Google Ads MCP Server: One Hosted Server for Google Ads, Meta Ads, and GA4

What This MCP Server Does and Who It Is For

The Model Context Protocol (MCP) is the open standard that lets AI assistants use external tools. This repository is an MCP server that exposes three advertising platforms as tool sets within a single connection: the Google Ads API, the Meta Marketing API (covering Facebook and Instagram ads), and the Google Analytics 4 Data API. A fourth platform, Google Search Console, is also included.

The target users are marketers and agencies who work across Google Ads and Meta Ads daily and want to manage campaigns, pull performance reports, research keywords, upload creatives, and build audiences from inside an AI assistant rather than navigating multiple dashboards. The server is built and operated by Ryze AI, and the repository serves as documentation and a public presence for the product.

Connecting takes a single command in Claude Code:

claude mcp add ryze --transport http https://connector.get-ryze.ai/mcp

For claude.ai, Claude Desktop, ChatGPT, and Cursor, the process is: Settings > Connectors > Add custom connector, then paste `https://connector.get-ryze.ai/mcp` and sign in with the Google or Facebook account that owns the ad accounts. No API keys, no developer token, and nothing to install locally.

How the Server Works: Remote MCP over Streamable HTTP

The server runs as a remote MCP endpoint over Streamable HTTP, meaning nothing runs on the user's machine. All tool calls go to `https://connector.get-ryze.ai/mcp`, which proxies them to the respective platform APIs. Authentication uses OAuth 2.1 with PKCE: the user signs in with their Google or Facebook account during setup, and the server holds a delegated token that it uses to make API calls on their behalf.

Reads are immediate: a tool call to list campaigns or pull a performance report returns data directly. Writes are staged: any operation that would change budget amounts, bids, pause or unpause a campaign, add keywords, or create new ads goes through an approval step before being applied. The README describes this as approval-gated writes.

This staging model means accidental or hallucinated writes do not take effect automatically. An AI assistant cannot change ad spend without the user seeing and approving the proposed change. For agencies running client accounts, this is a meaningful constraint that separates reading data (always safe) from making changes (always reviewed).

Agency setups with multiple accounts are supported. On the Google Ads side, the `listAccessibleCustomers` tool returns every account reachable with the login, and callers pass a `customerId` per tool call. Manager accounts pass an additional `loginCustomerId`. The Meta Ads side works through `listAdAccounts` in the same pattern.

Tool Coverage Across the Three Platforms

The repository reports over 250 tools split across the platforms: 150 or more for Google Ads, 80 or more for Meta Ads, and 20 or more for GA4. The detailed tool lists are documented in the google-ads-mcp/ and meta-ads-mcp/ subdirectories of the repository, with a machine-readable index in llms.txt for AI clients.

On the Google Ads side, the tools cover campaign management, ad group operations, keyword research and management, bid adjustments, budget changes, creative uploads, and performance reporting. On the Meta side, the tools span campaign creation, ad set management, audience building, creative handling, and reporting for Facebook and Instagram ads. The GA4 tools handle data reporting and analytics queries.

The repository also mentions TikTok, LinkedIn, and Microsoft Ads as additional platforms accessible through the same connector, though the README focuses on Google Ads, Meta Ads, GA4, and Search Console as the primary set documented here.

All commands output in a format the AI assistant can read and act on. The llms.txt file at the root of the repository is an agent-readable index that documents the available tools in a structured way, following a convention for AI-accessible documentation.

Comparison with the Official Google Ads MCP and Meta's MCP

Google publishes an official MCP server for Google Ads at github.com/googleads/google-ads-mcp. It is an open-source Python project that runs locally, but it is read-only by design, requires a Google Ads developer token, and needs a Google Cloud project for setup. It is appropriate for engineering teams who want a self-hosted, code-first, read-only integration.

Meta publishes a hosted MCP server at mcp.facebook.com/ads that covers Facebook and Instagram ads only and is hosted by Meta. It supports both reads and writes but handles only the Meta advertising ecosystem.

The Ryze AI server covers both platforms plus GA4 in a single connection. The README comparison table makes the tradeoffs explicit: the official Google server is the better choice when a local, read-only setup is the requirement; the Ryze server is the better choice when writes, Meta Ads, and analytics are all needed in one place. The two can run alongside each other without conflict, since they use different transport configurations.

For the specific question of whether the servers overlap in functionality: they do not conflict, but a team running the official Google server read-only for code-level work and the Ryze server for marketing operations would be running two separate tool sets rather than one. The Ryze server's write capability is the decisive differentiator.

Limitations and What the Server Cannot Do

The server is hosted infrastructure, not a self-deployable binary. Users have no option to run it on their own servers. The server endpoint is `https://connector.get-ryze.ai/mcp`, and all traffic goes through Ryze's systems. Teams with strict data residency requirements or policies against routing advertising data through third-party services should evaluate that against their compliance requirements before connecting.

The free tier covers all 250+ tools. Paid plans from Ryze are for the autopilot feature, which runs ad accounts automatically. The MCP itself is free to use for manual AI-assisted work. The README does not describe pricing for the autopilot plans beyond noting they exist.

The repository has no GitHub releases and the primary language field shows as unknown, reflecting that the server-side implementation is not in this repository. The repository is primarily documentation, configuration files, and the tool descriptions that the server exposes. The last push was on 2026-09-25.

License and Integration Options

The repository is licensed under MIT. The license covers the documentation and configuration files in the repository; the server itself is Ryze's hosted product.

The server works with any MCP-compatible client. The README explicitly lists Claude Code (`claude mcp add`), claude.ai, Claude Desktop, ChatGPT, Cursor, Windsurf, and n8n. The n8n integration is through n8n's MCP node, which can call any Streamable HTTP MCP endpoint. This makes the server usable in automated workflows that chain AI reasoning with ad platform operations.

The CHANGELOG.md in the repository documents server updates. The configs/ and docs/ directories hold additional integration documentation per client type, including screenshots for the setup flow in each supported application.

Editorial conclusion

This MCP server is the right choice for marketers and agencies who manage Google Ads, Meta Ads, and GA4 together and want to control them through Claude, ChatGPT, or Cursor without configuring three separate tools. It is not the right choice for engineering teams that need a local, self-hosted, code-first setup with full control over the MCP binary, since the server runs on Ryze's infrastructure and requires a live connection to the connector endpoint. Before adopting it for production ad management, verify that the approval gating on write operations matches your internal change-control requirements, since budget and bid changes go through a staging step rather than taking effect immediately.

Frequently asked questions

Official sources

1. irinabuht12-oss/google-meta-ads-ga4-mcp on GitHub
2. Issues
3. License: MIT
4. Project website
5. README