Back to Blog
MCP GuidesJul 23, 202613 min read

Google Ads MCP for ChatGPT and Claude — Start Free

BizSidekick Team
BizSidekick TeamJul 23, 2026

Quick answer

Google Ads MCP lets an MCP-compatible AI client call a defined set of Google Ads tools. The current official Google server is read-only: it can discover accessible accounts, run GAQL searches, and retrieve resource metadata, but it cannot change bids, pause campaigns, or create assets.

That makes the official server useful for reporting and investigation. If you need store context, approvals, or supported write actions, treat those as a separate workflow and verify the exact capabilities of the connection you choose. BizSidekick can run inside ChatGPT or Claude, bring permitted business sources into one task, and prepare supported actions for approval. You can install the Plugin without making a payment.

BS

BizSidekick task

Illustrative workflow

Analysis ready
Which campaigns increased spend while conversion value declined over the last 30 days?

I queried the connected Google Ads customer and compared the current 30-day window with the previous one. Three campaigns match the condition. I prepared the metrics and GAQL scope for review; no account changes were made.

Google Ads accountCampaign report30-day comparison

Largest exception

Non-brand Search · spend +24%

Conversion value declined 11% in the same window. Investigate tracking, search terms, and conversion lag before acting.


What is Google Ads MCP?

MCP stands for Model Context Protocol. In practical terms, a Google Ads MCP server sits between an AI client and Google Ads. It describes the tools the client may call, validates the input those tools require, and returns structured results the model can use in a response or workflow.

The server is not the model and it is not the Google Ads account. It is an adapter with a deliberately limited surface. One implementation might expose read-only reporting. Another might offer write tools. Two products that both say “Google Ads MCP” can therefore have different authentication methods, tool names, data coverage, and action boundaries.

Governed workflow

Analysis can run repeatedly. Material account changes stop for review.

  1. Ask1

    AI client

    ChatGPT or Claude receives a natural-language business question.

  2. Call2

    MCP tool

    The client selects an exposed tool and sends validated arguments.

  3. Query3

    Google Ads

    The server authenticates and requests data for the permitted account.

  4. Explain4

    Structured result

    The client turns the returned fields into an answer with scope and evidence.

This separation is useful because the tool boundary is inspectable. It also means you should not assume the phrase “connected to Google Ads” tells you what is available. Check the server's tools and the account permissions.


What the official Google Ads MCP server can do

Google's current developer-toolkit release exposes three core tools: list_accessible_customers for account discovery, search for GAQL queries, and get_resource_metadata for understanding resources and fields. It runs locally over standard input/output and supports OAuth or service-account authentication, subject to the requirements documented by Google.

Current official release

Official Google Ads MCP tool boundary

Capabilities reflect Google’s documentation reviewed on July 23, 2026. Recheck the release before implementation because the server can evolve.

OptionAvailableTypical useBoundary
Account discoveryYesList customer accounts accessible to the authenticated identityAccess still depends on Google Ads permissions
GAQL searchYesRetrieve campaign, ad group, keyword, and performance fieldsThe query must use valid resources, fields, and date rules
Resource metadataYesInspect fields before composing a queryMetadata does not change the account
Campaign changesNoNot available in the current official serverCannot modify bids, pause campaigns, or create assets

See Google's official Google Ads API MCP Server guide for current setup instructions and limitations. The read-only boundary is an important product fact, not a weakness to hide: it makes the server appropriate for analysis while preventing accidental account mutations through that implementation.


Google Ads MCP server vs a managed AI Plugin

A self-configured MCP server is a good fit when a technical team wants direct control over runtime, credentials, and queries. A managed Plugin is useful when the business task spans multiple systems or needs workspace permissions, approvals, and an action history. The choice is operational, not ideological.

Decision guide

Choose the connection model that matches the work

BizSidekick does not turn the official read-only server into a write server. Any additional source or action must come from a separately supported connection with its own permissions.

OptionOfficial Google MCPManaged Plugin workflowQuestion to ask
Primary jobGoogle Ads discovery and reportingConnected business investigation and governed workIs the task only inside Google Ads?
Setup ownerTechnical operatorWorkspace administrator or operatorWho will maintain credentials and access?
Store contextNot part of the Google serverAvailable when the store is separately connectedDoes the decision require margin, refunds, or inventory?
Write actionsNot supported in the current releaseOnly supported actions allowed by provider and account permissionWhich exact field can change, and who approves it?
Audit modelBuild around client and local runtime logsUse workspace task, approval, and provider-result historyWhere will the team review what happened?

BizSidekick's Google Ads MCP landing page explains the connected workflow. The Google Ads integration page covers the advertising signals and ecommerce use cases.


Use Google Ads MCP in ChatGPT or Claude

The business question can stay the same in both environments. What changes is how the MCP server or Plugin is installed and authenticated. Keep setup instructions separate from the reusable task so the team does not duplicate measurement logic for each AI client.

Run it where you work

The business task stays the same. The setup path changes.

Task in ChatGPT

Compare campaign spend and conversion value for the last 30 days. Return the three largest negative changes with account, campaign, metric, and date range. Do not change the account.

Configure the intended MCP server or install BizSidekick, authenticate the correct account, then verify the available tools before running the task.

  1. 1Add the Google Ads connection or install the BizSidekick Plugin.
  2. 2Authenticate and verify the customer account ID.
  3. 3Inspect the available tool names and read/write boundary.
  4. 4Run the bounded query and review the returned fields.

A Google Ads MCP workflow for ecommerce

Google Ads can report attributed conversion value, but an ecommerce decision may also need realized revenue, refunds, product margin, or inventory. Those fields do not appear merely because the Google Ads server uses MCP. They must come from another permitted source and be joined with an agreed key and time window.

  1. Query campaign spend, conversion value, conversions, and the relevant campaign identifiers from Google Ads.
  2. Retrieve store revenue, refunds, and product costs for the matching reporting window from the separately connected store.
  3. State the attribution and margin definitions before comparing results. Do not silently treat two different revenue measures as equivalent.
  4. Return a short exception list with source, account, campaign, metric, date range, and the reason each item crossed the guardrail.
  5. Prepare any follow-up action separately and stop for approval when the connection supports a material change.

For a detailed profitability example, read Google Ads automation with an AI agent. That guide covers margin, refunds, guardrails, and action review without repeating the MCP setup material here.


Permissions, approvals, and write-action boundaries

The safest default is read-only access to the smallest useful account scope. If a different provider later exposes write tools, enable only the actions the workflow needs. A proposed change should name the account, campaign or resource, field, old value, new value, reason, and reviewer.

Approval required

Review a separately supported budget action

Illustrative action

This illustration shows the information a reviewer should see. The current official Google Ads MCP server cannot execute this change.

Daily budget
$300/day
$240/day
Review label
None
Margin check
Keep current settingsApprove 2 changes

Do not infer write access from an AI response that sounds confident. Confirm the tool, the provider result, and the action log. If the provider does not support the operation, the correct outcome is an explicit limitation and an unchanged account.


Start with one Google Ads MCP task

A good first task is read-only, bounded, and easy to check in the Google Ads interface. It names the date window, comparison, requested output, and no-change boundary.

Copy and run

Read-only Google Ads MCP review

Review the last 30 days of Google Ads campaign performance. Find campaigns with rising spend and falling conversion value. Explain the evidence and prepare a review list. Do not make account changes.

Run the prompt once, inspect the customer account and returned metrics, and save the useful query. Only then add store context or recurring execution. If you prefer a managed setup, open the BizSidekick Plugin page and install the Plugin without making a payment. Product usage and subscription terms are separate from installation.


Frequently asked questions

What is a Google Ads MCP server?

A Google Ads MCP server exposes defined Google Ads tools to an MCP-compatible AI client. The current official Google server supports account discovery, GAQL search, and resource metadata through a read-only interface.

Can the official Google Ads MCP server change campaigns?

Not in its current release. Google describes the official server as read-only and says it cannot modify bids, pause campaigns, or create assets. Write capabilities offered by another connection should be evaluated separately.

Does Google Ads MCP work with Claude?

Yes, when the client supports MCP and the server is configured with the required authentication. BizSidekick also provides a Plugin setup path for Claude so the business task can use connected workspace permissions and approvals.

Does Google Ads MCP work with ChatGPT?

Yes, when the Google Ads MCP server is available through a compatible ChatGPT connection and configured with the required authentication. BizSidekick can also be installed as a Plugin in ChatGPT for connected ecommerce workflows.

Is MCP the same as the Google Ads API?

No. The Google Ads API is the underlying programmatic interface. An MCP server exposes a selected set of tools in a format an MCP-compatible AI client can use. The server decides which API operations are available.

Do I need a separate MCP server for store data?

A standalone Google Ads MCP server only provides the tools it implements. If the task needs orders, refunds, inventory, or product margin, connect those sources separately or use a managed workflow that brings the permitted sources together.

Can I install BizSidekick without paying?

Yes. No payment is required to install the BizSidekick Plugin. Product usage and subscription terms are separate and are shown on the Plugin page.

Share this article

BizSidekick operator call to action background

Install the Plugin. Start with Read-only Google Ads Analysis.

No payment is required to install. Connect only the account you intend to use, verify the available tools, and keep material changes behind an approval.

Install Free