# Google Ads MCP Guide 2026 | AI-Powered Campaign Analysis # How to Get Started with Google Ads MCP (2026 Setup Guide) > **Quick answer:** Google Ads MCP connects your Google Ads account to an AI assistant (like Claude) so you can analyze campaigns in plain English. In 2026 there are two ways to set it up. **Path A — Google’s official server**: open-source, self-hosted, and **read-only by design** (built for diagnostics and analytics), best for developers comfortable with a Google Cloud project and OAuth. **Path B — a hosted/managed option** (e.g., GrowthSpree): no developer keys or Python, connect via OAuth and a drag-and-drop extension in **under 5 minutes**. Both read live data; neither of these changes your account without you. > **TL;DR:** Getting Google Ads into an AI assistant used to mean custom API code. Not anymore. Google shipped an official, open-source Google Ads MCP server in 2026 — read-only, GAQL-based, and self-hosted, ideal for developers and analysts. For marketers who don’t want to touch a terminal, hosted options connect the same data in under five minutes: register, authorize Google Ads with OAuth (read-only), install a Claude Desktop extension, paste a token, and start asking questions. This guide covers both paths step by step, what read-only vs read-write means (and how draft-first third-party tools add safe write access), the prompts to run first, and how to keep it secure and multi-account. ## The two setup paths at a glance | | Google’s official server (Path A) | Hosted / managed (Path B) | |---|---|---| | Setup | Self-hosted, GCP + OAuth | No-code, OAuth + extension | | Time | Longer (developer setup) | Under 5 minutes | | Access | Read-only | Read-only | | Best for | Developers, analysts | Marketers, agencies | | Cost | Free, open-source | Often free (e.g. GrowthSpree) | | Extras | Raw GAQL interface | Prompt library, HubSpot, multi-account | *Reflects the Google Ads MCP landscape as of mid-2026; confirm current tool availability and beta status on each provider’s docs.* The way to answer “where am I wasting money?” in Google Ads has changed. Instead of exporting a report and building a pivot table, you ask your AI assistant and it reads the live data. Here’s exactly how to set that up — whichever path fits your team. ## What is Google Ads MCP? Google Ads MCP is a connection that lets an AI assistant read your Google Ads data through the [Model Context Protocol](https://modelcontextprotocol.io/docs/getting-started/intro) — the open standard [from Anthropic](https://www.anthropic.com/news/model-context-protocol). An MCP server sits between the assistant and the Google Ads API: you ask a question in natural language, the server runs the query, and the assistant returns a human-readable answer built from your live account data. For the wider ecosystem, see our [complete guide to MCP servers for B2B SaaS marketing](https://www.growthspreeofficial.com/blogs/mcp-servers-b2b-saas-marketing-complete-guide). ## What changed in 2026: Google’s official server The big update is that Google shipped its own [official, open-source Google Ads MCP server](https://github.com/googleads/google-ads-mcp). Per [Google’s developer documentation](https://developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server), the initial version is **read-only, designed for diagnostics and analytics**. It exposes a small set of tools — list accessible accounts, run a [GAQL](https://developers.google.com/google-ads/api/docs/query/overview) search, and fetch resource metadata — and it cannot modify bids, pause campaigns, or create assets. That read-only-by-design choice is deliberate: it makes the official server safe for analysis out of the box. > **Key takeaway:** Google’s official server is free and authoritative, but it’s a developer tool — self-hosted, GAQL-first, and read-only. If Google Ads analysis is your job and you’re comfortable with a Google Cloud project, it’s an excellent foundation. ## Path A: Google’s official server (developer setup) Use this if you have (or can get) engineering support. At a high level: 1. Create or choose a Google Cloud project and install the gcloud CLI (authenticated, active project set). 1. Get your Google Ads developer token (a 22-character string) from the API Center. 1. Create OAuth credentials — an OAuth2 Client ID and Client Secret pair. 1. Add the server to your MCP client’s config (e.g., settings.json for Claude Code, mcp.json for Cursor) with your developer token and OAuth credentials. 1. Optionally deploy to Google Cloud Run to share the server across agents, and use tools_config.yaml to enable or disable specific tools. 1. Restart the client and start querying with GAQL-backed natural-language questions. *Google’s own docs and GitHub repo are the source of truth for the exact commands and environment variables — follow them for current syntax.* ## Path B: Hosted / managed setup (no code, under 5 minutes) Use this if you’re a marketer who wants the data without the developer work. A hosted option (like GrowthSpree’s, built on top of Google’s MCP) removes the tokens, Python, and GitHub steps: 1. Register for the hosted Google Ads MCP with your email — no developer keys or manual tokens. 1. Authorize your Google Ads account via OAuth and grant read-only permission; optionally connect HubSpot (read-only) for revenue data. 1. Copy your secure MCP token from the dashboard. 1. Download the provided .MCPB extension file, then install Claude Desktop and open it. 1. Drag and drop the .MCPB file into Claude Desktop’s Extensions, paste your token, and toggle the extension ON. 1. Open a new chat and start asking questions about your account. For a fuller walkthrough and optimization patterns, see the [Google Ads MCP definitive guide for SaaS](https://www.growthspreeofficial.com/blogs/google-ads-mcp-definitive-guide-saas), or the broader [connect Google Ads to AI and start chatting](https://www.growthspreeofficial.com/blogs/how-to-connect-google-ads-with-ai-and-start-chatting-with-your-data). ## Read-only vs read-write Google’s official server — and most hosted analytics options, including GrowthSpree’s — are read-only: the assistant can analyze your data but can’t change campaigns, budgets, or bids. That’s the safe default for reporting and diagnostics. If you specifically want the AI to make changes, several third-party servers add read-write with a draft-first safety layer, where the agent stages changes and nothing goes live until you approve. Choose read-only for analysis; only add read-write when you want execution, and keep a human approving anything that touches live budget. ## What you can ask once connected This is where MCP pays off — each prompt replaces minutes of dashboard digging: - *“Which campaigns wasted the most money last week?”* - *“Why did CPA spike over the last 14 days?”* - *“Show me my underperforming keywords and search terms.”* - *“Give me a 7-day executive summary with recommended actions.”* Hosted options often ship a prompt library for deeper analyses — root-cause diagnostics, trend breakdowns, and client-ready summaries. For the diagnostic workflow, pair this with our [Google Ads root cause analysis guide](https://www.growthspreeofficial.com/blogs/google-ads-root-cause-analysis-mcp-claude), and to act on findings, the [negative keyword template](https://www.growthspreeofficial.com/blogs/b2b-saas-google-ads-negative-keyword-list-template-save-10k). ## Security and multi-account - **Read-only OAuth.** Access is granted via OAuth with read-only scope — the assistant reads data but can’t alter your account. - **Your data isn’t used to train models.** With a reputable hosted option, processing happens in your session and data isn’t used for AI training; confirm each provider’s policy. - **Revoke anytime.** You can remove access in under a minute by toggling off the extension and revoking the connection. - **Multi-account.** MCP supports multiple accounts — connect via your Google MCC to run cross-account analysis. MCC gives access; MCP adds the AI intelligence on top. ## Common mistakes to avoid - **Choosing the wrong path.** Don’t self-host Google’s server if you just want quick analysis — use a hosted option; don’t expect a no-code tool to give raw GAQL control. - **Expecting write access from read-only servers.** Google’s official server and most analytics servers can’t change campaigns by design. - **Granting write access without guardrails.** If you use a read-write server, require draft-first approval on every change. - **Skipping the developer token step (Path A).** The official server needs a valid Google Ads developer token and OAuth credentials to start. - **Trusting one number blindly.** Cross-check platform-reported metrics against your CRM for pipeline truth. ## Frequently Asked Questions ### Q1. What is Google Ads MCP? A connection that lets an AI assistant like Claude read your Google Ads data through the Model Context Protocol, so you can analyze performance, find wasted spend, and get recommendations conversationally instead of building reports. ### Q2. Does Google have an official Google Ads MCP server? Yes. Google shipped an official, open-source Google Ads MCP server in 2026. It’s read-only by design — built for diagnostics and analytics — and exposes tools like list accessible accounts and a GAQL search. ### Q3. Is Google Ads MCP read-only or read-write? Google’s official server is read-only, as are most hosted analytics options (including GrowthSpree’s). Some third-party servers add read-write with a draft-first approval layer for making changes. ### Q4. How do I set it up without coding? Use a hosted option: register with your email, authorize Google Ads via OAuth (read-only), copy your token, install a Claude Desktop extension, paste the token, and toggle it on — under 5 minutes, no developer keys. ### Q5. What do I need for Google’s official server? A Google Cloud project, the gcloud CLI, a Google Ads developer token (22 characters), and OAuth Client ID/Secret credentials. You add the server to your MCP client’s config and can optionally deploy it to Cloud Run. ### Q6. Can it change my campaigns? Not with Google’s official server or a read-only hosted option — those only analyze data. Read-write third-party servers can make changes, ideally draft-first so nothing goes live without your approval. ### Q7. What can I ask it? Questions like “which campaigns wasted the most money last week,” “why did CPA spike,” “show me underperforming keywords,” or “give me a 7-day executive summary with recommended actions.” ### Q8. Does it work with multiple accounts? Yes. MCP supports multi-account analysis — connect through your Google MCC to run cross-account queries. MCC provides access; MCP adds the AI intelligence layer. ### Q9. Is it secure? Access uses OAuth with read-only scope, and reputable hosted options don’t use your data to train models and let you revoke access in under a minute. Confirm each provider’s specific security and data policy. ### Q10. Does it work with assistants other than Claude? Yes. MCP is assistant-agnostic — Google’s server and most others work with Claude, Gemini, Cursor, and other MCP-compatible clients; hosted marketer options commonly target Claude Desktop. ### Q11. How is MCP different from an MCC? An MCC (My Client Center) gives you access to multiple Google Ads accounts. MCP adds AI-driven analysis, contextual understanding, and natural-language querying on top of that data. ### Q12. How long does setup take? A hosted option takes under 5 minutes. Google’s official self-hosted server takes longer because of the Google Cloud project, developer token, and OAuth configuration. ## Get started If you want analysis fast, take the hosted path: [connect the free Google Ads MCP](https://www.growthspreeofficial.com/resources/google-ads-mcp) and start asking “where am I wasting money?” in minutes. If you prefer the raw, official route, follow [Google’s developer guide](https://developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server). Either way, the [best AI marketing MCP servers guide](https://www.growthspreeofficial.com/blogs/best-ai-marketing-mcp-servers-b2b-saas) helps you plan the rest of your stack. --- **About the author:** Ishan Manchanda is Co-Founder at GrowthSpree, a B2B SaaS marketing agency (Google Partner, HubSpot Solutions Partner, 4.9/5 on G2). GrowthSpree runs Google Ads MCP on every client account across 300+ B2B SaaS brands and $60M+ in managed spend, using read-only AI analysis with senior operators making the optimization calls.