Playbooks · Teams
Google's Organization Policy MCP server is GA
Written November 14, 2026
Google's Organization Policy MCP server is generally available: agents can list constraints, read effective policy and manage exemptions across the hierarchy.
What changed
Google has made its Organization Policy MCP server generally available, alongside natural-language constraint authoring powered by Gemini. Organization Policy is the Google Cloud service that sets guardrails on resource configurations across the resource hierarchy, from a catalog of over 100 predefined policies, plus custom constraints for rules the catalog does not cover. Until now, an administrator who wanted to know what actually applies at a project had to work through the hierarchy chain by hand. The MCP server lets agents and pipelines work with Organization Policy directly.
Google runs the server itself. The Organization Policy MCP reference gives a single global endpoint, https://orgpolicy.googleapis.com/mcp, and states the setup plainly: MCP servers must be enabled and authentication set up before use. The reference describes the server's job as giving tools to discover, manage and audit organization policies and custom constraints across Google Cloud resources.
The tool surface
Google's announcement groups the tools by the job they do:
- Constraint discovery:
ListConstraintsandGetCustomConstraintlet an agent work out which guardrails apply to a project, a folder or an organization. - Policy evaluation:
GetEffectivePolicycalculates the effective policy after hierarchy inheritance and tag-based conditions are taken into account. Google's stated use is testing whether a configuration change will be compliant before it is made. - Managing policy:
CreatePolicyandCreateCustomConstraintlet platform teams manage guardrails through conversational workflows.
The reference lists twelve tools in total. Alongside the five above it names ListPolicies, GetPolicy, ListCustomConstraints, UpdatePolicy, UpdateCustomConstraint, DeletePolicy and DeleteCustomConstraint. Two behaviours in the reference matter for anyone handing these tools to an agent. An update performs a full overwrite of the policy or constraint. Deleting a policy makes the resource inherit from its parent again, or fall back to the constraint's default behaviour.
The work Google frames for it
Google names three kinds of work for the server: posture audits, drift detection and exemption management. Its example combines a coding agent with the MCP tools to build a custom dashboard that shows the constraints applied to an organization and picks out the folders and projects that have been exempted from a policy. The same tools can sit inside custom agents and skills a team builds to manage Organization Policy its own way. In each case the point is the same: the agent reads the hierarchy as it stands, instead of an administrator reconstructing it level by level.
Gemini writes the custom constraint
Custom constraints are where the second half of the announcement lands. Writing one used to mean writing the rule in common expression language (CEL), which calls for a close understanding of Google Cloud resource schemas. Administrators can now describe the constraint in natural language through a Gemini prompt and refine it in conversation. Google's examples include requiring agent-identity in Agent Runtime, requiring agents there to use an approved Agent Gateway, and stopping users from turning off auto-upgrades on GKE. Gemini can also generate compliant and non-compliant test resource configurations, so a constraint can be tested in dry run mode before it is enforced. For teams that would rather start from an existing rule, Google Cloud Professional Services maintains a sample custom constraints library built on CIS and PCI standards.
Keep the read side and the write side apart
The tool list mixes two very different kinds of access. The discovery and evaluation tools answer questions: what constraints exist, what a custom constraint says, what policy is effective at a resource once inheritance is worked through. The create, update and delete tools change the guardrails themselves, and an update overwrites what was there. A sensible first deployment gives an agent the read side for audits and exemption work, and treats the write tools as a separate decision. Where a write is allowed, Google's own material points at the guardrail that already exists: test the constraint in dry run mode before enforcing it.
Related playbooks
- Iterable's hosted MCP server reaches your AI tools
- Retool's MCP server is generally available
- Sherweb's MCP server gives MSPs their own data
- Fulcrum's MCP sign-in goes OAuth (alpha)
Sources
- AI-powered cloud governance: What's new in Organization Policy — Google Cloud Blog
- MCP Reference: orgpolicy.googleapis.com — Google Cloud Documentation
Back to: Teams playbooks hub · Also: Copilot's model-policy API is GA at four levels · Retool's MCP server is generally available · Sherweb's MCP server gives MSPs their own data