Create IAM Access policies

Console

To create an Access policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. To add an IAM Access policy for Agent Gateway egress, click add Create.
4. In Policy details, do the following:
   1. Click the Policy field.
   2. In the drop-down list, select Create new policy.
   3. In the Policy name field, enter the name of the policy. The policy ID is automatically generated based on the policy name.
   4. Optional: Click Edit to edit the policy ID. In the Policy ID field, you can change the name of the policy ID.
   5. In Add Rules, expand the default rule, Rule 1 or click Add a rule.
5. Create one or more rules for your Access policy. View examples of different policies.
6. To create your policy, click Create.

After you save your policy, the following happens:

• The IAM iap.resources.egressViaIAP permission is granted to the principals in your policy.

• The policy is automatically bound to the selected project.

• All Agent Gateway instances in this project can enforce any policy associated with the project.

gcloud

To create an Access policy in the gcloud CLI, do the following:

1. Create a JSON-formatted policy file. Access policies can contain one or more rules. Each rule controls access from one or more agent principals to one or more target resources.

   [
     {
       "description": "RULE_DESCRIPTION",
       "effect": "RULE_EFFECT",
       "principals": [
         AGENT_PRINCIPALS
       ],
       "operation": {
         "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
       },
       
       "conditions": {
         "iap.googleapis.com": {
           "expression": "CEL_EXPRESSION"
         }
       }
       
     }
   ]
   

   Replace the following:

   • RULE_DESCRIPTION: the description of the rule
   • RULE_EFFECT: the effect of the rule, which is either ALLOW or DENY
   • AGENT_PRINCIPALS: the principals for the source agent identity. Agent principals can be an array of one or more individual principals or principal sets:
     • Built-in agent identities: principal://TRUST_DOMAIN/AGENT_UNIQUE_IDENTIFIER

       The structure of AGENT_UNIQUE_IDENTIFIER depends on the Google Cloud service hosting the agent:

       • Agent Runtime (Reasoning Engine): principal://agents.global.org-123456789012.system.id.goog/resources/aiplatform/projects/9876543210/locations/us-central1/reasoningEngines/my-agent
       • Gemini Enterprise (Discovery Engine): principal://agents.global.org-123456789012.system.id.goog/resources/discoveryengine/projects/9876543210/locations/global/engines/my-engine/assistants/default_assistant/agents/default/core_assistant
       • Cloud Run (Agent Identity): principal://agents.global.org-123456789012.system.id.goog/resources/run/projects/9876543210/locations/us-central1/services/my-cloud-run-agent

       Replace the following:

       • TRUST_DOMAIN: the trust domain of the project or organization that contains the agent principal (for example, agents.global.org-ORGANIZATION_ID.system.id.goog or agents.global.proj-PROJECT_NUMBER.system.id.goog)
       • AGENT_UNIQUE_IDENTIFIER: the resource URI path of the agent instance
     • Workload Identity Federation (custom or external agents): principal://iam.googleapis.com/projects/PROJECT_NUMBER/locations/global/workloadIdentityPools/POOL_ID/subject/SUBJECT—for example, principal://iam.googleapis.com/projects/1234567890/locations/global/workloadIdentityPools/my-agent-identity/subject/ns/default/sa/my-agent

       Replace the following:

       • PROJECT_NUMBER: the project number of the project that contains the workload identity pool
       • POOL_ID: the ID of the workload identity pool
       • SUBJECT: the subject identifier of the agent principal
   • CONDITION: the condition parameter, formatted as follows:

     "conditions": {
       "iap.googleapis.com": {
         "expression": "CEL_EXPRESSION"
       }
     }
     

     Replace CEL_EXPRESSION with a conditional expression—for example:

     "destination.agent_registry.mcp_server.name == '/projects/corp-apim-prod/locations/us-east1/mcpServers/finance-data-service' && destination.agent_registry.mcp_server.tool.name == 'updateRecord'"
     

     This expression must be a valid CEL expression. It must evaluate to true for access to be allowed.

   In all agent egress policies, you must set the permission to iap.resources.egressViaIAP.

   For more information, see Access policy rules and Example policies.

2. To create the Access policy, run the following command:

   gcloud iam access-policies create POLICY_NAME \
       --details-rules=POLICY_FILE \
       --project=PROJECT_ID \
       --location=global
   

   Replace the following:

   • POLICY_NAME: the name of the policy
   • POLICY_FILE: the path to the policy file—for example: my-policy.json
   • PROJECT_ID: the project ID that contains the policy

REST API

To create and bind a policy using the REST API, do the following:

1. Save the JSON-formatted policy details to a file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         RULES
       ]
     }
   }
   

   Replace RULES with one or more rules.

2. To create the Access policy, make a POST request to the accessPolicies endpoint:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

Console

To create an Access policy with both allow and deny rules in the Google Cloud console, do the following:

1. In Add Rules, configure the allow rule:
   1. In Rule description, enter the description for the allow rule.
   2. In Rule effect, select Allow.
   3. Select the agent principals and target resources, and add the allow CEL conditions.
   4. To save the allow rule, click Save.
2. To add the deny rule, click Add a rule:
   1. In Rule description, enter the description for the deny rule.
   2. In Rule effect, select Deny.
   3. Select the agent principals and target resources, and add the deny CEL conditions.
   4. To save the deny rule, click Save.
3. To create the policy, click Create.

gcloud

To create an Access policy file containing both allow and deny rules, do the following:

1. Save the following to your JSON-formatted policy file:

   [
     {
       "description": "ALLOW_RULE_DESCRIPTION",
       "effect": "ALLOW",
       "principals": [
         "AGENT_PRINCIPALS_ALLOW"
       ],
       "operation": {
         "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
       },
       "conditions": {
         "iap.googleapis.com": {
           "expression": "CEL_EXPRESSION_ALLOW"
         }
       }
     },
     {
       "description": "DENY_RULE_DESCRIPTION",
       "effect": "DENY",
       "principals": [
         "AGENT_PRINCIPALS_DENY"
       ],
       "operation": {
         "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
       },
       "conditions": {
         "iap.googleapis.com": {
           "expression": "CEL_EXPRESSION_DENY"
         }
       }
     }
   ]
   

   Replace the following:

   • ALLOW_RULE_DESCRIPTION: the description of the allow rule
   • DENY_RULE_DESCRIPTION: the description of the deny rule
   • AGENT_PRINCIPALS_ALLOW: the agent principals that you want to allow—for example, principal://agents.global.proj-PROJECT_NUMBER.system.id.goog/resources/aiplatform/projects/PROJECT_ID/locations/LOCATION/reasoningEngines/AGENT_NAME
   • AGENT_PRINCIPALS_DENY: the agent principals that you want to deny—for example, principal://agents.global.proj-PROJECT_NUMBER.system.id.goog/resources/aiplatform/projects/PROJECT_ID/locations/LOCATION/reasoningEngines/AGENT_NAME
   • CEL_EXPRESSION_ALLOW: the CEL expression for the allow rule—for example, destination.agent_registry.mcp_server.name == "/projects/my-project/locations/us-east1/mcpServers/finance-data-service" && destination.agent_registry.mcp_server.tool.name == "getStatements"
   • CEL_EXPRESSION_DENY: the CEL expression for the deny rule—for example, destination.agent_registry.mcp_server.name == "/projects/my-project/locations/us-east1/mcpServers/finance-data-service" && destination.agent_registry.mcp_server.tool.name == "updateRecord"

2. To create the Access policy, run the following command:

   gcloud iam access-policies create POLICY_NAME \
       --details-rules=POLICY_FILE \
       --project=PROJECT_ID \
       --location=global
   

   Replace the following:

   • POLICY_NAME: the name of the policy
   • POLICY_FILE: the path to the policy file—for example: my-policy.json
   • PROJECT_ID: the project ID that contains the policy

3. To activate the policy, bind the Access policy to your project.

REST API

To create an Access policy containing both allow and deny rules using the REST API, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "ALLOW_RULE_DESCRIPTION",
           "effect": "ALLOW",
           "principals": [
             "AGENT_PRINCIPALS_ALLOW"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "CEL_EXPRESSION_ALLOW"
             }
           }
         },
         {
           "description": "DENY_RULE_DESCRIPTION",
           "effect": "DENY",
           "principals": [
             "AGENT_PRINCIPALS_DENY"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "CEL_EXPRESSION_DENY"
             }
           }
         }
       ]
     }
   }
   

   Replace the following:

   • ALLOW_RULE_DESCRIPTION: the description of the allow rule
   • DENY_RULE_DESCRIPTION: the description of the deny rule
   • AGENT_PRINCIPALS_ALLOW: the agent principals that you want to allow
   • AGENT_PRINCIPALS_DENY: the agent principals that you want to deny
   • CEL_EXPRESSION_ALLOW: the CEL expression for the allow rule
   • CEL_EXPRESSION_DENY: the CEL expression for the deny rule

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

3. To activate the policy, bind the Access policy to your project.

Console

After you save your policy, the following happens:

• The IAM iap.resources.egressViaIAP permission is granted to the principals in your policy.

• The policy is automatically bound to the selected project.

• All Agent Gateway instances in this project can enforce any policy associated with the project.

gcloud

To bind the policy to the target resource, run the following command:

gcloud iam policy-bindings create BINDING_NAME \
    --policy="projects/PROJECT_ID/locations/global/accessPolicies/POLICY_NAME" \
    --target-resource="TARGET_RESOURCE" \
    --project=PROJECT_ID \
    --location=global

Replace the following:

• BINDING_NAME: the name for your policy binding
• PROJECT_ID: your Google Cloud project ID
• POLICY_NAME: the name of your access policy
• TARGET_RESOURCE: the full resource URI for the project, formatted as follows:

      //cloudresourcemanager.googleapis.com/projects/PROJECT_ID
      

REST API

To bind the policy to the target resource, make a POST request to the policyBindings endpoint:

curl -X POST \
  "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/policyBindings?policyBindingId=BINDING_NAME" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d '{
    "policy": "projects/PROJECT_ID/locations/global/accessPolicies/POLICY_NAME",
    "target": {
      "resource": "TARGET_RESOURCE"
    }
  }'

Replace the following:

• PROJECT_ID: the project ID
• BINDING_NAME: a name for the binding
• POLICY_NAME: the name of the policy that you created earlier in this document
• TARGET_RESOURCE: the full resource URI for the binding target, which can be one of the following:
  • Project: //cloudresourcemanager.googleapis.com/projects/PROJECT_ID
  • Agent Gateway: //networkservices.googleapis.com/projects/PROJECT_ID/locations/LOCATION/agentGateways/GATEWAY_NAME

Console

In the Google Cloud console, edit your Access policy and do the following:

1. Edit your existing rule. To add a rule, click Add a rule.
2. Rule description: a human-readable description for this rule
3. Select principals: The source agent principals that the rule manages access from. Principals can be one of the following:
   • All agents in this project: All agents that exist in the current project.
   • Individual agents: One or more specific principals that you can select. Agents are defined by their agent identities.
   • Custom principal set: One or more principal sets that you can specify by entering a principal identifier. Principal identifiers must start with principalSet:// and refer to agent identities.
4. Effect: the rule's effect, which can be one of the following:
   • Allow: allows the principals to access the resources
   • Deny: disallows the principals from accessing the resources
5. Select resources: One or more types of destination resources that you can select. Resource types include the following:
   • Registry: All resources that are registered in an Agent Registry location
   • Agent: A destination agent that is registered in Agent Registry

   • MCP server: A destination MCP server that is registered in Agent Registry. You can use conditions to further control MCP server access.

   • Endpoint: A destination endpoint that is registered in Agent Registry

   • Unregistered endpoint: A destination endpoint that is not registered in Agent Registry and is accessed through an external URL. You can use conditions to further control unregistered endpoint access.

gcloud

Rules in a JSON-formatted file have the following general format:

{
  "description": "RULE_DESCRIPTION",
  "effect": "EFFECT",
  "principals": [
    "AGENT_PRINCIPALS"
  ],
  "operation": {
    "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
  },
  "conditions": {
    "iap.googleapis.com": {
      "expression": "CONDITION"
    }
  }
}

The policy file contains the following:

• RULE_DESCRIPTION: a human-readable description for this rule

• EFFECT: the rule's effect, which can be one of the following:

• ALLOW allows the principal to access the resources

• DENY disallows the principal from accessing the resources

• CONDITION: a CEL expression that conditionally evaluates to true or false. If the expression evaluates to true, then the rule goes into effect. Conditions can also contain references to specific resources that principals can access:

• destination.agent_registry: the destination resource is registered in Agent Registry.

• destination.unregistered: the destination resource is unregistered. It is accessed through a URL. Access can be controlled in the condition.

To learn more about conditions, see Conditions. Learn more about conditions attributes in CEL attributes.

REST API

To create and bind a policy using the REST API, do the following:

1. Save the JSON-formatted policy details to a file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "RULE_DESCRIPTION",
           "effect": "EFFECT",
           "principals": [
             "AGENT_PRINCIPALS"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "CONDITION"
             }
           }
         }
       ]
     }
   }
   

   Add one or more rules to the details field. In each rule, replace the following:

   • RULE_DESCRIPTION: one or more rules

   • EFFECT: the rule's effect, which can be one of the following:

     • ALLOW allows the principal to access the resources

     • DENY disallows the principal from accessing the resources

   • AGENT_PRINCIPALS: the agent principals that you want to grant access to.

   • CONDITION: a CEL expression that conditionally evaluates to true or false. If the expression evaluates to true, then the rule goes into effect. Conditions can also contain references to specific resources that principals can access:

     • destination.agent_registry: the destination resource is registered in Agent Registry.
     • destination.unregistered: the destination resource is unregistered. It is accessed through a URL. Access can be controlled in the condition.

   To learn more about conditions, see Conditions. Learn more about conditions attributes in CEL attributes.

2. To create the Access policy, make a POST request to the accessPolicies endpoint:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

3. To bind the policy to the target resource, make a POST request to the policyBindings endpoint:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/policyBindings?policyBindingId=BINDING_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d '{
     "policy": "projects/PROJECT_ID/locations/global/accessPolicies/POLICY_NAME",
     "target": {
       "resource": "TARGET_RESOURCE"
     }
   }'
   

   Replace the following:

   • PROJECT_ID: the project ID
   • BINDING_NAME: a name for the binding
   • POLICY_NAME: the name of the policy that you created earlier in this document
   • TARGET_RESOURCE: the full resource URI for the project formatted as follows: //cloudresourcemanager.googleapis.com/projects/PROJECT_ID

Console

In the Google Cloud console you can create conditions by doing the following:

1. In the Select resource(s) section of your rule, select one of the following tabs:

   • Standard: In the Standard tab, select one or more resource types and then use the condition builder or the condition editor to create a condition.

   • Custom: In the Custom tab, you can use the condition editor to create a custom condition in which you can use CEL attributes to specify your destination resources and other conditions.

2. To save the condition along with your rule, click Save.

gcloud

In the policy file, conditions determine which destination resources your agent principals can access and under what conditions they are accessed.

Conditions are defined in the conditions field of a rule. For example:

"conditions": {
  "iap.googleapis.com": {
    "expression": "destination.agent_registry.mcp_server.name == '/projects/my-project/locations/us-east1/mcpServers/finance-data-service' && destination.agent_registry.mcp_server.tool.name == 'getStatements'"
  }
}

For more information, see Example policies.

REST API

In the policy file, conditions determine which destination resources your agent principals can access and under what conditions they are accessed.

Conditions are defined in the conditions field of a rule. For example:

"conditions": {
  "iap.googleapis.com": {
    "expression": "destination.agent_registry.mcp_server.name == '/projects/my-project/locations/us-east1/mcpServers/finance-data-service' && destination.agent_registry.mcp_server.tool.name == 'getStatements'"
  }
}

For more information, see Example policies.

Console

To create an agent-to-registry policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. In Policy details, select a policy or create a new one. If you're creating a new policy, enter a policy name—for example, allow-registry-access.
4. In Add Rules, configure the rule:
   • Rule description: Allow all agents in project to access us-central1 registry
   • Rule effect: Allow
   • Select principals: In Principal sets, select All agents in this project.
   • Select resource(s): Select Standard > Registry, and select us-central1.
5. To save the rule, click Save.
6. To create the policy, click Create.

To learn how to create and bind policies, see Create IAM Access policies.

gcloud

The following example policy shows an Access policy rule that allows all agents in a project access to all destination resources in a specific registry:

{
  [
    {
      "description": "Allow all agents in project 9876543210 to access all destination resources in the us-central1 registry",
      "effect": "ALLOW",
      "principals": [
        "principalSet://agents.global.org-123456789012.system.id.goog/attribute.platformContainer/aiplatform/projects/9876543210"
      ],
      "operation": {
        "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
      },
      "conditions": {
        "iap.googleapis.com": {
          "expression": "destination.is_registered == true && destination.agent_registry.location == 'us-central1'"
        }
      }
    }
  ]
}

To activate the policy, bind the Access policy to your project.

REST API

To create an Access policy that allows all agents in a project access to all destination resources in a specific registry, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "Allow all agents in project 9876543210 to access all destination resources in the us-central1 registry",
           "effect": "ALLOW",
           "principals": [
             "principalSet://agents.global.org-123456789012.system.id.goog/attribute.platformContainer/aiplatform/projects/9876543210"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "destination.is_registered == true && destination.agent_registry.location == 'us-central1'"
             }
           }
         }
       ]
     }
   }
   

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

To activate the policy, bind the Access policy to your project.

Console

To create an agent-to-agent policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. In Policy details, select a policy or create a new one. If you're creating a new policy, enter a policy name—for example, allow-orchestrator-to-support-agent.
4. In Add Rules, configure the rule:
   • Rule description: Allow orchestrator agent to invoke customer-support agent
   • Rule effect: Allow
   • Select principals: Select Specific agents, and select orchestrator-agent.
   • Select resource(s): Select Standard > Agent, and select customer-support-agent.
5. To save the rule, click Save.
6. To create the policy, click Create.

To activate the policy, bind the Access policy to your project.

gcloud

The following example policy shows an IAM allow rule that allows an orchestrator agent to invoke a specific subagent registered in the Agent Registry:

{
  [
    {
      "description": "Allow orchestrator agent to invoke customer-support agent",
      "effect": "ALLOW",
      "principals": [
        "principal://agents.global.org-123456789012.system.id.goog/resources/aiplatform/projects/9876543210/locations/us-central1/reasoningEngines/orchestrator-agent"
      ],
      "operation": {
        "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
      },
      "conditions": {
        "iap.googleapis.com": {
          "expression": "destination.agent_registry.agent.name == '/projects/9876543210/locations/us-central1/agents/customer-support-agent'"
        }
      }
    }
  ]
}

To create and bind the policy by using the gcloud CLI, see Create IAM Access policies.

REST API

To create an IAM policy with an allow rule that allows an orchestrator agent to invoke a specific subagent registered in the Agent Registry, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "Allow orchestrator agent to invoke customer-support agent",
           "effect": "ALLOW",
           "principals": [
             "principal://agents.global.org-123456789012.system.id.goog/resources/aiplatform/projects/9876543210/locations/us-central1/reasoningEngines/orchestrator-agent"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "destination.agent_registry.agent.name == '/projects/9876543210/locations/us-central1/agents/customer-support-agent'"
             }
           }
         }
       ]
     }
   }
   

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

To activate the policy, bind the Access policy to your project.

Console

To create an agent-to-MCP server policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. In Policy details, select a policy or create a new one. If you're creating a new policy, enter a policy name—for example, allow-github-read-only.
4. In Add Rules, configure the rule:
   • Rule description: Allow read-only access to GitHubTool on MCP server
   • Rule effect: Allow
   • Select principals: Select the agent principal—for example, my-ae-agent.
   • Select resource(s): Select the Custom tab.
   • Conditions: In the condition editor, enter:

     destination.agent_registry.mcp_server.tool.name == 'GitHubTool' && destination.agent_registry.mcp_server.tool.annotations.read_only_hint == true

5. To save the rule, click Save.
6. To create the policy, click Create.

To activate the policy, bind the Access policy to your project.

gcloud

The following example policy shows an IAM allow rule that allows a Workload Identity Federation-formatted agent principal read-only access to a tool called GitHubTool:

{
  [
    {
      "description": "Allow read-only access to GitHubTool on MCP server",
      "effect": "ALLOW",
      "principals": [
        "principal://iam.googleapis.com/projects/9876543210/locations/global/workloadIdentityPools/POOL_ID/subject/ns/default/sa/my-ae-agent"
      ],
      "operation": {
        "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
      },
      "conditions": {
        "iap.googleapis.com": {
          "expression": "destination.agent_registry.mcp_server.tool.name == 'GitHubTool' && destination.agent_registry.mcp_server.tool.annotations.read_only_hint == true"
        }
      }
    }
  ]
}

Replace the following:

• POOL_ID: the workload identity pool ID. Depending on your pool type, format the ID as follows:
• Google-managed pool: PROJECT_ID.svc.id.goog
• Self-managed pool: POOL_NAME.global.POOL_HOST_PROJECT_NUMBER.workload.id.goog

To create and bind the policy by using the gcloud CLI, see Create IAM Access policies.

REST API

To create an IAM Access policy that allows an agent using Workload Identity Federation read-only access to a tool called GitHubTool, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "Allow read-only access to GitHubTool on MCP server",
           "effect": "ALLOW",
           "principals": [
             "principal://iam.googleapis.com/projects/9876543210/locations/global/workloadIdentityPools/POOL_ID/subject/ns/default/sa/my-ae-agent"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "destination.agent_registry.mcp_server.tool.name == 'GitHubTool' && destination.agent_registry.mcp_server.tool.annotations.read_only_hint == true"
             }
           }
         }
       ]
     }
   }
   

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • POOL_ID: the workload identity pool ID. Depending on your pool type, format the ID as follows:
   • Google-managed pool: PROJECT_ID.svc.id.goog
   • Self-managed pool: POOL_NAME.global.POOL_HOST_PROJECT_NUMBER.workload.id.goog
   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

To create and bind the policy by using the REST API, see Create Access policies.

Console

To create an agent-to-endpoint policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. In Policy details, select a policy or create a new one. If you're creating a new policy, enter a policy name—for example, allow-translator-to-endpoint.
4. In Add Rules, configure the rule:
   • Rule description: Allow translator agent to access translation endpoint
   • Rule effect: Allow
   • Select principals: Select Specific agents, and select translator-agent.
   • Select resource(s): Select the Custom tab.
   • Conditions: In the condition editor, enter:

     destination.agent_registry.endpoint.name == '/projects/9876543210/locations/us-central1/endpoints/translation-service'

5. To save the rule, click Save.
6. To create the policy, click Create.

To activate the policy, bind the Access policy to your project.

gcloud

The following example shows an IAM allow policy that allows an agent to access a registered service endpoint:

[
  {
    "description": "Allow translator agent access to registered translation endpoint",
    "effect": "ALLOW",
    "principals": [
      "principal://agents.global.org-123456789012.system.id.goog/resources/aiplatform/projects/9876543210/locations/us-central1/reasoningEngines/translator-agent"
    ],
    "operation": {
      "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
    },
    "conditions": {
      "iap.googleapis.com": {
        "expression": "destination.agent_registry.endpoint.name == '/projects/9876543210/locations/us-central1/endpoints/translation-service'"
      }
    }
  }
]

To activate the policy, bind the Access policy to your project.

REST API

To create an IAM allow policy that allows an agent to access a registered service endpoint, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "Allow translator agent access to registered translation endpoint",
           "effect": "ALLOW",
           "principals": [
             "principal://agents.global.org-123456789012.system.id.goog/resources/aiplatform/projects/9876543210/locations/us-central1/reasoningEngines/translator-agent"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "destination.agent_registry.endpoint.name == '/projects/9876543210/locations/us-central1/endpoints/translation-service'"
             }
           }
         }
       ]
     }
   }
   

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

To activate the policy, bind the Access policy to your project.

Console

To create an agent-to-unregistered endpoint policy in the Google Cloud console, do the following:

1. In the Google Cloud console, go to the Policies page:

   Go to Policies

2. In the project selector, select the project that contains the agent gateways that will use the Access policy.
3. In Policy details, select a policy or create a new one. If you're creating a new policy, enter a policy name—for example, allow-ocr-api-access.
4. In Add Rules, configure the rule:
   • Rule description: Allow POST access to external OCR service
   • Rule effect: Allow
   • Select principals: Select the agent principal (for example, my-ae-agent).
   • Select resource(s): Select Standard > Unregistered endpoint.
   • Conditions: Click the Custom tab.
   • In the condition editor, enter:

     destination.unregistered.host.endsWith('example-ocr.com') && destination.unregistered.path.startsWith('/v2/process') && destination.unregistered.method == 'POST'

5. To save the rule, click Save.
6. To create the policy, click Create.

To activate the policy, bind the Access policy to your project.

gcloud

The following example shows an Access policy that allows an agent to access an unregistered endpoint with a path that starts with /v2/process and ends with example-ocr.com:

[
  {
    "description": "Allow POST access to unregistered OCR endpoint",
    "effect": "ALLOW",
    "principals": [
      "principal://iam.googleapis.com/projects/9876543210/locations/global/workloadIdentityPools/POOL_ID/subject/ns/default/sa/my-ae-agent"
    ],
    "operation": {
      "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
    },
    "conditions": {
      "iap.googleapis.com": {
        "expression": "destination.unregistered.host.endsWith('example-ocr.com') && destination.unregistered.path.startsWith('/v2/process') && destination.unregistered.method == 'POST'"
      }
    }
  }
]

Replace the following:

• POOL_ID: the workload identity pool ID. Depending on your pool type, format the ID as follows:
• Google-managed pool: PROJECT_ID.svc.id.goog
• Self-managed pool: POOL_NAME.global.POOL_HOST_PROJECT_NUMBER.workload.id.goog

To activate the policy, bind the Access policy to your project.

REST API

To create an Access policy that allows an agent to access an unregistered endpoint with a path that starts with /v2/process and ends with example-ocr.com, do the following:

1. Save the following to a JSON-formatted policy file named agent-access-policy.json:

   {
     "details": {
       "rules": [
         {
           "description": "Allow POST access to unregistered OCR endpoint",
           "effect": "ALLOW",
           "principals": [
             "principal://iam.googleapis.com/projects/9876543210/locations/global/workloadIdentityPools/POOL_ID/subject/ns/default/sa/my-ae-agent"
           ],
           "operation": {
             "permissions": ["iap.googleapis.com/resources.egressViaIAP"]
           },
           "conditions": {
             "iap.googleapis.com": {
               "expression": "destination.unregistered.host.endsWith('example-ocr.com') && destination.unregistered.path.startsWith('/v2/process') && destination.unregistered.method == 'POST'"
             }
           }
         }
       ]
     }
   }
   

2. To create the policy, run the following curl command:

   curl -X POST \
   "https://iam.googleapis.com/v3beta/projects/PROJECT_ID/locations/global/accessPolicies?accessPolicyId=POLICY_NAME" \
   -H "Authorization: Bearer $(gcloud auth print-access-token)" \
   -H "Content-Type: application/json" \
   -d @agent-access-policy.json
   

   Replace the following:

   • POOL_ID: the workload identity pool ID. Depending on your pool type, format the ID as follows:
   • Google-managed pool: PROJECT_ID.svc.id.goog
   • Self-managed pool: POOL_NAME.global.POOL_HOST_PROJECT_NUMBER.workload.id.goog
   • PROJECT_ID: the project ID
   • POLICY_NAME: the policy name

To activate the policy, bind the Access policy to your project.