Execution pattern
Agents call connections_list, filter by provider or environment, then use connection_exec with the exact returned connection name.
Connections are secure runtime targets that hold credentials, defaults, and usage guardrails for external systems. Agents discover assigned connections, choose the exact connection by name, and execute through policy, approval, and sandbox checks.

What this page covers
Supported targets include cloud, Kubernetes, SSH, Google Workspace, GitHub, SharePoint, databases, monitoring, ITSM, and custom systems.
Scope access with selected agents, optional workflows, command families, user grants, group grants, and service principals.
Use selected-agent scope for normal operation; all-agents scope should be rare and reviewed with command-family limits.
Operational detail
Connections are secure runtime targets that hold credentials, defaults, and usage guardrails for external systems. Agents discover assigned connections, choose the exact connection by name, and execute through policy, approval, and sandbox checks.
Agents call connections_list, filter by provider or environment, then use connection_exec with the exact returned connection name.
Secrets are managed server-side and materialized only for authorized execution, with temporary files for keys or tokens when needed.
Use stable names that include provider, customer, account, environment, or purpose so agents select the right target.
Related pages