Claude for Google Ads: Connect It, Then Ask Better Questions

Anthropic's own one-line distinction is the clearest: MCP gives Claude access to tools, skills teach Claude how to use them. A plugin, in the Google Ads sense, is a bundle aimed at developers. Here is how the three map onto Google Ads.

• Connector (MCP server)

  Gives Claude live access to the account: list accounts, run queries, read campaigns and metrics. Without one, Claude only knows what you paste in.

• Skill

  A folder of instructions, scripts and resources Claude loads when a task needs it, such as how to run an account diagnosis. No data of its own.

• Plugin

  Google's Google Ads API Developer Assistant, for Antigravity and Claude Code. For people writing code against the API, not for reading a report.

The Google Ads skills Google publishes

Google ships agent skills for the Google Ads API, installed with one command. The documentation names three: google-ads-api-mcp-setup, google-ads-api-account-diagnostics and google-ads-api-quickstart. Skills are an open standard (agentskills.io), so the same folder works in Claude Code and other agents that read it.

npx skills add google/skills/skills/ads

In the Claude apps, skills are available on every plan, need code execution turned on, and are enabled under Customize, then Skills. You can also write your own: a skill that holds your agency's audit checklist and naming conventions turns "audit this account" into the same audit every time, whoever asks.

Claude's connector directory lists no ads connector published by Google, so there is no one-click official option. As of October 2026 the choice is between these three.

Adding a self-hosted server in Claude

1. Deploy Google's server to Cloud Run (the steps are in the Google Ads MCP guide) and copy the URL plus /mcp.
2. In Claude, go to Settings, then Connectors, and add a custom connector with that URL. On Team and Enterprise, an Owner adds it for the organisation.
3. Connect and sign in with the Google account that has access to the Google Ads accounts.
4. In a chat, enable the connector and start with: "List the Google Ads accounts you can access."

Custom connectors work on every Claude plan; Free allows one. In Claude Code the same deployment is one line, followed by the /mcp command inside Claude Code to sign in.

claude mcp add --transport http google-ads https://<your-cloud-run-url>/mcp

A read-only connector is perfect for an audit: Claude can look at everything and change nothing. Good audit prompts name the date range, the account and what "bad" means to you. Vague prompts get generic advice. These are written to be pasted as they are, with your own thresholds.

1. "For the last 30 days, list every campaign with its budget, bid strategy, spend, conversions and cost per conversion. Flag any campaign limited by budget."
2. "Show the search terms that spent more than $100 in the last 60 days with zero conversions. Group them by theme and suggest negatives, but do not assume any are wrong without showing me the spend."
3. "Which conversion actions are counted in the Conversions column? List each with its category and whether it is primary. Point out any that are not close to revenue."
4. "Compare device performance per campaign for the last 90 days. Where is mobile's cost per conversion more than 50% above desktop?"
5. "List ad groups with only one active ad, and ads that are disapproved or limited."
6. "Find keywords with high spend and a cost per conversion more than twice the account average over the last 90 days."
7. "Check the account structure: campaigns that target the same keywords, broad match keywords without a smart bidding strategy, and paused campaigns still holding budget."
8. "Summarise what changed in performance month over month, and tell me which query you ran for each number."

The last line of that final prompt is a habit worth keeping in every prompt: asking Claude to show the GAQL it ran lets you check the date range and filters, which is where most wrong answers come from.

Reporting prompts work best when they fix the shape of the answer: which columns, which period, what comparison. Then the same prompt next week produces a report you can put next to last week's.

1. "Weekly report: spend, clicks, conversions and cost per conversion per campaign, this week against last week, as a table, with three sentences on what moved."
2. "For a client update, summarise the month in five bullet points a non-marketer can read. No jargon, no metric without its number."
3. "Which five campaigns spent the most this quarter, and what share of total conversions did they produce?"
4. "Across all accounts under this manager account, rank clients by spend change month over month."
5. "Build a table of Performance Max and Search campaigns side by side for the last 30 days, with the same columns for both."

With Google's server, Claude is an analyst with read access. It cannot change bids or budgets, pause or enable anything, or create campaigns, ads or assets. As we read the documentation, it also cannot return keyword ideas or search volumes, because those come from a separate API call the server does not expose. Ask for any of those and a well-behaved Claude will tell you it has no tool for it.

If you want Claude to act, a third-party connector such as Zapier MCP offers actions like setting a campaign's status, billed per call. Before you give an assistant write access to a client account, decide who confirms a change and where it is logged. Many teams keep Claude read-only and make the change themselves, with Claude's analysis as the brief. The comparison of options is in Ads MCP servers compared.

Most "chatgpt google ads plugin" searches end at the same place: ChatGPT connects to MCP servers as custom apps in developer mode. That works with Google's server deployed to Cloud Run, since Google lists the chatgpt.com redirect URI for it, or with a third-party connector that supports ChatGPT.

• Plans: developer mode is on ChatGPT on the web for Plus, Pro, Business, Enterprise and Edu.
• Workspaces: on Business, Enterprise and Edu an admin or owner has to enable developer mode first.
• Requirement: the server has to be a public HTTPS endpoint with streamable HTTP, usually at /mcp. A local stdio server does not reach ChatGPT without a tunnel.

The prompts above work in ChatGPT unchanged. The difference between the two assistants matters less than the difference between the data sources you give them.

Every prompt above returns Google's numbers, and for a lead business Google's conversion is usually a form fill or a call. Claude will rank campaigns by cost per conversion faithfully. It cannot tell you that the campaign with the cheapest leads has the lowest close rate, because Google Ads does not know which leads closed unless you send that back as an offline conversion.

That is the argument of Why your ad platform's MCP gives Claude the wrong CPL, with a worked example. The fix is not a better prompt. It is a second source in the same chat.

LeadJourney tracks every click, form fill and call server-side and first-party, keeps the gclid, and joins it to the deal in your CRM: HubSpot, Salesforce, Pipedrive, Close, Attio, GoHighLevel, ActiveCampaign, Odoo, Zoho CRM or Microsoft Dynamics 365. So each Google Ads campaign carries its qualified leads, booked demos and won revenue next to the spend the Google Ads integration imports, and the stages you map go back to Google Ads as conversions.

The LeadJourney MCP server puts that in front of Claude: one endpoint, https://api.leadjourney.io/mcp, added as a custom connector with OAuth (step by step on LeadJourney for Claude, and for ChatGPT on LeadJourney for ChatGPT). Connected next to a Google Ads server, the audit prompt becomes "rank campaigns by cost per won deal", and the answer comes from the CRM.

• Reads: journeys and touchpoints, spend by platform, campaign and ad set, CRM stages, qualified counts, booked demos and won revenue, channels and landing pages.
• Changes, inside LeadJourney only: reports and dashboards it builds and saves, the attribution model and lookback window.
• Never: pausing a campaign, moving budget, creating ads or editing a CRM record. It acts only when you send a message.

The MCP connection is included from the Growth plan; below it, it is an add-on at €49 a month for advertisers or €99 a month flat for agencies. LeadJourney also has Atlas, its built-in AI analyst in the dashboard and in Slack, for teams that would rather not set up a connector at all.

• Google Ads MCP: the official server, its limits and alternatives
• Ads MCP servers compared: Meta, Google, LinkedIn, TikTok and more
• Why your ad platform's MCP gives Claude the wrong CPL
• Claude for Meta Ads: connector, skills and prompts
• Which ad copy attracts leads that close
• MCP vs API vs Zapier for ad reporting
• LeadJourney for Claude
• The Claude integration