Docs · Connector reference

Google Cloud connector reference

Operations, inputs, troubleshooting, and questions for the Google Cloud connection.

Set up and verify the Google Cloud connection

Choose one task and its account, and match the reads and actions on the overview page to it before granting provider permissions.
Prepare the connection settings from the overview page, connect the provider in Deixic, and choose which tools your agent can use.
Try a small task with a known result, and compare the returned identity and resource with what the account owner expects.
Review the retrieved information, retain its identifiers, and keep any later change in a separate, explicitly authorized step.

Information your agent can use

Resource	Record identity	Start with	Page coverage
Cloud assets	`name`	Top level	Cursor
Storage security	`name`	Top level	Cursor
Network security	`name`	Top level	Cursor
Identity security	`name`	Top level	Cursor
Iam policies	`name`	Top level	Cursor

Record identity: Match on these fields before joining or deduplicating results; where none is specified, check the returned record first.
Start with: Resolve the parent identifier first.
Cursor: Cursor pagination. Continue from the returned cursor; an initial page is not a complete inventory.

Supported operations

Review the inputs before granting an action to an agent. Required inputs must be supplied, optional inputs narrow the request, and provider permissions apply in addition to Deixic policy.

Read: Retrieves information; check the returned scope and identifiers before use.

Operation	Effect	Approval	Provider permissions	Inputs
Iam Workload Identity Provider Get	Read	By policy	https://www.googleapis.com/auth/cloud-platform	3

Inputs by operation

Iam Workload Identity Provider Get Read

Inputs for Gcp Iam Workload Identity Provider Get
Input	Requirement	Accepted value
`pool_id`	Required	stringMinimum characters: 4Maximum characters: 32
`project_number`	Required	stringMinimum characters: 1Maximum characters: 19
`provider_id`	Required	stringMinimum characters: 4Maximum characters: 32

When the Google Cloud task does not produce the expected result

Connected, but the expected data is missing

Check the identity behind the bearer token connection and its access to the specific resource; a successful connection does not establish access to every object.

Compare the lookup against one known record first, and keep an empty response separate from a permissions error.

The answer covers only part of the data

Check the resource’s page coverage, query filters, and result limit, and have the agent state the account and time window it inspected, so a partial response yields a scoped answer.

The agent keeps doing the same work

Inspect the action history for repeated targets and inputs, keep the successful result’s identifiers with the task, and refresh a lookup only when the information changed.

The next step needs a write

Use a separate authorized action or a person to make the change; broader credentials do not add a write operation to this connection.

See how an action proceeds and how to review an approval.

Questions about the Google Cloud connector

Can I use my existing agent with Google Cloud?

Yes: connect your agent to Deixic through MCP, choose which connected tools it can use, and Deixic checks calls routed through its gateway; Dex is available if your team also wants an agent.

Can an agent change data in Google Cloud?

This connector supports reads. Use a configured connection to retrieve information and prepare a recommendation; it provides no action for changing provider data, even if the provider’s API offers writes.

What do I need to connect Google Cloud?

Prepare Short-lived access token (managed exchange or supplied alternative), Scope type (projects, folders, or organizations), Project ID or project, folder, or organization number; the connection uses Bearer token, and the connected identity’s provider permissions determine which resources it can access.

Does connecting Google Cloud keep every answer up to date?

No: an earlier answer is still an earlier answer, so for a decision that depends on current state, have the agent refresh the source and retain the lookup time, and save stable instructions separately from facts that can change.

Google Cloud connector overview Docs All connectors