Best MCP Server for Google Ads 2026: 9 Options Compared

Checked October 30-31, 2026. Prices and features come from each vendor's own site and docs, linked under every tool and again at the end.

The short version: if you only want to read Google Ads data in Claude and you're comfortable running a server from the terminal, the free official Google Ads MCP does the job. The moment you want the assistant to actually change campaigns on client accounts, you need a server with a real approval flow. For PPC agencies and freelancers who run Google Ads and Meta Ads for many clients, AdSparky is our pick: it works in Claude, ChatGPT, Claude Code and Codex, every change goes through a dry run, a before/after preview and your approval, and it never deletes anything. It's in free beta, no card: adsparky.com.

AdSparky is our product - I'm the founder, so I link a source for every claim about other tools and say where AdSparky falls short.

Google Ads MCP servers at a glance#

Want to see the approval flow on your own accounts? Join the free AdSparky beta - setup takes a couple of minutes in the browser.

What an MCP server for Google Ads actually is#

MCP (Model Context Protocol) is the open standard that lets an AI assistant like Claude or ChatGPT talk to outside software. A Google Ads MCP server is the translator in the middle: the assistant asks for data or proposes a change, and the server talks to the Google Ads API (Google's official channel for reading and changing accounts).

Instead of pasting exports into chat, you type "show campaigns that spent over $500 in 30 days with zero conversions" and get the answer from live data.

How I compared them#

I didn't run multi-week tests of every tool on the same account, and I won't pretend I did. On October 30-31, 2026 I went through each vendor's site, pricing and docs and scored five things: read or write; safety (dry run - the change is prepared and validated but not sent - plus preview, approval, undo, and whether it can delete); Meta Ads included; browser or terminal; and price in the vendor's currency.

The tools, one by one#

1. AdSparky - our pick for agencies on Google + Meta#

What it does: a remote MCP server plus a web panel that connects Google Ads and Meta Ads (plus GA4, Google Tag Manager, Merchant Center and the client's site through a WordPress plugin) to Claude, ChatGPT, Claude Code and Codex. You sign in with OAuth, nothing to install. Over 300 tools; about 30 core ones show in chat, the rest via built-in tool search.

Why I think it's the most comfortable way to run client accounts from chat:

• Every change goes dry run -> before/after preview -> human approval, in the chat or in the panel's approval queue.
• It never deletes - it only pauses. New campaigns are created paused.
• Full change history with author, source and reason, and undo is usually one step (exceptions are flagged in the preview).
• Per-client limits (max budget, untouchable campaigns, allowed platforms) that the AI can only tighten from chat. Loosening them takes a human in the panel. One operation can move a budget by at most +30% / -50%.
• A morning review across all clients with the urgent stuff on top, client memory (what counts as a lead, what's off-limits), goals like max CPA, and PDF reports for clients.
• Numbers come from the API, not the model's memory. Small sample? It says "not enough data".
• Database and engine are hosted in the EU (Ireland and Frankfurt), keys are encrypted, agency data is separated, and we don't train AI on your data. That matters if your clients ask GDPR questions.

Price: free beta, no card. Pricing will be announced before the beta ends. You need your own Claude or ChatGPT subscription.

To be fair: it's a beta, and it covers Google Ads and Meta Ads only (no Microsoft, TikTok or LinkedIn yet).

More on the safety model: How to connect Claude to Google Ads. Source: adsparky.com.

2. Official Google Ads MCP (Google)#

What it does: an MCP server from the Google Ads API team with three tools: run a GAQL query (Google Ads Query Language, SQL-like), describe a resource's fields, and list accessible accounts. Google's docs call it "strictly read-only": it won't change bids, pause campaigns or create anything.

Price: free, code on GitHub.

Pros: straight from Google, free, supports manager accounts (MCC).

Cons: read only. You need a Google Cloud project with the Google Ads API enabled and an access level, OAuth credentials, Python and pipx. It runs locally or on a server you deploy yourself. Google Ads only. Per Google's docs, developer tokens were retired on September 9, 2026 and access now ties to the Cloud project, while the GitHub README still mentions a token - follow Google's docs if they disagree.

Sources: GitHub, Google docs.

3. Optmyzr MCP (Sidekick)#

What it does: the long-running PPC optimization platform, now with an MCP server. In Claude or ChatGPT you run analyses, reports and workflows like in the Optmyzr UI. Covers Google, Microsoft, Amazon, Meta, LinkedIn and Yahoo.

Safety: per Optmyzr's help center, changes made through MCP land on the Change Requests page, get checked against account rules, and a team member applies them.

Price: from $209 a month (the pricing page headline says "starting at $209 a month"); the exact amount depends on monthly ad spend. MCP access is ticked in all three plans (Essentials, Premium, Enterprise).

Best for: existing Optmyzr agencies. See also: Optmyzr alternatives. Sources: pricing, MCP guide.

4. Opteo MCP#

What it does: Google Ads recommendations and alerts, with a remote MCP server. The assistant reads account data in Opteo and runs actions in Opteo.

Safety: "Your assistant asks you to confirm before it runs anything that changes your account."

Price: Basic $129/mo (10 accounts), Professional $249/mo (25), Agency $499/mo (75), Enterprise custom.

Cons (from their site): Google Ads only. The help page doesn't list which actions MCP exposes.

Sources: pricing, Opteo MCP.

5. Ryze AI#

What it does: an AI platform for Google, Meta, TikTok, LinkedIn and Microsoft Ads with an MCP connector for Claude and ChatGPT.

Safety: the pricing page lists "approval controls on every change" on every plan.

Price: Paid Ads $89/mo, Agency $279/mo (unlimited client workspaces, white-label). 7-day trial. Claims "2,000+" customers on its site.

Head-to-head: Ryze AI alternative. Source: get-ryze.ai/pricing.

6. Markifact#

What it does: hosted MCP and AI agents for 50+ marketing platforms, including Google, Meta, TikTok and Microsoft Ads. Works in Claude, ChatGPT, Gemini and Cursor.

Safety: "Nothing goes live without you" - you approve in chat or Slack.

Price: Pro $79/mo (1 seat), Team $249/mo (white-label, shared workspace, API), Managed custom. Usage is billed in credits per operation, so cost scales with how much you run.

Head-to-head: Markifact alternative. Source: markifact.com/pricing.

7. Adspirer#

What it does: MCP for ChatGPT, Claude, Claude Code and others. Google, Meta, LinkedIn, TikTok, Amazon and ChatGPT Ads. Listed in the ChatGPT and Claude directories.

Safety: "New campaigns stay paused" and "Every high-impact change stays yours to approve". The site doesn't define "high-impact", so I couldn't tell which changes skip approval.

Price: Free (15 tasks/mo), Plus $40, Pro $83, Max $167/mo, agency on request.

Source: adspirer.com/pricing.

8. AdKit#

What it does: MCP and CLI for seven platforms: Meta, Google, TikTok, Reddit, LinkedIn, X and Microsoft. Claude connector and ChatGPT app.

Safety: "Every change your agent makes is a draft by default" until you approve it in the AdKit dashboard.

Price: $29/mo (1 project), $49/mo (unlimited projects). Claims "6,000+ users".

Source: adkit.so/pricing.

9. Pipeboard#

What it does: MCP servers for Meta, Google, TikTok and more. Works with Claude, ChatGPT, Claude Code, Cursor, Codex and n8n.

Safety: read-only tokens and scoping to selected accounts and tools (plan-dependent). For Google Ads I didn't find a separate approval step described; the site advises reviewing what you ask the assistant and using scoped access to set boundaries.

Price: free plan (2 ad accounts, 30 AI tool calls a week), paid tiers with 3, 10 and 25 account slots. The pricing page localizes currency - from Poland it showed PLN, so check the page for your region.

Head-to-head: Pipeboard alternative. Source: pipeboard.co/pricing.

When to choose AdSparky#

• You run Google Ads and Meta Ads for clients. Compare cost per lead across both platforms in one sentence instead of two tools and a spreadsheet. The official Google server and Opteo don't cover Meta at all.
• You want write access without the anxiety. Every change is a dry run first, you see before/after, and nothing hits the account until a human approves. The official Google server can't write, and several tools above describe approval only in a tagline.
• You never want "deleted" to be possible. AdSparky pauses, never deletes. A paused campaign comes back with its history; a deleted one doesn't.
• You manage 10, 20, 50 accounts. The morning review across every client, urgent items on top, is the single biggest time saver I've had in my own agency work.
• Juniors propose, seniors approve. Team roles plus an approval queue in the panel, and automations that only prepare changes unless the owner turns on autopilot.
• Clients have hard rules. Per-client budget caps and untouchable campaigns that the AI can't loosen from chat. A +30% / -50% budget guard on every single operation.
• You don't want a terminal. It runs in claude.ai or ChatGPT in the browser. Paste a URL, sign in, done.
• You care about EU data residency. Database in Ireland, engine in Frankfurt, encrypted keys, no training on your data.
• You want a flat start. Free beta, no card, no credits to count.

When to choose something else#

• Read-only questions on one account and a developer on hand: the official Google Ads MCP.
• You already live in Optmyzr or Opteo: start with their MCP.
• You need TikTok, LinkedIn, Microsoft or Amazon Ads in the same chat today: Ryze, Markifact, Adspirer, AdKit or Pipeboard.

How to choose#

Read or write? For writes, pick a tool that documents approval and undo in its docs, not just on the homepage. Meta Ads too? Then the official Google server and Opteo are out - see the best Meta Ads MCP list. Already paying for a PPC tool? Try its MCP first. And find out what happens when the assistant misreads an instruction on a client account.

FAQ#

Is there a free MCP server for Google Ads?#

Yes. The official Google Ads MCP is free and open source, but read only and self-hosted. Pipeboard (30 calls a week) and Adspirer (15 tasks a month) have free plans. AdSparky is free during the beta.

Can the official Google Ads MCP change campaigns?#

No. Google's documentation says it is "strictly read-only": no bid changes, no pausing, no creating resources.

Is AdSparky better than Optmyzr or Opteo for an agency?#

Depends on your stack. If you run Google and Meta for many clients and want to work in Claude or ChatGPT with approvals, per-client limits and EU hosting, AdSparky is built exactly for that. Opteo's pricing and features cover Google Ads only. If your team already runs on Optmyzr, its MCP plugs into what you have.

Do I need a Google Ads developer token?#

Per Google's docs, developer tokens were retired on September 9, 2026 and API access now ties to a Google Cloud project. With remote tools like AdSparky, Opteo or Optmyzr you just sign in with your Google account.

Sources#

All checked October 30-31, 2026.

• Official Google Ads MCP: github.com/googleads/google-ads-mcp, developers.google.com - MCP server, API policy
• Optmyzr: optmyzr.com/pricing, Sidekick MCP User Guide
• Opteo: opteo.com/pricing, Opteo MCP
• Ryze AI: get-ryze.ai, get-ryze.ai/pricing
• Markifact: markifact.com, markifact.com/pricing
• Adspirer: adspirer.com, adspirer.com/pricing
• AdKit: adkit.so, adkit.so/pricing
• Pipeboard: pipeboard.co, pipeboard.co/pricing
• AdSparky: adsparky.com

If you run Google and Meta for clients, try the approval flow on your own accounts. AdSparky is in free beta, no card: adsparky.com.