Google Ads MCP Server: Official vs Hosted (2026)
Adspirer Team
GOOGLE ADS MCP
A Google Ads MCP server connects an AI agent — ChatGPT, Claude, Cursor — to your Google Ads account so you can manage campaigns in plain English. There's a free official server from Google and hosted servers like Adspirer. The difference that matters: read-only vs read-and-write, and how much plumbing you maintain.
-
Run Search, PMax, and reporting from ChatGPT or Claude
-
Choose self-managed official access or a hosted OAuth setup
-
Read-and-stage by default — deletes are confirmation-gated
A Google Ads MCP server is the bridge that lets an AI agent read and manage your Google Ads account through conversation instead of the dashboard. MCP — the Model Context Protocol — is the open standard Anthropic created so AI models can call real tools. A Google Ads MCP server exposes campaign management, reporting, and optimization as those tools.
Ask “which campaigns are burning budget with no conversions?” and the agent pulls your live data, answers, and stages the fix. There are a few ways to get there — including a free one from Google itself — so let’s be precise about the trade-offs.
What you can do with a Google Ads MCP server
Capabilities through conversation
Reading and managed actions, depending on the server.
-
Pull performance — Spend, conversions, CPA, ROAS, impression share across campaigns and ad groups, any date range.
-
Audit wasted spend — Mine search terms, surface zero-conversion spend, stage negative keywords.
-
Launch campaigns — Create a Search or PMax campaign from a brief — created paused for review.
-
Adjust bids & budgets — Shift budget to winners, tune device/geo/audience bids, with rationale in chat.
-
Read Auction Insights — See which competitors gained impression share and why your CPCs moved.
-
Audit conversion setup — Verify tags fire once, pass values, and use the right attribution model.
The capability that separates servers is write access. Google’s official server is intentionally read-only; other providers may expose campaign changes. That does not make one universally better. It makes the intended workflow the deciding factor.
Under the hood, a broad server needs a way to keep its tool surface usable. Adspirer uses search_tools to find the right operation for a request, then get_tool_schema to retrieve its parameters before acting. That discovery layer lets one connection cover Google, Meta, Amazon, LinkedIn, TikTok, and ChatGPT Ads without loading every tool into one prompt. Current surface: MCP docs.
Official Google Ads MCP Server (2026) vs a hosted server
Google publishes an official, open-source Google Ads MCP server. It is Python-based, Google-only, and currently read-only. It runs over stdio by default and can be operated locally or deployed to Cloud Run. The official guide requires a Google Cloud project, developer token, and OAuth 2.0 or service-account credentials.
Its current tools are list_accessible_customers, search, and get_resource_metadata. Together they support account discovery, GAQL reporting, and schema lookup; they do not modify bids, pause campaigns, or create assets.
DECIDE
Google Ads MCP server options
| Adspirer | Google's official MCP | Community repo | DIY (Ads API) | |
|---|---|---|---|---|
| Setup | Hosted OAuth connection | Run locally or on Cloud Run + credentials | Usually self-hosted | Build it yourself |
| Read performance | Yes | Yes | Varies | Yes (you build it) |
| Write / manage ads | Yes — staged | No (read-only) | Varies | Yes (you build it) |
| Multi-platform | Yes — +Meta, Amazon, ChatGPT Ads, LinkedIn, TikTok | No (Google only) | No | No |
| Maintained for you | Yes | You operate its runtime | On the maintainer | On you |
| Safety rails | Yes — deletes hard-gated | N/A (read-only) | Rarely | You build them |
| Pricing model | Free tier + paid usage | Open source + your runtime | Varies | Your engineering and runtime costs |
Comparison note: This comparison reflects publicly available information gathered when the article was prepared. Features, pricing, hosting, and policies can change. Check each provider’s current documentation and decide based on your own requirements. If a detail is outdated or incorrect, let us know; we’ll review the source and correct it.
If you only need to read Google Ads data, are comfortable operating the credentials and runtime, and do not need other platforms in the same workflow, Google’s official MCP is a solid free choice. If you want the agent to act or prefer a managed multi-platform connection, compare hosted read-and-write options. Full breakdown: when Google’s free MCP server fits.
If the appeal of a self-managed repo is inspectable logic, Adspirer’s Community Plugins offer a different model: open-source, read-only analysis extensions on top of a maintained hosted connection. That does not replace the control of running Google’s server yourself, but it can reduce the amount of infrastructure an agency maintains.
Connect Google Ads in about two minutes
Adspirer is a hosted Google Ads MCP server with read-and-write access and built-in approval controls. It also covers Meta, Amazon, ChatGPT Ads, LinkedIn, and TikTok through the same connection.
Sign up and grab the MCP URL
Create a free account at adspirer.ai. The server URL is https://mcp.adspirer.com/mcp.
Add the server to your AI client
ChatGPT: open Apps, search for Adspirer, click Connect, and sign in, then use @adspirer in a chat. Claude: add Adspirer under Customize → Connectors. Plan availability and organization controls can change, so use the current client documentation. Walkthroughs: connect Claude to Google Ads · connect ChatGPT to Google Ads.
OAuth into Google Ads
Sign in with the same Google account you use for Ads Manager. No developer token to request, no API client to configure — the server handles it.
Run your first prompt
Adspirer won’t delete anything without an explicit confirmation — destructive tools are hard-gated behind a confirm_delete flag on every platform. New campaigns are created paused, pausing a live campaign needs explicit confirmation, and the agent stages changes for your approval. Read-and-write does not mean unsupervised.
Common questions
Frequently asked questions
Setup
Capabilities
Pricing
Related reading
- What is MCP? Model Context Protocol explained
- Why not just use Google’s free MCP server?
- Connect Claude to Google Ads
- Connect ChatGPT to Google Ads
- The 10 best ad MCP servers in 2026
- AdWords automation without scripts
Run Google Ads from your AI agent.
Connect Google Ads to ChatGPT or Claude through Adspirer's MCP server in about two minutes — read and write, with safety rails. Free tier, no credit card.
Try Adspirer freeMore articles to read
ChatGPT for Marketing: The Complete Guide [2026]
How to use ChatGPT for marketing in 2026 — real workflows, 15 copy-paste prompts, and how to connect it to your Google Ads and Meta Ads for live campaign management.
Claude for Marketing: A Practical Guide for 2026
Claude for marketing: connect live ad data, organize brand context, analyze campaigns, create reports, and choose between Chat, Cowork, and Claude Code.
Google Ads Automation: Tools & Workflows (2026)
Compare Google Ads automation tools, Google AdWords automation with scripts and rules, and AI agents. Learn what to automate and what to review.
Stop Switching Between Ad Platforms: Manage Google, Meta, LinkedIn, and TikTok in One Conversation
You don't need four browser tabs. Connect Claude or ChatGPT to Google Ads, Meta Ads Manager, LinkedIn Campaign Manager, and TikTok Ads through one Adspirer connection and manage all of them from a single conversation.