How to Connect Google Ads and Meta Ads to AI with MCP: Steps, Benefits and Risks

A year ago, getting an AI assistant to look inside a Google Ads account meant exporting CSVs and pasting them into a chat. Today both Google and Meta publish official MCP servers: connectors that let Claude, ChatGPT, Cursor and other AI tools query your ad accounts directly, in plain English.

That changes how paid media work gets done. It also opens a door that deserves a lock. This guide covers what MCP is, how to connect Google Ads and Meta Ads step by step, what the connection is genuinely good for, and the risks you need to manage before an AI touches a live budget.

Quick answer: MCP (Model Context Protocol) is an open standard that lets AI assistants call external tools. Google's official Google Ads MCP server is self-hosted and read-only: you run it on your own machine with Google Cloud credentials and it answers reporting questions. Meta's official Ads MCP server is hosted at mcp.facebook.com/ads: you add it as a connector, sign in with Facebook, and the AI can report on and also create and edit campaigns. Connect with the narrowest access you can, and keep a human approving every change.

Key takeaways

• Google Ads MCP: official, open source, self-hosted, read-only. Needs a Google Cloud project, OAuth credentials and an approved developer token.
• Meta Ads MCP: official, hosted at mcp.facebook.com/ads, read and write. Sign in with Facebook; no developer app needed. Open beta since 29 April 2026.
• Benefits: instant reporting, cross-platform comparisons, faster audits and more strategist time for decisions.
• Risks: unintended spend, confident but wrong analysis, over-broad access, prompt injection and exposed credentials.
• Controls: start read-only, share only the accounts needed, create new campaigns paused, confirm every change and keep an audit log.

What is MCP in Google Ads and Meta Ads?

MCP is a common plug that lets an AI assistant use outside tools and data. Anthropic introduced the Model Context Protocol in 2024, and it has since become the standard way AI apps connect to software. An "MCP server" wraps a product's API as a set of tools the AI can call: "list accessible accounts", "run this report", "create this ad set".

For advertisers, that means you can ask questions such as:

• "Which search terms spent more than ₹5,000 last month without a conversion?"
• "Compare cost per lead across my Meta ad sets for the last 14 days and flag anything above target."
• "Show me every Google Ads campaign whose budget ran out before 6pm this week."

The AI writes the query, calls the platform's API through the MCP server, and explains the result. No exports, no pivot tables.

What is the difference between the Google Ads MCP and the Meta Ads MCP?

The main difference is control: Google's official Google Ads MCP server only reads data and runs on your own machine, while Meta's official Ads MCP server is hosted by Meta and can create and edit campaigns.

How do you connect Google Ads to Claude with MCP?

To connect Google Ads to Claude, enable the Google Ads API in a Google Cloud project, authorise your Google account with the Google Ads scope, and add Google's open-source google-ads-mcp server to your AI client's MCP configuration. The seven steps below take 30–60 minutes the first time.

Google's server runs locally, so the work is mostly getting credentials in place. You need a Google Ads account (ideally a manager account) and access to Google Cloud.

Step 1: Create a Google Cloud project and enable the Google Ads API

In the Google Cloud console, create a project (or reuse one), open APIs & Services → Library, search for Google Ads API and click Enable.

Step 2: Create OAuth credentials

Go to APIs & Services → Credentials → Create credentials → OAuth client ID, choose Desktop app, and download the JSON file. If Google asks you to configure the OAuth consent screen first, do that, and add your own Google account as a test user.

Step 3: Get a developer token

In your Google Ads manager account, open Admin → API Center and apply for a developer token. A new token starts with test-account access; to query real accounts you need at least Explorer access, which Google grants on application. Recent versions of the server treat the token as optional at startup, but production accounts will still reject queries without an approved one.

Step 4: Authorise your Google account

Install the Google Cloud CLI and create Application Default Credentials that include the Google Ads scope:

gcloud auth application-default login \
  --scopes https://www.googleapis.com/auth/adwords,https://www.googleapis.com/auth/cloud-platform \
  --client-id-file=PATH_TO_YOUR_CLIENT_JSON

A browser window opens. Sign in with the Google account that has access to your Google Ads accounts and approve the request. The command prints the path of the credentials file it saved.

Step 5: Install Python and pipx

The server is a Python package run through pipx. Install Python 3.10 or later, then run pip install pipx and pipx ensurepath.

Step 6: Add the server to your AI client

Add this block to your client's MCP configuration (Claude Desktop, Claude Code, Cursor and VS Code all use the same mcpServers format):

{
  "mcpServers": {
    "google-ads-mcp": {
      "command": "pipx",
      "args": ["run", "--spec", "git+https://github.com/googleads/google-ads-mcp.git", "google-ads-mcp"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "PATH_TO_CREDENTIALS_JSON",
        "GOOGLE_PROJECT_ID": "YOUR_PROJECT_ID",
        "GOOGLE_ADS_DEVELOPER_TOKEN": "YOUR_DEVELOPER_TOKEN",
        "GOOGLE_ADS_LOGIN_CUSTOMER_ID": "YOUR_MANAGER_ACCOUNT_ID"
      }
    }
  }
}

Use your manager account ID without dashes for GOOGLE_ADS_LOGIN_CUSTOMER_ID. Restart the client.

Step 7: Test it

Ask: "List the Google Ads accounts I can access." The assistant should call list_accessible_customers and return your customer IDs. Then try a real question, such as "Show me last month's cost, clicks and conversions by campaign for account 123-456-7890."

The official server exposes three tools: list_accessible_customers, search (which runs Google Ads Query Language reports) and get_resource_metadata (which tells the AI which fields exist). It cannot change bids, pause campaigns or create ads. If you want an AI to make changes in Google Ads, you need a third-party server or your own build on the Google Ads API.

How do you connect Meta Ads to Claude with MCP?

To connect Meta Ads to Claude, add https://mcp.facebook.com/ads as a custom connector, sign in with Facebook, and choose the ad accounts to share. No developer app, API keys or code are needed.

Meta launched its Ads AI Connectors in open beta on 29 April 2026 (Meta announcement). The MCP server is hosted by Meta, so there is nothing to install.

Step 1: Check your access

You need a Facebook profile with access to the business portfolio and ad accounts you want to connect. To set rules for what AI agents may do (Step 5), you need full control of the business portfolio.

Step 2: Add the connector to your AI tool

• Claude (web or desktop): open Settings → Connectors → Add custom connector, paste https://mcp.facebook.com/ads and click Add.
• Claude Code: run claude mcp add --transport http meta-ads https://mcp.facebook.com/ads.
• ChatGPT, Cursor and other MCP clients: add a custom MCP connector pointing at the same URL.

Step 3: Sign in with Facebook and choose what to share

Click Connect. Meta's Facebook Login window opens. Sign in, choose the business portfolios and ad accounts the AI may use, and approve the permissions. Grant only the accounts the assistant actually needs.

Step 4: Test it

Ask: "Which Meta ad accounts can you see?" Then try reporting: "Show spend, results and cost per result by ad set for the last 7 days." The server covers reporting, campaign/ad set/ad creation and editing, catalogs, pixel and dataset diagnostics, A/B tests and lift studies, and activity logs.

Step 5: Set ads MCP server rules

Since 16 July 2026, anyone with full control of a business portfolio can set ads MCP server rules that govern what AI agents are allowed to do on the ad account, from budget changes to catalog updates. Do this before anyone uses write actions. A sensible starting point: allow reporting and drafting, block budget increases and activation.

What are the benefits of MCP for Google Ads and Meta Ads?

• Answers in seconds instead of hours. Questions that used to mean an export, a spreadsheet and half an hour of pivoting become one sentence.
• Cross-platform views. With both servers connected, one assistant can compare Google and Meta cost per lead side by side for the same period.
• Faster audits. Wasted search terms, disapproved ads, budget-limited campaigns, broken conversion signals and overlapping audiences can be surfaced across every account in one pass.
• Better use of specialist time. When the AI does the data pulling, the strategist spends more time on decisions: what to test, where to move budget, what to tell the client.
• Consistent checks. The same audit prompt runs the same way every week, so problems are caught on a schedule rather than when someone remembers to look.
• Faster builds (Meta, or custom Google tools). Drafting ad sets, ad variations and campaign structures from a brief takes minutes, as long as a human reviews them before launch.

What are the risks of connecting AI to your ad accounts?

MCP gives an AI the same power as the API behind it. Treat it like giving a new hire the keys.

• Unintended spend. An AI that can edit budgets or activate campaigns can spend real money on a misunderstanding. A request for "double the budget on the winners" can be read more broadly than you meant.
• Wrong answers stated confidently. An AI can pick the wrong date range, a mismatched conversion action, or misread a metric, and present the result as fact. Check the numbers behind any decision.
• Over-broad access. Connecting every business portfolio and client account when you needed one widens the damage if something goes wrong.
• Prompt injection. Text inside data the AI reads (an ad name, a landing page, a document) can contain instructions aimed at the AI. Assistants that can act as well as read need guardrails for this.
• Credential exposure. Google's local setup involves OAuth credential files and a developer token. Stored carelessly, in a shared folder or a code repository, they grant access to your accounts.
• Client data and compliance. Agencies must be sure clients have agreed to AI tools processing their account data, and that the setup follows Google's API Services User Data Policy and Meta's platform terms.
• Beta changes. Meta's connectors are in beta and tools may change. A workflow that works today may need adjusting next month.

How do you reduce the risks of AI in ad accounts?

1. Start read-only. Use MCP for reporting and audits first. Google's official server is read-only by design.
2. Connect only the accounts you need. Choose specific ad accounts at sign-in rather than everything.
3. Use Meta's MCP server rules to block budget increases and activations by AI.
4. Create everything paused. Ask the AI to create new campaigns, ad sets and ads in a paused state, then review and launch them yourself.
5. Require a preview before every change. Good write tools show exactly what they will change and wait for explicit confirmation.
6. Keep an audit trail. Review Meta's activity log and Google Ads change history regularly, and log every AI action on your side too.
7. Store credentials like passwords. Keep OAuth files and developer tokens out of shared drives and code repositories, and revoke access you no longer use from Google Account permissions and your Meta business settings.

Is PPCPartner an MCP-driven Google Ads agency?

Yes. PPCPartner is an MCP-driven Google Ads agency: AI connected through MCP does the data work, and a human strategist approves every change.

PPCPartner runs its Google Ads management on MCP. Our strategists work alongside AI connected directly to the accounts we manage, so reporting, search term reviews, budget pacing checks and audits that used to take hours now take minutes, across every account.

We built it with the guardrails above, because we manage real client budgets:

• Reads through Google's official Google Ads MCP server, so reporting comes straight from the Google Ads API.
• Changes through our own write layer on the Google Ads API, where every action is a dry run by default and only goes live after a strategist confirms it.
• Deletions are restricted to campaigns our system created and labelled, so nothing a client built by hand can be removed by an AI.
• Every change is logged in an audit trail, on top of Google Ads' own change history.
• Meta Ads through Meta's official Ads MCP server, with new campaigns created paused for human review.

The AI does the data work; people make the decisions and own the outcomes. That is how we deliver faster analysis without handing your budget to a black box. You keep full ownership of your account, and you can revoke our access at any time. See how we handle Google data in our Privacy Policy.

If you are an agency, the same MCP-driven workflow powers our white-label Google Ads service. To build agents of your own, see our guide to building AI agents for a digital marketing agency.

Want to see what an MCP-driven audit finds in your account? Get a free Google Ads audit.