERVICE_ID}.mtls.googleapis.com" \  --interfaces="protocolBinding=jsonrpc,url=https://${LOCATION}-${SERVICE_ID}.googleapis.com" \  --interfaces="protocolBinding=jsonrpc,url=https://${LOCATION}-${SERVICE_ID}.mtls.googleapis.com" \  --interfaces="protocolBinding=jsonrpc,url=https://${SERVICE_ID}.${LOCATION}.rep.googleapis.com"

4. GKE Agents & MCP Servers (Automatic Registration)

Configure GKE Deployment manifests for automatic discovery in Agent Registry (using metadata.name as the display name):

Component Typemetadata.labels (Deployment)metadata.annotations (Deployment)spec.template.metadata.annotations (Pod)
GKE Agentregistry.gke.io/functional-type: "AGENT" (plus apphub.cloud.google.com/functional-type: "AGENT" in annotations for compatibility)a2a-protocol.org/agent-card: Declares card.endpoint (/.well-known/agent-card.json), card.protocol (HTTP), card.port (8080)iam.gke.io/spiffe-identity-type: agent-identity
GKE MCP Serverregistry.gke.io/functional-type: "MCP_SERVER" (plus apphub.cloud.google.com/functional-type: "MCP_SERVER" in annotations for compatibility)modelcontextprotocol.info/urls: In-cluster URL list (- https://my-mcp-server.default.svc.cluster.local/mcp)<br>modelcontextprotocol.info/capabilities: card.endpoint ("/mcp"), card.protocol ("HTTP")iam.gke.io/spiffe-identity-type: agent-identity

Use scripts/annotate_gke.py [blocked] to automatically add registry.gke.io/functional-type, apphub.cloud.google.com/functional-type, and the Pod template iam.gke.io/spiffe-identity-type: "agent-identity" annotation to existing YAML files (target path is required; supports --dry-run):

bash
./scripts/annotate_gke.py /path/to/folder_or_file.yaml --type AGENT./scripts/annotate_gke.py /path/to/folder_or_file.yaml --type MCP_SERVER

GKE Agent (Deployment + Service) & MCP Server (Deployment) Templates:

yaml
apiVersion: apps/v1kind: Deploymentmetadata:  name: my-agent  labels:    registry.gke.io/functional-type: "AGENT"  annotations:    apphub.cloud.google.com/functional-type: "AGENT"    a2a-protocol.org/agent-card: |      card:        endpoint: /.well-known/agent-card.json        protocol: HTTP        port: 8080spec:  selector:    matchLabels:      app: my-agent  template:    metadata:      labels:        app: my-agent      annotations:        iam.gke.io/spiffe-identity-type: agent-identity    spec:      containers:      - name: server        image: us-central1-docker.pkg.dev/PROJECT_ID/my-repo/my-agent:1.0.0        ports:        - containerPort: 8080---apiVersion: v1kind: Servicemetadata:  name: my-agent-servicespec:  type: ClusterIP  ports:  - port: 8080    targetPort: 8080    protocol: TCP    name: http  selector:    app: my-agent
yaml
apiVersion: apps/v1kind: Deploymentmetadata:  name: my-mcp-server  labels:    registry.gke.io/functional-type: "MCP_SERVER"  annotations:    apphub.cloud.google.com/functional-type: "MCP_SERVER"    modelcontextprotocol.info/urls: |      - https://my-mcp-server.default.svc.cluster.local/mcp    modelcontextprotocol.info/capabilities: |      card:        endpoint: "/mcp"        protocol: "HTTP"spec:  selector:    matchLabels:      app: my-mcp-server  template:    metadata:      labels:        app: my-mcp-server      annotations:        iam.gke.io/spiffe-identity-type: agent-identity    spec:      containers:      - name: server        image: us-central1-docker.pkg.dev/PROJECT_ID/my-repo/my-mcp-server:1.0.0

5. Cloud Run Agents & MCP Servers (Automatic Registration via --functional-type)

Register Cloud Run services and jobs automatically using gcloud beta run:

[!WARNING] Critical Cloud Run Guardrails:

  1. Immutability: Once --functional-type or --identity-type is set on a Cloud Run service or job, it is immutable and cannot be changed or cleared.
  2. --functional-type=agent Requires --identity-type=agent-identity: Always include --identity-type=agent-identity whenever --functional-type=agent is used (certificates are provisioned by default; opt out with --no-identity-certificate).
  3. Updating Existing Services (--no-traffic Guardrail): Transitioning an existing service to --identity-type=agent-identity assigns a new SPIFFE Agent Identity principal (principal://agents.global.org-ORGANIZATION_ID.system.id.goog/resources/run/projects/PROJECT_NUMBER/locations/REGION/services/SERVICE_NAME) that does not inherit the previous service account's IAM permissions. Recommend --no-traffic on gcloud beta run services update to stage IAM bindings before shifting traffic.
  4. Cloud Run Jobs Cannot Be MCP Servers: Only Cloud Run services support --functional-type=mcp-server; Cloud Run jobs only support --functional-type=agent.
bash
# Deploy or update a Cloud Run service as an AI Agent or MCP Servergcloud beta run deploy SERVICE_NAME --image=IMAGE_URL --region=us-central1 \  --functional-type=agent --identity-type=agent-identitygcloud beta run services update SERVICE_NAME --region=us-central1 \  --functional-type=agent --identity-type=agent-identity --no-trafficgcloud beta run deploy SERVICE_NAME --image=IMAGE_URL --region=us-central1 \  --functional-type=mcp-server --identity-type=IDENTITY_TYPEgcloud beta run services update SERVICE_NAME --region=us-central1 --functional-type=mcp-server
# Create or update a Cloud Run Job as an AI Agentgcloud beta run jobs create JOB_NAME --image=IMAGE_URL --region=us-central1 \  --functional-type=agent --identity-type=agent-identitygcloud beta run jobs update JOB_NAME --region=us-central1 \  --functional-type=agent --identity-type=agent-identity
# Inspect Agent Identity or delete Cloud Run workloadsgcloud beta run revisions describe REVISION_NAME --region=us-central1gcloud beta run jobs executions describe EXECUTION_NAME --region=us-central1gcloud run services delete SERVICE_NAME --region=us-central1gcloud run jobs delete JOB_NAME --region=us-central1

6. Bindings

Connect a source agent to a target resource or an Auth Provider connector:

bash
gcloud agent-registry bindings create BINDING_NAME --location=us-central1 \  --display-name="My Custom Binding" --source-identifier="SOURCE_ID" --target-identifier="TARGET_ID"gcloud agent-registry bindings create BINDING_NAME --location=us-central1 \  --display-name="Auth Provider Binding" --source-identifier="SOURCE_ID" \  --auth-provider="projects/PROJECT_ID/locations/REGION/connectors/AUTH_PROVIDER_ID"gcloud agent-registry bindings list --location=us-central1gcloud agent-registry bindings describe BINDING_NAME --location=us-central1gcloud agent-registry bindings update BINDING_NAME --display-name="New Display Name" --location=us-central1gcloud agent-registry bindings delete BINDING_NAME --location=us-central1

7. Skills & Skill Revisions

Manage reusable AI Skills, Skill Revisions, and Skill IAM policies using gcloud agent-registry skills.

[!IMPORTANT] Skills Guardrails (--location, private- Prefix & --target-state):

  • Supported Locations: Skills only work in "global", "us", or "eu" multi-regions. If a user asks for any single region (e.g. us-central1), do not pass that region to gcloud. Explain that Skills are only supported in global, us, or eu, preview the command using a concrete supported multi-region (e.g. --location=us, --location=eu, or --location=global), and ask them to confirm which multi-region to use.
  • Automatic private- Prefix: Running gcloud agent-registry skills create my-skill automatically prepends private-, creating skills/private-my-skill. Subsequent describe, update, delete, revisions, and *-iam-policy commands must pass private-my-skill (passing my-skill fails with NOT_FOUND).
  • Target State & Scanning: --target-state=active|disabled|draft sets the desired lifecycle state on creation (active requires an initial revision via --payload or --gcs-source-uri; deprecated and decommissioned are also valid on update). Under the default enabled-blocking scanning policy, --target-state=active enters DRAFT (STATE_DRAFT) until the background security scan passes with no findings.
bash
# List, create (local ZIP or GCS URI), describe, update, and delete Skillsgcloud agent-registry skills list --location=usgcloud agent-registry skills create my-skill --location=us \  --display-name="My Custom Skill" --description="Skill description" --type=simple --target-state=active --payload=/path/to/archive.zipgcloud agent-registry skills create my-skill --location=us \  --display-name="My Custom Skill" --target-state=disabled --gcs-source-uri=gs://my-bucket/skill.zipgcloud agent-registry skills describe private-my-skill --location=usgcloud agent-registry skills update private-my-skill --location=us --display-name="Updated Skill Name" --target-state=activegcloud agent-registry skills delete private-my-skill --location=us
# Manage Skill Revisions and Resource-Level IAM Policies (using private-my-skill)gcloud agent-registry skills revisions create REVISION_ID --skill=private-my-skill --location=us --payload=/path/to/archive.zipgcloud agent-registry skills revisions list --skill=private-my-skill --location=usgcloud agent-registry skills revisions describe REVISION_ID --skill=private-my-skill --location=usgcloud agent-registry skills revisions download REVISION_ID --skill=private-my-skill --location=us --destination=./skill-payload.zipgcloud agent-registry skills revisions delete REVISION_ID --skill=private-my-skill --location=usgcloud agent-registry skills get-iam-policy private-my-skill --location=usgcloud agent-registry skills set-iam-policy private-my-skill policy.json --location=usgcloud agent-registry skills add-iam-policy-binding private-my-skill --location=us \  --member='user:[email protected]' --role='roles/agentregistry.viewer'gcloud agent-registry skills remove-iam-policy-binding private-my-skill --location=us \  --member='user:[email protected]' --role='roles/agentregistry.viewer'

8. AI Applications (ai-applications)

An AI Application (projects/{project}/locations/{location}/aiApplications/{ai_application}, including the regional default AI Application) groups agents, MCP servers, and endpoints under a governance perimeter.

bash
# List, filter, describe, and search AI Applicationsgcloud agent-registry ai-applications list --location=us-central1gcloud agent-registry ai-applications list --location=us-central1 \  --filter="attributes.environment.type=production AND attributes.criticality.type=high"gcloud agent-registry ai-applications describe my-app --location=us-central1gcloud agent-registry ai-applications describe default --location=us-central1gcloud agent-registry ai-applications search --location=us-central1 --search-string="displayName:Support*"
# Create, update, and delete an AI Applicationgcloud agent-registry ai-applications create my-app --location=us-central1 \  --display-name="Customer Support App" --description="AI Application orchestrating customer support agents" \  --environment-type=production --criticality-type=high \  [email protected] [email protected] [email protected]gcloud agent-registry ai-applications update my-app --location=us-central1 \  --display-name="Updated Support App" --environment-type=staging --criticality-type=mediumgcloud agent-registry ai-applications delete my-app --location=us-central1 --force
# Inspect & search subresources scoped to an AI Application (agents, mcp-servers, endpoints)gcloud agent-registry ai-applications agents list --ai-application=my-app --location=us-central1gcloud agent-registry ai-applications agents describe AGENT_ID --ai-application=my-app --location=us-central1gcloud agent-registry ai-applications agents search --ai-application=my-app --location=us-central1 --search-string="displayName:Assistant*"gcloud agent-registry ai-applications mcp-servers list --ai-application=my-app --location=us-central1gcloud agent-registry ai-applications endpoints list --ai-application=my-app --location=us-central1
# Manage Resource-Level IAM Policies on an AI Applicationgcloud agent-registry ai-applications get-iam-policy my-app --location=us-central1gcloud agent-registry ai-applications set-iam-policy my-app policy.json --location=us-central1gcloud agent-registry ai-applications add-iam-policy-binding my-app --location=us-central1 \  --member='group:[email protected]' --role='roles/agentregistry.viewer'gcloud agent-registry ai-applications remove-iam-policy-binding my-app --location=us-central1 \  --member='group:[email protected]' --role='roles/agentregistry.viewer'

9. Management Boundary Setup (Admin Governance)

Configure Folder/Organization hub-and-spoke perimeters using gcloud resource-manager (boundary-configs, capability-configs, boundaries, and tags bindings). Read Management Boundaries Guide [blocked] for the full IAM table, 6 interactive scoping questions, and 6-step provisioning workflow.

  • Always Use Project Numbers (projects/<PROJECT_NUMBER>): Cloud Resource Manager v3 requires project numbers (not project IDs) for --management-project=projects/<PROJECT_NUMBER> and --parent=//cloudresourcemanager.googleapis.com/projects/<PROJECT_NUMBER>. When given a project ID, output both gcloud projects describe PROJECT_ID --format="value(projectNumber)" and the capability-configs create command template (--types=agent-management by default) in the same response.
  • Set BoundaryConfig Before Attaching Boundaries & Isolate dev/prod Hubs: Set boundary-configs update --tag-key=... before attaching boundaries. Use separate CapabilityConfig + Management Project + Boundary per environment (dev vs. prod) for strict catalog and IAM isolation.
  • Tag Rebinding Requires Delete-Then-Create: Moving a project between tag values (e.g., dev to prod) fails with FAILED_PRECONDITION if you only run create; always run tags bindings delete first, then tags bindings create.
  • Safe 3-Step Teardown Order: Always output all 3 commands in order when deleting a boundary and capability config (noting that BYO projects are preserved with isManagementProject=false, while Google-managed hubs are deleted):
    bash
    gcloud resource-manager capability-configs update CAPABILITY_CONFIG_ID --folder=FOLDER_ID --clear-boundariesgcloud resource-manager boundaries delete BOUNDARY_ID --folder=FOLDER_IDgcloud resource-manager capability-configs delete CAPABILITY_CONFIG_ID --folder=FOLDER_ID

Advanced Filtering & Search

Filtering (list --filter)

  • URN Substring Filtering (agentId & mcpServerId):
    • agentId: urn:agent:projects-{project-number}:projects:{project-number}:locations:{location}:{platform-segments}
    • mcpServerId: urn:mcp:projects-{project-number}:projects:{project-number}:locations:{location}:{platform-segments}
    • Platform segments: aiplatform:reasoningEngines:{id} (Vertex AI Reasoning Engine), container:clusters:{cluster}:k8s:namespaces:{ns}:apps:deployments:{deploy} (GKE), run:services:{svc} or run:jobs:{job} (Cloud Run).
  • Attribute Mapping:
    • "runtime" ("agent engine", "reasoning engine", "vertex ai" -> reasoningEngine): attributes."agentregistry.googleapis.com/system/RuntimeReference".uri
    • "identity": attributes."agentregistry.googleapis.com/system/RuntimeIdentity".principal
    • "tool" / "tool name" (on mcp-servers list): tools.name
bash
# Filter agents by runtime (Vertex AI Reasoning Engine) or identity principalgcloud agent-registry agents list --location=us-central1 \  --filter="attributes.\"agentregistry.googleapis.com/system/RuntimeReference\".uri:reasoningEngine"gcloud agent-registry agents list --location=us-central1 \  --filter="attributes.\"agentregistry.googleapis.com/system/RuntimeIdentity\".principal:service-123456789012"
# Filter by agentId / mcpServerId substring or MCP tool namegcloud agent-registry agents list --location=us-central1 --filter="agentId:reason"gcloud agent-registry mcp-servers list --location=us-central1 --filter="mcpServerId:run"gcloud agent-registry mcp-servers list --location=us-central1 --filter="tools.name:search_documents"

Searching Agents, MCP Servers, and Skills (search)

  • Agents (gcloud agent-registry agents search): Supports --search-string with exact (=), token (:), and prefix (*) matching across agentId, displayName, skills.name, skills.id, skills.tags, skills.description, and boolean OR/AND. Note: Values with colons (URNs) must be quoted, and wildcard * is not supported on skills.description (use token match : instead).
  • MCP Servers (gcloud agent-registry mcp-servers search): Supports --search-string across mcpServerId, name, and displayName.
  • Skills (gcloud agent-registry skills search): Supports --query and --search-type=keyword or --search-type=semantic in global, us, or eu.
bash
# Search Agentsgcloud agent-registry agents search --location=us-central1 --search-string="agentId:\"urn:agent:projects-123456789012*\""gcloud agent-registry agents search --location=us-central1 --search-string="displayName:MyAgent*"gcloud agent-registry agents search --location=us-central1 --search-string="skills.name:translation"gcloud agent-registry agents search --location=us-central1 --search-string="skills.id:translate-text"gcloud agent-registry agents search --location=us-central1 --search-string="skills.tags:productivity"gcloud agent-registry agents search --location=us-central1 --search-string="skills.description:database"gcloud agent-registry agents search --location=us-central1 --search-string="analytics OR reporting"
# Search MCP Serversgcloud agent-registry mcp-servers search --location=us-central1 --search-string="mcpServerId:\"urn:mcp:projects-123456789012*\""gcloud agent-registry mcp-servers search --location=us-central1 --search-string="displayName:MyMcp*"
# Search Skills (Keyword and Semantic)gcloud agent-registry skills search --location=us --query="sql" --search-type=keywordgcloud agent-registry skills search --location=us --query="summarize support tickets" --search-type=semantic

Python ADK Integration

Whenever the user asks to discover or connect to registered Agents (get_remote_a2a_agent) or MCP Servers (get_mcp_toolset) in Python using Google's Agent Development Kit (google-adk[a2a,agent-identity,mcp]), read and follow ADK Integration Guide [blocked].


Documentation Fallback (developerknowledge MCP Server)

For CLI synopses, read gcloud Command Reference [blocked]. If a user asks about a feature, Terraform resource (google_agent_registry_service), or error not covered in SKILL.md or references/, query the developerknowledge MCP server (search_documents + get_documents, or answer_query) or fetch the public .md.txt documentation pages:

  • https://docs.cloud.google.com/agent-registry/overview.md.txt
  • https://docs.cloud.google.com/agent-registry/automatic-registration.md.txt
  • https://docs.cloud.google.com/agent-registry/manual-registration.md.txt
  • https://docs.cloud.google.com/agent-registry/register-agents.md.txt
  • https://docs.cloud.google.com/agent-registry/register-mcp-servers.md.txt
  • https://docs.cloud.google.com/agent-registry/register-endpoints.md.txt
  • https://docs.cloud.google.com/agent-registry/register-skills.md.txt
  • https://docs.cloud.google.com/agent-registry/manage-skills.md.txt
  • https://docs.cloud.google.com/agent-registry/skills-architecture.md.txt
  • https://docs.cloud.google.com/agent-registry/scan-skills.md.txt
  • https://docs.cloud.google.com/agent-registry/boundaries-and-topology.md.txt
  • https://docs.cloud.google.com/agent-registry/access-control.md.txt
  • https://docs.cloud.google.com/agent-registry/roles-permissions.md.txt
  • https://docs.cloud.google.com/agent-registry/a2a-protocol.md.txt
  • https://docs.cloud.google.com/agent-registry/resolve-endpoints-and-build-orchestrators.md.txt
  • https://docs.cloud.google.com/agent-registry/search-agents-and-tools.md.txt
  • https://docs.cloud.google.com/agent-registry/identity-and-identifiers.md.txt
  • https://docs.cloud.google.com/agent-registry/troubleshooting.md.txt
  • https://docs.cloud.google.com/run/docs/ai/agent-platform-features.md.txt

Error Handling

  1. Verify Cloud SDK version (560.0.0+, updated for GA skills) and beta component (2026.03.09+ for Cloud Run).
  2. Verify --location / --region: Google Cloud remote MCP servers are discovered in global; Skills only support global, us, or eu.
  3. Ensure JSON strings passed to --interfaces, --agent-spec-content, or --mcp-server-spec-content are properly single-quoted for the shell.
  4. Confirm required APIs (agentregistry.googleapis.com, run.googleapis.com, apphub.googleapis.com) and IAM roles are enabled on the target project.

來源與署名

來源:google/skills位於skills/cloud/agent-platform-agent-registry提交5fb62c2

授權條款: 無授權條款

內容歸原作者所有。SourceWeft 從公開儲存庫中收錄這些內容。

在 README 中新增徽章

向讀者展示該技能已收錄於 SourceWeft,徽章連結到本頁。

「Listed on SourceWeft」徽章預覽
[![Listed on SourceWeft](https://sourceweft.com/badge/skills/gh-google-skills-agent-platform-agent-registry-skills-cloud-agent-platform-agent-registry-f5009a66af5e1058.svg)](https://sourceweft.com/skills/gh-google-skills-agent-platform-agent-registry-skills-cloud-agent-platform-agent-registry-f5009a66af5e1058?utm_source=readme&utm_medium=badge)

檢舉或申請下架