Resources · Claude connector

Claude Google Compute Engine connectorWhat Claude can do in your Google Compute Engine account.

The Claude Google Compute Engine connector exposes 29 tools. 23 read your infrastructure, 6 launch, power, resize or delete VMs, 0 are undocumented. Here: what Claude can inspect, what it can change, and the Google Cloud roles that gate it.

Verified Trustpilot reviews · AI, automation & growth agency

Overview

What the Claude Google Compute Engine connector changes for cloud teams

Answering a simple infrastructure question in Google Cloud often means clicking through the console or remembering gcloud flags. With this connector, you ask Claude in plain words: which VMs are running in a zone, which disks sit idle, what a reservation covers. It reads your Compute Engine project itself and, when you ask, can launch, power or delete machines too.

An inventory in one question. A team lead preparing a cost review asks what is running. list_instances returns the VMs of a zone, list_disks shows each disk's last attach time, and Claude points out the ones nobody seems to use.

A quick fix on a stuck machine. An on-call engineer sees a frozen server. reset_instance restarts it, then get_zone_operation confirms the operation reached DONE without error.

Capacity and commitments explained. Before a launch, list_reservations and list_commitments show what capacity is held and when commitments end, so finance and engineering read the same picture.

What it does not do: the sheet exposes no tool that changes disks, snapshots, templates, reservations or commitments; those are read-only here, and the machine type can only change on a stopped VM. Google also notes that calls may be routed through any region. Nothing runs by itself either: Claude acts while it answers you, never on a schedule. Scheduled shutdowns belong to an automation tool, a different job presented on the Integrations hub.

Vocabulary

The vocabulary in one minute

Five terms you will see while connecting Compute Engine.

Connector
The link you set up once between Claude and an account you already own, so Claude can work in it while it answers you.
Tool
One named action a connector gives Claude. Claude picks the ones it needs; the directory sheet lists every one by its exact name.
Authorization
Google's own sign-in screen, where you hand Claude the access it will use. Granted once per person, and you can revoke it later.
Approval
The confirmation Claude waits for before it goes ahead with something that changes your project, shown in the chat when it matters.
MCP
The shared standard connectors are built on: it is what lets an assistant like Claude talk to an outside service such as Google Cloud.
Connect

Connect Google Compute Engine to Claude in three steps

  1. 01

    Find the connector

    Open Claude's settings, go to Customize then Connectors, and look for Google Compute Engine. On a Team or Enterprise workspace, an Owner or Primary Owner has to enable it for the organization before members can sign in.

  2. 02

    Sign in with Google

    Click Connect on its row and sign in with your Google account in the window Google opens. If the connection ever breaks, use Disconnect and connect again.

  3. 03

    Check the scopes you grant

    Read Google's authorization screen before you accept. It belongs to Google, not to Claude, and sets what the access covers; Google lists a read-only scope and a read-write one for Compute Engine. You can revoke it later from your Google account.

Tools

The 29 tools, sorted by what they do

Google Compute Engine gives Claude 29 tools: 23 that read your account, 6 that change something in it.

Twenty-three tools read, six change your machines. Names stay exactly as Claude shows them.

  • 23 read
  • 6 write

What Claude reads (23)

23 tools

Twenty-three tools that look at VMs, groups, templates, disks, reservations, commitments and operations without touching them.

get_instance_basic_info

Pulls the essentials of one VM: name, ID, status, machine type, when it was made and any attached accelerators. Claude uses it to answer a quick question about a single machine.

When it helps
you want to know whether the staging server is running and which machine type it uses.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_instance_attached_disks

Shows every disk attached to a given VM, with details such as type, mode, boot flag, size, device name and architecture.

When it helps
before resizing a database server, you check which disk is the boot disk and how big the data disk is.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_instances

Returns the VMs of a project in one zone, each with its name, ID, status, machine type and accelerators. It is the usual starting point for an inventory.

When it helps
a team lead asks which machines are up in a zone before the monthly cost review.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_instance_group_manager_basic_info

Describes a single managed instance group: the template it follows, its base instance name, target sizes for running, stopped and suspended instances, and its current status.

When it helps
you want to confirm which template the web tier's group currently relies on before a release.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_instance_group_managers

Lists the managed instance groups of a project in one zone, each with its template, base name, target sizes and status, so Claude can compare them at a glance.

When it helps
you are auditing which autoscaled groups exist across a zone.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_managed_instances

Lists the instances inside one managed group, with each one's ID, URL, status and current action, which shows what the group is doing right now.

When it helps
a rollout seems stuck and you want to see which members are still recreating.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_instance_templates

Brings back a project's instance templates with name, ID, description, machine type and region, giving Claude the full catalogue of blueprints your teams rely on.

When it helps
you need to find the template your team built for GPU workloads last quarter.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_instance_template_basic_info

Gives the short version of one instance template: name, ID, description, machine type and region. Enough for Claude to confirm what a template is meant for.

When it helps
two templates have similar names and you want their descriptions side by side.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_instance_template_properties

Opens the full instance properties of a template: tags, machine type, network interfaces, disks, metadata, service accounts, scheduling options, labels, accelerators, reservation affinity and shielded or confidential settings.

When it helps
a security review needs to know which service account a template grants its VMs.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_disk_basic_info

Reports one disk's name, ID, description, size, type, status, and the last moments it was attached or detached from a machine.

When it helps
you found an unattached disk on the bill and want to know when it was last in use.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_disk_performance_config

Reads the performance profile of a disk: type, size, provisioned IOPS and throughput, physical block size, storage pool and access mode.

When it helps
an application feels slow on disk and you want the provisioned throughput before blaming the code.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_disks

Returns all disks of a project in one zone, each with size, type, status and last attach or detach time, so Claude can spot the idle ones.

When it helps
a finance controller asks for disks that have sat unattached for a while.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_accelerator_types

Lists the accelerator types offered in a zone, including whether each is deprecated and the maximum number of cards per instance.

When it helps
a data scientist wants to know which GPUs a zone offers before planning a training run.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_machine_types

Returns the machine types available in a zone with their guest CPUs, memory, accelerators and architecture. It needs a project and a zone to answer.

When it helps
you are sizing a new service and want the options with enough memory in your zone.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_images

Shows a project's images with name, ID, status and family, which tells Claude which operating system images your teams have on hand to boot from.

When it helps
you want to know whether the hardened image your platform team maintains is still marked ready.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_zone_operation

Reads the state of a zone operation: status, errors, warnings and HTTP error details. Google's reference tells Claude to use it to confirm that a VM action finished with status DONE.

When it helps
after a VM restart request, you want proof it completed without error.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_reservation_basic_info

Gives the basics of one capacity reservation: its zone, its status, whether a specific reservation is required to use it, and the commitments linked to it.

When it helps
you want to know whether a reservation is tied to a commitment before releasing capacity.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_reservation_details

Returns the detailed view of a reservation: machine type, accelerators and local SSDs it covers, aggregate properties such as VM family, sharing settings and resource status.

When it helps
a platform engineer checks whether a reservation is shared with another project.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_reservations

Lists the reservations of a project in one zone with status, required-reservation flag and linked commitments, which gives Claude a quick overview of held capacity.

When it helps
before a product launch, you review what capacity is already held in the target zone.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_commitments

Returns the committed-use contracts of a project in one region, with their status, plan, type, the resources they cover and their begin and end dates.

When it helps
procurement wants to see which commitments expire next quarter so renewals can be planned in time.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

get_commitment_basic_info

Summarizes a single commitment: status, plan, type, the resources it covers and the dates that frame it. Claude can then explain what was committed, and until when.

When it helps
a finance analyst asks what one specific commitment actually covers before the budget meeting.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_commitment_reservations

Shows which reservations sit under a given commitment, with the machine type, accelerators, local SSDs, sharing settings and resource status of each one.

When it helps
you want to know how a commitment's capacity is split across reservations and projects.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

list_snapshots

Gathers the snapshots of a project with status, disk size, storage bytes and source disk, so Claude can tell what is backed up and from which disk.

When it helps
before an upgrade, you check that the database disk has a recent snapshot.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

What Claude changes (6)

6 tools

Six tools that launch, power, resize or delete virtual machines. No source describes a confirmation specific to any of them, so the general rule below applies.

create_instance

Approval: see the rule

Spins up a new virtual machine from a project, a zone and a name. Without further detail, Google says it defaults to the e2-medium machine type and a Debian image.

What Claude asks for
no source describes a confirmation for this tool, so the general approval rule applies.
When it helps
a developer needs a throwaway test machine in the same zone as staging.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

delete_instance

Approval: see the rule

Deletes a virtual machine from a project and zone. Google tags this tool as destructive in its reference.

What Claude asks for
the sources describe no confirmation specific to this deletion, so it falls under the default rule further down.
When it helps
cleaning up the test machines left behind after a demo.
Watch out
Claude should only treat it as done once the operation reports DONE without errors.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

start_instance

Approval: see the rule

Starts a stopped virtual machine, identified by project, zone and name, and returns an operation to follow.

What Claude asks for
Google documents no confirmation here, so the general approval rule is what covers it.
When it helps
bringing the demo server back up on the morning of a client meeting.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

stop_instance

Approval: see the rule

Stops a running virtual machine. It is also the step that must come before a machine-type change.

What Claude asks for
no confirmation is documented for stopping a VM, so the default rule on approvals applies.
When it helps
shutting down a forgotten GPU machine on a Friday evening to stop the meter.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

reset_instance

Approval: see the rule

Resets a virtual machine and hands back an operation to check.

What Claude asks for
nothing in Google's reference describes a confirmation for a reset, so the general rule is the one that applies.
When it helps
a frozen machine no longer answers on SSH and a restart is the quickest way out.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

set_instance_machine_type

Approval: see the rule

Switches a stopped VM to another machine type, for more or fewer CPUs and memory.

What Claude asks for
no source spells out a confirmation for this resize, so the default approval rule covers it.
When it helps
a reporting server needs more memory for month-end.
Watch out
Google's description only covers a stopped instance, so stop it first.

Sourcedocs.cloud.google.com · April 23, 2026 ↗

Approvals

What Claude asks before it acts

By default, Claude stops and asks for your go-ahead before any action it carries out on an account for you. That covers the six VM tools here, since no source sets a different rule for them, delete_instance included, which Google flags as destructive.

On Team and Enterprise, workspace owners decide whether members may let some actions through without being asked each time, and they can limit what a connector may do for the whole organization. On the Google Cloud side, IAM allow and deny policies can block tools by principal, by tool name or by their read-only attribute. Claude works with your Google Cloud permissions and nothing more: what you cannot do in the console, it cannot do either.

Plans

Which plans it is available on

None of the 819 sheets in the official directory shows plan availability. Connector by connector, that answer is published nowhere: a real gap in the catalogue.

The general rule is published: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile. On Team and Enterprise, an Owner or Primary Owner enables a connector for the organization before members connect. For the up-to-date status of this one, check the Google Compute Engine sheet in the official directory. On Google's side, the right IAM roles on the project are required.

Limits

Where the Claude Google Compute Engine connector stops

A connector is not an automation. Claude calls these tools while it answers you; no VM shuts down on its own at night because of this connector.

Only Google documents this connector; Claude's help center has no article on it. A tool list is an observed floor, not a promise, and the partner badge is not a security audit: Anthropic says on every sheet that it neither chooses the tools a publisher exposes nor guarantees their behaviour. Connect it only if you trust the publisher, here Google. Google adds that the service is switched on when the Compute Engine API is enabled on the project.

Need help

Need help connecting Google Compute Engine to Claude?

A person reads every message.

FAQ

Questions about the Claude Google Compute Engine connector

01What can Claude do with the Google Compute Engine connector?
Claude can list and inspect VMs, managed instance groups, instance templates, disks, images, snapshots, reservations, commitments, machine and accelerator types, and follow zone operations. It can also launch a VM, start, stop or reset one, resize a stopped VM and delete a VM. You ask in plain language, for example which machines are idle in a zone, and Claude picks the tools and answers in the conversation. Every change stays tied to your own permissions.
02Can Claude delete or change my Compute Engine resources?
Yes, six tools change your project, all of them about virtual machines: launching one, deleting one, starting, stopping or resetting one, and changing the machine type of a stopped one. Google flags the deletion tool as destructive. The sheet offers no tool that changes disks, snapshots, templates, reservations or commitments, which Claude can only read. A tool list is still a floor rather than a ceiling, so check your own connector settings.
03Does Claude ask before it launches or deletes a VM?
By default, yes: Claude asks for confirmation before any action it carries out on an account for you. No source describes a confirmation specific to any of the six VM tools, so that default rule covers them. On Team and Enterprise workspaces, owners decide whether members may let some actions through without being asked, and on the Google Cloud side, IAM deny policies can block the tools that are not read-only.
04Which Claude plans is the Google Compute Engine connector available on?
No official source publishes plan availability connector by connector, and none of the directory sheets shows it. The general rule says remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, and that on Team and Enterprise an Owner or Primary Owner enables them first. The connector's sheet in the official directory is the only place showing its current status. Google separately requires IAM roles on the project.
05Does Claude see every project in our Google Cloud organization?
No. Claude works with the Google Cloud permissions of the person who connected and nothing more. Google requires a dedicated tool-call role on the project to call the tools at all, plus the Compute Engine roles needed for each operation, and administrators can add IAM deny policies by principal or tool. Most tools also ask for a project and a zone or region, so Claude looks where you point it.
06Why did Claude stop my VM before resizing it?
Because Google describes the machine-type change for a stopped instance only. To move a VM to a bigger or smaller machine type, Claude has to call stop_instance first, then set_instance_machine_type, and then start_instance once you want it back. Each step returns an operation, and Google tells Claude to continue only when that operation reports DONE without errors. Plan the downtime before asking for a resize. The change itself takes the project, the zone, the instance name and the new machine type.
07Should I use Claude or an automation tool with Compute Engine?
They do different jobs, so this is not a ranking. Claude with this connector suits questions and one-off actions you steer in a conversation: an inventory, a restart, a resize you want explained. An automation tool suits work that must run on a schedule or react to an event without anyone asking, such as stopping test machines every evening. The connector never fires by itself, so unattended jobs belong on the automation side.