Google Ads MCP for Claude Code: Run PMax and Search
Connect a Google Ads MCP to Claude Code. Manage PMax, Search, Shopping campaigns, audiences, and conversions from your terminal in 2026.
Agentic Orchestration Kit for Claude Code.
Problem: Running Google Ads at scale means living inside the Ads Editor, copy-pasting GAQL into the API explorer, and waiting weeks for a developer token to graduate from Test to Standard access. When a Performance Max campaign misfires, you are five tabs deep before you have pulled the data.
Quick Win: A google ads mcp server connects Claude Code to your Google Ads account so the agent can run GAQL, audit campaigns, and (with the right server) edit budgets, swap bidding strategies, and refresh asset groups. As of May 2026 you have four ready-made options plus a DIY path: Google's official read-only MCP shipped April 28, Pipeboard's hosted server with 35 plus tools, Composio's HTTP toolkit, community servers like cohnen/mcp-google-ads, and a custom wrapper for teams with compliance requirements.
This pillar maps the google ads mcp claude code landscape: which server to pick, how to get a developer token past the approval backlog, the eight performance max mcp and Search workflows that pay back the install on day one, and how Google Ads slots into the Meta plus Shopify operator stack.
The Google Ads MCP Landscape in 2026
Unlike the Meta side, where Meta shipped both an official CLI and a hosted MCP in late April 2026, Google's first-party Google Ads MCP is deliberately narrow. The April 28, 2026 release at github.com/googleads/google-ads-mcp exposes three tools (list_accessible_customers, search, get_resource_metadata) and runs read-only. Apache 2.0, Python, 428 stars. There is no first-party path to "let Claude pause my underperforming PMax asset group."
That gap is why the third-party Google Ads MCP market is busier than the third-party Meta market. The official Meta connector exposes the full Marketing API, so brokers compete on convenience. The official google ads mcp exposes a read-only slice, so brokers compete on capability.
The four google ads mcp servers for google ads claude code in May 2026:
- Official Google Ads MCP (
googleads/google-ads-mcp). Read-only, three tools, stdio transport, your developer token and OAuth credentials. Right for analytics-heavy work where you want first-party trust. - Pipeboard Google Ads MCP (
google-ads.mcp.pipeboard.co). Hosted HTTP transport, 35 plus tools covering keyword management, bid optimization, GAQL, and campaign creation. OAuth 2.0, no local credentials. Right for full-control PMax and Search operations. - Composio Google Ads (
composio.dev/toolkits/googleads). HTTP MCP, OAuth managed by Composio, integrated into the Tool Router so Google Ads sits next to Slack, HubSpot, Notion, and the rest of the Composio catalog. Right if Google Ads is one of fifteen toolkits in the same session. - Community servers like
cohnen/mcp-google-ads(577 stars, MIT, GAQL-focused) andamekala/ads-mcp(multi-platform, 100 plus tools across Google, Meta, LinkedIn, TikTok). Read-mostly, self-hostable.
The decision tree is simple. Read-heavy analytics goes to the official server. Write-heavy google ads mcp claude code operations go to Pipeboard. Multi-platform toolkits go to Composio. Compliance needs go to a community server you can fork. If you have not wired an MCP server in Claude Code before, our MCP basics guide is the protocol primer.
The Official Google Ads MCP: Read-Only and That Is Fine
Google's google-ads-mcp is the cleanest place to start, and the read-only constraint is more feature than bug for a first install. You cannot drop a budget by 90 percent on a misread instruction. You cannot mutate an asset group. What you can do is everything analytical: GAQL queries, customer-id discovery, resource-metadata introspection.
Install in Claude Code:
The server takes three env variables: GOOGLE_APPLICATION_CREDENTIALS pointing at your OAuth JSON, GOOGLE_PROJECT_ID for the Cloud project, and GOOGLE_ADS_DEVELOPER_TOKEN for the 22-character token. Set them in ~/.claude/settings.json under the env block, restart Claude Code, and you are connected.
The three tools are surgical. list_accessible_customers returns every Google Ads customer ID the authenticated user can reach. search executes any GAQL query you can write, covering performance metrics, budgets, ad-group structure, keyword data, and conversion fields. get_resource_metadata returns metadata about resource types, useful when Claude needs to reason about which fields exist before constructing a query.
Real workflow: ask Claude to "list every Performance Max campaign across all accessible accounts that spent more than $500 in the last seven days but had a conversion-value-to-cost ratio below 2.0, sorted by spend." Claude calls list_accessible_customers, iterates GAQL queries, aggregates, and returns a Markdown table grounded in live API responses.
The hard limit is that you cannot act on what you found. The natural follow-up (pause the worst, reallocate budget, swap bidding strategies) needs a write-capable server. Most operators end up running the official server for read trust and Pipeboard or Composio for execution. Running both costs context, which is why MCP Tool Search matters once you cross three connected servers.
Pipeboard Google Ads MCP: The Write Surface
Pipeboard ships the broadest write surface in the third-party google ads mcp market. The hosted HTTP server at https://google-ads.mcp.pipeboard.co/ exposes 35 plus tools covering keyword management, bid optimization, GAQL, and campaign creation. Skip the developer-app plumbing, hand off OAuth to a hosted broker, trade self-hosting for setup speed.
That is the entire setup. The first tool call triggers a browser OAuth handshake against Pipeboard's authorization endpoint, which proxies your Google identity through a token-exchange flow.
The capability list that matters for a real pmax claude code workflow:
- Campaign creation and modification. Spin up a Search or Performance Max campaign, set the bidding strategy, attach the conversion goal, and configure the daily budget from a single prompt. The asset-group-and-asset bulk-mutate that Google's API requires (asset group plus minimum-required assets in one request) is handled by Pipeboard's tool layer, not by you.
- Keyword management. Bulk-add keywords, push negative keywords across Search campaigns, run search-term reports, identify wasted-spend queries.
- Bid optimization. Switch a campaign from Maximize Conversions to Target ROAS. Adjust tCPA targets in batch.
- Budget pacing. Pull month-to-date spend, compare to plan, reallocate across campaigns to hit a monthly cap.
- GAQL queries. Same surface as the official server, useful when you want one MCP for both read and write.
The tradeoff is the broker. Your operations move through Pipeboard's infrastructure, a third party in the auth chain. For agencies and in-house teams running their own spend, this is a fine trade. For regulated client data, the calculus shifts. Match Pipeboard's security documentation against your contract obligations.
Composio Google Ads: When You Run Twenty Toolkits at Once
Composio's Google Ads toolkit is a different shape. It is a single MCP entry inside Composio's Tool Router, alongside Slack, HubSpot, Linear, Stripe, and roughly two hundred other apps. The Google Ads surface is narrower than Pipeboard's, with five primary actions at writing time: create customer list, get campaign by ID, get campaign by name, get customer lists, and add or remove members from a customer list. Enough for audience workflows, thinner than Pipeboard for full campaign editing.
The YOUR_MCP_URL_HERE placeholder is generated from your Composio dashboard when you connect the Google Ads integration. Composio handles OAuth, token refresh, and scopes on their side. The only secret you store locally is your Composio API key.
The strategic case shows up when Google Ads is one tile in a bigger workflow. Example: "Pull yesterday's Google Ads spend by campaign, attribute revenue from Stripe, write a summary to the marketing-daily Slack channel, update the daily-spend row in the Notion ops dashboard." Pipeboard handles step one. Composio runs all four through one Tool Router call with one auth boundary and one billing relationship.
If your workflow is self-contained, Composio is overkill. If it is the spine of a multi-tool loop touching CRM, billing, comms, and analytics, it earns the install.
Building Your Own: Google Ads API Plus a Wrapper
This is not a fourth vendor -- it is the path for teams that need to own the entire data perimeter. The MCP spec is small. A Google Ads API client is a few hundred lines. A custom MCP server that wraps it is another few hundred. For a senior engineer this is two days of work, with no third-party trust boundary.
Why anyone does it: compliance teams need provable data perimeters; custom surfaces let you bake business rules into tool definitions (every pause_campaign call can require a manager-approved ticket ID); agencies running 50 client accounts need cleaner per-client token isolation than the brokers expose.
What you build: a Python server using the Google Ads Python client library wrapped in an MCP transport. Claude Code's custom integrations guide walks the wrapper pattern. Register tools (run_gaql_query, pause_campaign, update_budget, create_asset_group), each a thin layer over the Python client, then claude mcp add it like any other server.
The cost is maintenance. Google updates the Ads API on a roughly twelve-month cycle of breaking changes. The official MCP and third-party servers track those updates. Your custom server tracks them when you remember. For most operators this is the wrong tradeoff. For a few, it is the only one that meets the compliance bar.
OAuth, Developer Tokens, and the Approval Process
Every google ads mcp setup post skips this section. We are going to fix that, because the developer token approval is currently the longest pole in any google ads mcp claude code install.
A Google Ads developer token is a 22-character string from the API Center page of your Google Ads manager account. Three access levels:
- Test Account Access: 15,000 operations per day, test accounts only, instant.
- Basic Access: 15,000 operations per day, production accounts, two-business-day review.
- Standard Access: unlimited operations, production accounts, ten-business-day review.
Reality in 2026: the Google Ads Developer Blog posted February 6 an update acknowledging that both access tiers were running beyond their stated review windows -- Basic Access averaging 14-plus business days against the two-day SLA, Standard Access running four-plus weeks against the ten-day target. Google cited a surge in new developer-token applications driven by AI-agent tooling as the primary cause. Every Claude Code, ChatGPT, and Cursor user wiring up a google ads mcp is adding pressure to the same review queue.
Practical implications. Apply for Basic Access early, even if you only plan to use the official read-only MCP. Be specific in the use case ("internal team analyzing campaigns through Claude Code with the official Google Ads MCP server" beats "AI integration"). Add the Cloud project number if you completed OAuth verification. Use a Test Account while Basic Access sits in queue.
For Pipeboard and Composio, you skip the developer-token application entirely. Both brokers provision a shared token under their own Google Cloud project and authenticate your specific Google Ads account through the OAuth flow on top. This is the single biggest reason to consider a third-party server while evaluating. Operators who picked Pipeboard were running real campaigns from Claude Code seven minutes after install while their Basic Access application sat in queue for two weeks.
Eight PMax and Search Workflows From Your Terminal
This is the operator-grade performance max mcp and Search list -- every workflow run on real accounts, not press-release theater.
1. Asset group creation for Performance Max. "Create a new asset group on the Summer Sale 2026 PMax campaign with these five product images, three headlines, two long headlines, four descriptions, and these audience signals." Claude builds the bulk-mutate request that Performance Max requires (asset group plus minimum-required assets in one call) and executes through Pipeboard or your custom server.
2. Audience signals for PMax campaigns. Audience signals are the highest-leverage knob on Performance Max. Claude reads your top-spending customer list, pushes it as a custom audience, and attaches it as an audience signal to the right asset group in one prompt. Pays back the install cost on most accounts.
3. Conversion uploads via the Conversion Adjustments API. "Upload these 1,200 offline conversions from the Shopify export, dedupe against existing uploads from the last 30 days, and report any rejections." Claude validates the schema, batches the API call, and writes a CSV of rejected rows.
4. Negative keyword sweeps across Search campaigns. "Pull every search term from the last 14 days, identify queries with high impressions and zero conversions, add the worst 50 as account-level negative keywords." Wasted spend that was sitting in your account for a month gets clipped in 90 seconds.
5. Search term reports with intent classification. Claude pulls the full search-terms report and classifies each query by intent. High-impression informational queries become content briefs for your blog. Commercial queries with low conversion become candidates for new ad groups. Google Ads becomes the keyword input for your SEO program rather than a separate silo.
6. Bid strategy switches with dry-run. "If I switched Brand Search from Manual CPC to Maximize Clicks, what does the historical data suggest the spend curve would look like?" Claude pulls the data, models the curve, reports a recommendation. You read, then ask Claude to make the switch.
7. Budget pacing across multi-campaign accounts. "It is the 22nd of the month, we are 40 percent through our $50,000 budget. Reallocate across the 12 campaigns so we hit $50,000 by month-end without dropping below 80 percent of the historical daily spend on the top three." Claude does the linear programming, you approve, Pipeboard executes the budget updates.
8. Attribution debugging across the conversion path. "We are seeing a 30 percent drop in reported conversions on the PMax campaign, but Shopify orders are flat." Claude pulls the conversion action setup, checks for tag-firing failures, validates the enhanced-conversion-data hashing, and surfaces the most likely cause.
The eight workflows are well within the API surface of any of the four servers covered. The operator question is which playbook you run them inside. The Shopify Kit ($199) ships 11 paid-ads files for exactly this stack, including a Google Performance Max playbook, a post-iOS-14 measurement stack, and the cross-channel diagnostic trees that triage workflow eight. The kit is the engagement framework. Claude Code plus the MCP is the execution layer.
Google Ads Plus Meta Plus Shopify: The Three-Channel Loop
Google Ads is the third leg of a three-channel operator stool. Meta on the social side, Shopify on the store side. Claude Code is where they share data and generate the cross-channel artifacts no single-platform UI can produce.
The shape of the loop. Creative generation lives in Higgsfield MCP (launched April 30, 2026). Distribution runs across Meta CLI and Meta MCP (launched April 29) and your Google Ads MCP. Conversion and store control runs through the Shopify AI Toolkit (launched April 9). Retention picks up where paid drops off through Klaviyo MCP.
The cross-channel question Claude can answer that no single-platform UI can: "We spend 40 percent on Meta, 35 percent on Google, 25 percent on TikTok. Native attribution overcounts because of last-click bias. Pull last-touch from Triple Whale, weight by marginal-ROAS curves from each platform's incrementality test, tell me where the next $10,000 should go." One prompt. Without the MCP stack, two days of spreadsheets and a $1,500 contractor invoice.
The diagnostic question Claude can answer when a campaign tanks: "PMax CPA jumped 40 percent Tuesday. Creative fatigue, audience saturation, a competitor bid spike, a tracking break, or yesterday's theme deploy?" Claude pulls Meta CAPI signal diagnostics, Google Ads conversion diagnostics, runs git log on the Shopify theme, and triangulates. Three weeks of consultant work in one session.
The Shopify Kit's cross-channel diagnostic-trees playbook walks this exact triage with the GAQL queries, Meta breakdown calls, Shopify metafield validations, and Klaviyo flow checks pre-built.
Connecting Google Ads to Triple Whale and Polar Analytics
Native attribution lies in every direction. Google overcounts conversions Meta drove. Meta overcounts conversions Google drove. Both undercount email and SMS. The DTC attribution layer sits on top of the platforms with cross-channel last-touch and incrementality models. Triple Whale and Polar Analytics are the two leaders as of May 2026, both with MCP-compatible surfaces you can wire next to your Google Ads MCP.
Triple Whale's MCP exposes the Sonar pixel data and the Lighthouse attribution model directly to Claude. Polar Analytics is similar: cross-channel ROAS by campaign, customer-LTV-by-source, and the marginal-ROAS curves you need to actually decide where the next dollar goes. Northbeam plays the same role with a stronger media-mix-modeling tilt. Pick one. Running two attribution layers simultaneously paralyzes decision-making.
The workflow that earns the install: the weekly cross-channel reallocation. Every Monday, Claude pulls last week's Google Ads, Meta, TikTok, and email-revenue numbers, normalizes through Triple Whale, computes marginal ROAS by channel, and writes a Markdown report with a recommended reallocation. You read it, push back on what looks wrong, approve. Claude executes. 25 minutes. Same job a $2,500-per-month attribution consultant did, running in your terminal.
If you are not yet running an attribution layer, native reporting through the MCP is enough to start.
Limits, Quotas, and the Beta Caveats
Things that bite when you start running Google Ads from Claude Code at scale.
API quotas matter. Basic Access caps you at 15,000 operations per day. A single GAQL query is one operation. A bulk-mutate updating 200 ad-group bids is one operation. Generous until you run a five-account weekly audit that fires 800 GAQL queries, then a daily dashboard refresh of 200 more.
Performance Max asset-group changes have caveats. Per the Google Ads API docs, asset groups in non-retail PMax campaigns must be created with minimum required assets in a single bulk-mutate, not piecemeal. Pipeboard handles this. Custom servers must implement the ordering correctly.
The April 2026 PMax asset-disapproval policy is live. As of April 7, 2026, all new Performance Max campaigns are subject to asset-level review from submission. Campaigns ship looking healthy, then get hit with disapprovals two days later. Build a daily disapproval-check workflow into your routine.
Rate limits are per-developer-token, not per-account. If you operate 30 client accounts under one developer token, all 30 share your 15,000-operation cap. Standard Access removes the daily cap; per-second rate limits still apply.
The official MCP is read-only and likely to stay that way. Google's choice to ship read-only first is consistent with constraining write operations when AI agents are the calling client. Do not plan a workflow around the official server gaining write in the next quarter.
Why Google's Own Agent Tools Aren't Enough
Google has been shipping AI inside the Ads UI for two years: Smart Bidding, asset auto-generation, Recommendations, the PMax suggestion engine. All real, all useful. The mismatch is not capability -- it is shape. These tools live inside Google Ads and only inside Google Ads. Every recommendation comes from inside the box.
A google ads mcp claude code setup inverts the box. The agent lives in your terminal with every other tool you have wired. "Is the Smart Bidding recommendation to raise CPA targets defensible given the gross-margin compression on the top three SKUs last week?" Smart Bidding cannot ask itself. Claude can. The google ads mcp supplies the Google-side data. The Shopify MCP supplies the margin data. The Meta MCP supplies the cross-channel ROAS context. Smart Bidding becomes one input, not the whole reasoning loop.
The right mental model is layered. Smart Bidding handles within-account, second-by-second optimization no human can match. Claude plus the MCPs handles cross-platform, cross-week, cross-portfolio strategic questions Smart Bidding has no inputs for. The PPC consultant and the in-house media buyer are no longer competing against Smart Bidding. They are competing against operators who have wired the performance max mcp stack and run it from one session.
Where to Go From Here
Pick a server today, install your google ads mcp claude code setup in 15 minutes, run the eight workflows on a Test Account or your real account, and decide before you commit to the multi-week Basic Access approval.
Read the MCP basics guide if you have not wired an MCP yet. Install the official Google Ads MCP for read-only analytics, or claude mcp add --transport http pipeboard-google-ads https://google-ads.mcp.pipeboard.co/ for full read-write. Wire the Meta MCP and Shopify AI Toolkit next. Add Klaviyo MCP for retention and Higgsfield MCP for creative. Enable MCP Tool Search once you cross three connected servers. Cursor users can follow our Cursor MCP setup walkthrough. Browse the curated MCP list for adjacent servers (GA4, Slack, Notion).
Google Ads MCP is the missing leg of the operator stack. Meta runs paid social. Shopify runs the store. Higgsfield generates creative. Klaviyo runs retention. Google Ads runs paid search and Performance Max. With all five wired, one operator in one session executes the work that twelve months ago required a senior media buyer, a PPC analyst, a retention strategist, a CRO consultant, and an attribution contractor.
That is what the Shopify Kit ($199) closes. Eleven paid-ads playbooks for the Meta-plus-Google-plus-TikTok stack, a dedicated Google PMax framework, the post-iOS-14 measurement stack, the cross-channel diagnostic trees that triage when a campaign tanks. Plus 31 SEO files that turn Google Ads search-term reports into a content roadmap, 10 CRO files for the landing-page side, eight retention flows, and eight named playbooks from Quick Wins Sprint to 90-Day Audit Sequence. Twenty-person Shopify consultancy worth of operator judgment in your Claude Code, paid once, updated forever. The MCPs give Claude hands. The kit tells the hands what to build.
Last updated on
