mcp-google-ads MCP — 사용 사례, 설치 & 라이브 데모

왜 쓰나요

핵심 기능

GAQL passthrough — full query power, not just pre-baked reports
OAuth (for individuals) or Service Account (for automation)
Campaign, ad group, ad, and keyword metrics
Impressions, clicks, conversions, cost, CTR, avg CPC
Python 3.11+; .env-based config

라이브 데모

실제 사용 모습

google-ads.replay ▶ 준비됨

0/0

설치

클라이언트 선택

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "google-ads": {
      "command": "uvx",
      "args": [
        "mcp-google-ads"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config 열기. 저장 후 앱 재시작.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "google-ads": {
      "command": "uvx",
      "args": [
        "mcp-google-ads"
      ],
      "_inferred": true
    }
  }
}

Cursor는 Claude Desktop과 동일한 mcpServers 스키마 사용. 프로젝트 설정이 전역보다 우선.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "google-ads": {
      "command": "uvx",
      "args": [
        "mcp-google-ads"
      ],
      "_inferred": true
    }
  }
}

Cline 사이드바의 MCP Servers 아이콘 클릭 후 "Edit Configuration" 선택.

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "google-ads": {
      "command": "uvx",
      "args": [
        "mcp-google-ads"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "google-ads",
      "command": "uvx",
      "args": [
        "mcp-google-ads"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json

{
  "context_servers": {
    "google-ads": {
      "command": {
        "path": "uvx",
        "args": [
          "mcp-google-ads"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add google-ads -- uvx mcp-google-ads

한 줄 명령. claude mcp list로 확인, claude mcp remove로 제거.

사용 사례

실전 활용법: mcp-google-ads

Weekly campaign health check across accounts

👤 Paid marketing managers, agencies ⏱ ~20 min intermediate

언제 쓸까: Weekly — which campaigns are burning budget with poor CTR / high CPA?

사전 조건

Google Ads Developer Token (approved) — ads.google.com > Tools > API Center — 1-3 business days
OAuth client OR service account with Ads access — Google Cloud Console

흐름

List accounts

List all Google Ads accounts I can access.✓ 복사됨

→ MCC + child accounts
Pull performance

For each account, pull last 7 days: campaign name, cost, clicks, conversions, CPA. Flag campaigns where CPA > $100.✓ 복사됨

→ Ranked hot-spot list
Propose action

For the worst 3 campaigns, suggest: pause, lower bid, or investigate further based on the numbers.✓ 복사됨

→ Action plan

결과: Weekly hotspot review in minutes.

함정

Attribution windows differ by conversion action — Include segments.conversion_action_name and check each separately

함께 쓰기: notion

Mine search terms for negative keywords

👤 Performance marketers ⏱ ~15 min intermediate

언제 쓸까: Regular account hygiene — catch wasteful queries before they add up.

흐름

Pull search terms

GAQL: search_term_view for last 30 days, columns [search_term, campaign, cost, conversions]. Cost > $10 and conversions = 0.✓ 복사됨

→ Wasteful query list
Group and recommend

Group by recurring phrases. Suggest negative keywords to add at account or campaign level.✓ 복사됨

→ Negative list with level

결과: A tidy negative-keyword list to apply.

함정

Broad-matching negatives can block legit traffic — Use exact or phrase negatives first; broad only when very confident

Audit ad copy performance across a campaign

👤 Copywriters, PPC leads ⏱ ~15 min intermediate

언제 쓸까: You launched 3 responsive ads and want to know which combos perform.

흐름

Pull ad stats

For campaign X, list all ads + their 30-day CTR + conv rate. Sort by conv rate desc.✓ 복사됨

→ Ad performance ranked
Recommend

What patterns distinguish top-performing ad copy? What should I iterate on?✓ 복사됨

→ Copy insights

결과: Data-driven creative iteration.

함정

Low-volume ads look worse than they are due to variance — Require minimum 100 impressions before comparing rates

조합

다른 MCP와 조합해 10배 효율

google-ads + notion

Weekly marketing digest

Pull last week's Google Ads performance + biggest wins/losses; create a Notion page in 'Marketing Weekly'.✓ 복사됨

google-ads + postgres

Join Google Ads data with your DB for full-funnel view

Get conversion counts from Google Ads for campaign X. Query the DB for actual closed deals originating from that campaign.✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출	비용
list_accessible_accounts		Discover the account graph	1 API call
run_gaql	customer_id, query: str	Any custom report	1 API call
campaign_performance	customer_id, date_range	Quick campaign digest	1 API call
ad_performance	customer_id, campaign_id?, date_range	Creative audit	1 API call

비용 및 제한

운영 비용

API 쿼터: Google Ads API: generous per-day quota; rate-limited per QPS
호출당 토큰: Results depend heavily on date range and segments — cap rows
금액: API is free; you pay for ads separately
팁: Always include a date range in GAQL. Running without LIMITs can return 100k rows.

보안

권한, 시크릿, 파급범위

최소 스코프: Read-only on the Ads accounts where possible

자격 증명 저장: Service account JSON or OAuth refresh token — env vars or secret manager

데이터 외부 송신: Performance data goes to your LLM provider

절대 부여 금지: Admin on the Ads account to an agent that does more than read

Developer Token approval has tiers — test tokens have low quota. Apply for production.
Service account must be added as a user in Ads MCC with appropriate role.

문제 해결

자주 발생하는 오류와 해결

AUTHENTICATION_ERROR or PERMISSION_DENIED

Service account email not added to MCC with access, or developer_token wrong. Check ads.google.com > Access & Security.

Too many metrics for segment

GAQL has restrictions on which metrics can be combined with which segments. Query a simpler subset.

확인: Read the GAQL compatibility table

Test token quota exceeded

Apply for production token; until then, test on small accounts with low query frequency.

대안

mcp-google-ads 다른 것과 비교

대안	언제 쓰나	단점/장점
Google Ads UI / Editor	You only occasionally pull reports	Manual; no LLM insight
Meta Ads MCP	You also run Meta ads	Different platform; own auth

mcp-google-ads