Why a paid search team wants Google Ads in ChatGPT
Jake Morgan runs paid search for a home goods brand. Every Monday his team rebuilds the same answers. They pull last week's spend by campaign. They find the search terms that cost money without a conversion. They check which budgets ran out before evening. Each answer is a report, an export and a spreadsheet. Google Ads in ChatGPT puts those questions in the chat the team already has open.
MCP (Model Context Protocol) is the standard way an AI assistant calls tools in other apps. Google documents its own Google Ads MCP server, which lets AI agents retrieve and analyze account data, create assets, upload videos and manage labels (developers.google.com, checked November 2026). It is a Python server that a team runs over stdio or on Cloud Run.
Elaichi's googleads connector takes a different route. It signs in to Google Ads with OAuth, the sign-in that lets an app act for an account without seeing its password. It carries reads and writes across campaigns, ad groups, ads, budgets, conversions and account access. The marketing team needs the reads. The writes are where money moves.
Which Google Ads role should the connecting account hold?
Read-only, when the team only needs reports. Google's help page lists five access levels on a Google Ads account: Email-only, Billing, Read-only, Standard and Admin (support.google.com, checked November 2026). Read-only users can view campaigns, use planning tools, and edit and run performance reports. Only Standard and Admin users can edit campaigns. Only Admin users can give account access.
The role matters because Google does not split access by OAuth scope. Google's access-model page says the API uses the same OAuth scopes for read-only and read-write operations (developers.google.com, checked November 2026). Instead, the API follows the user roles Google Ads supports. By Google's model, a connection signed in by a Read-only user carries Read-only's limits.
That makes the Google account behind the connection the first control. Connect with a Google account that holds Read-only on the ad accounts the team reports on. If the person connecting needs Standard access for their own work, use a separate Google account for the shared connection. Elaichi's rules then sit on top as the second control.
Which googleads tools answer the Monday report?
Six account and structure reads plus two query tools cover a reporting desk. googleads_customers_list_accessible_customers returns the Google Ads accounts the signed-in user can reach directly. list_all_googleads_campaigns returns campaigns with their budget, bidding strategy and campaign group. list_all_googleads_ad_groups, list_all_googleads_ad_group_ads, list_all_googleads_campaign_budgets and list_all_googleads_bidding_strategies fill in the structure under them.
Metrics come from the query tools. googleads_google_ads_search and googleads_google_ads_search_stream run any Google Ads Query Language (GAQL) query, and both are tiered read. Google's reporting overview says GAQL reports reach from a whole campaign down to the keywords that triggered an ad (developers.google.com, checked November 2026). Spend by campaign, cost per click by keyword and budget pacing are each one query. The search tool pages through results, and the stream tool returns every row in one response.
Search terms carry one trap. Google's search_term_view aggregates metrics by search term at the ad group level and leaves out Performance Max data (developers.google.com, checked November 2026). Google points to CampaignSearchTermView for Performance Max. An answer drawn only from the first view misses those search terms, so ask for both.
Which Google Ads tools spend money or pause campaigns?
Every tool that changes a campaign, a bid, a budget, an ad or a conversion action stays restricted for the marketing role. None of them is tiered destructive in Elaichi's catalog. The connector's destructive tools remove links, invitations, video uploads, text customization or an operation record. Removing a campaign goes through create_a_googleads_campaign, which is tiered write.
| What it changes | Tools | For the marketing role |
|---|---|---|
| Reports | googleads_google_ads_search, googleads_google_ads_search_stream, list_all_googleads_campaigns, list_all_googleads_campaign_budgets |
Allow |
| Campaigns and their status | create_a_googleads_campaign |
Restrict |
| Budgets | googleads_campaign_budgets_mutate, googleads_account_budget_proposals_mutate |
Restrict |
| Bids, keywords and ads | create_a_googleads_ad_group, googleads_ad_group_criteria_mutate, create_a_googleads_ad_group_ad, googleads_ads_mutate |
Restrict |
| Conversion tracking | googleads_conversion_actions_mutate |
Restrict |
| Bulk and applied changes | googleads_google_ads_mutate, googleads_recommendations_apply, googleads_campaign_drafts_promote, googleads_experiments_promote |
Restrict |
| Account access | googleads_customer_user_access_mutate, googleads_customer_user_access_invitations_mutate |
Restrict |
Two rows deserve a closer look. create_a_googleads_campaign reads like a create, but it calls campaigns:mutate to create, update or remove campaigns. Google's campaigns overview says a campaign's status is ENABLED or PAUSED, and that campaigns change through that same mutate service (developers.google.com, checked November 2026). Pausing or enabling a campaign therefore runs through this tool.
googleads_google_ads_mutate applies many kinds of operation in one atomic request, across campaigns, ad groups, ads, budgets and criteria. Restricting the single-purpose tools while it stays open leaves every one of those changes reachable.
How to hold the marketing role to reports in Elaichi
An allow rule on the marketing role that names the read tools is the control that holds. A restriction is a rule naming which connectors and which individual tools a target may reach. In Elaichi, restriction targets are role or user only; the organization default is the absence of a rule, which means allow everything. So marketing needs its own role, and each marketer holds that one role.
Name the six reads and the two query tools in the rule. The same rule, or a second allow rule on the role, must also name every other app marketing uses, as whole connectors. In Elaichi, once a role carries any allow rule, every connector and tool none of its allow rules names is blocked for that role. A HubSpot or Google Analytics connector left off the list is lost to the role. Do not name the googleads connector whole and some of its tools in one rule, because Elaichi refuses that overlap.
Block rules alone are the weaker choice on this connector. In Elaichi, a layer with only block rules has no allowlist, so everything it does not block passes it. On a connector with this many write tools, a block list that misses one leaves it open.
Add block rules on the tools in the table anyway, as a second line. Within a layer, a block always beats an allow. If someone later widens the allow rule to all of Google Ads, those tools stay restricted. Block rules match a tool's name or its operation, and how block and allow rules match tools explains the difference. A restriction change takes effect within about two minutes.
Connecting one Google Ads account for the whole team
One marketer connects the googleads connector with the Read-only Google account, then shares the connection with the team at use. Every call through a shared connection uses that one connection's credential, whoever makes it. In Elaichi, a call from an MCP client is recorded under the person who signed in, with surface mcp and the OAuth client named. So the audit trail still names the marketer behind each report.
Pick an owner who is staying. When a member leaves, Elaichi's offboarding lists the connections they own, and an admin transfers each one to another member. A transfer sets a new owner and leaves every share as it was. Offboarding when the agent holds access covers the rest of that list.
Every organization uses one address: https://api.elaichi.ai/mcp. On ChatGPT Business, only admins and owners create and publish custom apps for the workspace, and members then connect the published app (help.openai.com, checked October 2026). Each marketer signs in to Elaichi with their own OAuth grant. Connecting Elaichi to ChatGPT has the steps. In Elaichi, connected tools are never listed one by one, however few there are. So a ChatGPT control that gates tools by name acts only on execute_tool and its siblings as wholes. The marketing role's rules are the control for each Google Ads tool.
What to check before the team relies on it
Sign in from a test account on the marketing role and ask ChatGPT for last week's spend by campaign. The query should run. Then search for "budget" and "mutate". A restricted tool is left out of the tools Elaichi offers a client and cannot be called. Search still names it, flagged restricted, with no schema.
Then read the audit trail. Elaichi writes one entry for each connected-tool call that reaches execution (a call refused earlier writes none). Each entry records the tool, the connection reached, the outcome and the client, plus the one path argument that names the object, as the target id, and nothing else about the arguments. What an AI audit log must capture lists what to check in those rows.
Mind the quotas on both sides. Google counts each Search or SearchStream request as one operation against a daily quota (developers.google.com, checked November 2026). Google sets that quota per Google Cloud project, from 2,880 operations a day at Explorer access to unlimited at Standard (developers.google.com, checked November 2026). Elaichi allows 120 MCP requests every 60 seconds per OAuth grant.
When a marketer needs a restricted tool for a real change, they file an access request. An admin approves it in the console, under Governance, then Access requests. The grant lifts that block for that one person, and the role is not edited.
When Google's own MCP server covers the job
If one analyst wants reports and asset work in their own accounts, Google's Google Ads MCP server may be all they need. Google's guide describes its mode as read and partial mutate, with a configurable read-only mode (developers.google.com, checked November 2026). Its mutate tools propose a change and validate it before anything commits. It cannot create, modify, pause or enable campaigns, ad groups, ads, budgets or bids. For a reporting desk, that limit is a safety property in itself. It needs a Google Cloud project with the Google Ads API turned on, and production accounts need Explorer, Basic or Standard access.
Elaichi earns its place when several people share the work and the question crosses apps. Spend sits in Google Ads, pipeline sits in HubSpot and sessions sit in Google Analytics. One endpoint serves ChatGPT, Claude, Cursor and Microsoft Copilot alike, with one set of rules and one audit trail. The CRM side is in HubSpot for marketing in ChatGPT, and site traffic is in a Google Analytics MCP server for marketing. If one person only wants to try a few reports, you may not need a control plane yet. The rest of the stack is in the 900+ connectors.