MCP Server in action: Have you talked to your Google Ads account today?

The way we work with Google Ads accounts is fundamentally changing. What if you no longer had to navigate through tedious reports, but could simply talk to your account? “Show me the top performers of the last 30 days” or “Create a keyword analysis for campaign XYZ” – and within seconds you receive precise answers with concrete recommendations for action.

This is precisely what the combination of Claude AI and MCP (Model Context Protocol) servers makes possible. In this article, I’ll show you how we at Prange Webmasterei use this technology to revolutionize Google Ads management.

What is MCP and why is it a game-changer?

MCP servers provide structured data, tools, or actions that an AI model like Claude can interact with. Think of them as interfaces that allow AI access to internal APIs, databases or company systems – in our case, Google Ads accounts, customer databases and SEO tools.

Local vs. Remote MCP Server

Local installation:

• Quick installation on the desktop PC
• Usable only locally
• No authentication required
• Ideal for individual workstations

Remote Installation:

• Central installation on remote server
• Authentication required (security is your responsibility)
• Usable with multiple workstations
• Perfect for teams and agencies

The MCP server landscape for Google Ads

For our work, we use a combination of different MCP servers, each fulfilling specific tasks:

Work tools

Google Ads MCP Cohen

• Lists accounts on
• Executes GAQL (Google Ads Query Language) queries
• Direct access to account data
• GitHub: cohnen/mcp-google-ads

DataForSEO MCP

• SEO data requests
• Keyword research for the German market
• Competitive analysis
• DataForSEO MCP Documentation

Notion MCP

• Communication with Notion Pages and databases
• Structured data storage for customer information
• Ideal for master data management
• MCP Servers: Notion

Other useful servers

• Google Cloud : Generative editing of Google Cloud setups
• Google Search Console : Data retrieval from the Search Console

Setup: Configure Claude Desktop correctly

Which LLM clients are suitable?

Claude Desktop ✅

• MCP is directly integrated into the desktop app.
• Easy configuration via Settings
• Best integration and performance

ChatGPT Desktop ❌ (not yet)

• MCP is not yet integrated into the desktop app.
• Alternatives: Cursor IDE or VS Code Extension

MCP configuration in Claude Desktop

1. Open Claude Desktop
2. Settings > Developer > Local MCP Servers
3. Edit MCP configuration file

Example configuration for Google Ads Server:

{
  "mcpServers": {
  "googleAdsServer": {
  "command": "C:\\Users\\info\\claude\\mcp-google-ads\\.venv\\Scripts\\python.exe",
  "args": [
  "C:\\Users\\info\\claude\\mcp-google-ads\\google_ads_server.py"
  ],
  "env": {
  "GOOGLE_ADS_CREDENTIALS_PATH": "C:\\Users\\info\\claude\\mcp-google-ads\\credentials\\credentials.json",
  "GOOGLE_ADS_DEVELOPER_TOKEN": "your_token",
  "GOOGLE_ADS_LOGIN_CUSTOMER_ID": "xxx-xxx-xxxx"
  }
  }
  }
}

The three levels of context

In order for Claude to work effectively with Google Ads, three levels of context are needed for the LLM: knowledge about Google Ads, knowledge about the customer, and current data from the Google Ads account.

1. Google Ads Knowledge (GitHub as a Knowledge Base)

The foundation is a structured GitHub repository containing comprehensive Google Ads knowledge in Markdown format. Why GitHub?

• Version control : Every change is traceable.
• Markdown format : Optimally processed by LLMs
• Claude Desktop Integration : Direct access at the project level

Structure of the Knowledge Base:

├── GENERAL_RULES.MD # Routing & Standards
├── QUICK_COMMANDS.MD # Most frequently used commands
├── NOTION/
│ ├── API_GUIDE.MD
│ └── DATABASE_SCHEMA.MD
├── GOOGLE_ADS/
│ ├── GAQL_REFERENCE.MD # Central query collection
│ ├── API_LIMITATIONS.MD
│ ├── FUNDAMENTALS.MD
│ ├── REFERENCE/
│ │ ├── KEYWORDS.MD
│ │ ├── NEGATIVE_KEYWORDS.MD
│ │ ├── CAMPAIGN_TYPES.MD
│ │ └── ACCOUNT_STRUCTURE.MD
│ ├── SEARCH/
│ │ ├── CAMPAIGN_CREATION.MD
│ │ ├── TERMS_ANALYSIS.MD
│ │ └── CAMPAIGN_OPTIMIZATION.MD
│ ├── PMAX/
│ │ ├── CAMPAIGN_CREATION.MD
│ │ ├── ASSET_DESIGN.MD
│ │ └── VIDEO_CREATION.MD
│ ├── WORKFLOWS/
│ │ ├── ACCOUNT_HEALTH_CHECK.MD
│ │ ├── NEW_CUSTOMER_ONBOARDING.MD
│ │ └── PLACEMENT_ANALYSIS.MD
│ └── UTILITIES/
│ └── EDITOR_IMPORT.MD
└── DATA_FOR_SEO/
  ├── API_GUIDE.MD
  └── COST_OPTIMIZED_QUERIES.MD

Example: Explaining keywords correctly

Instead of simply saying “Claude should know his way around”, we document precisely:

1. Fundamentals & Normalization
2. Positive keywords (match types)
3. Negative keywords (match types)
4. Critical differences
5. Special characters in keywords
6. Best Practices & Workflows
7. GAQL queries for keyword analysis

GAQL Reference: The central query collection

A key element is theGAQL_REFERENCE.MD with all important queries:

-- Campaign Performance
SELECT  
  campaign.id,  
  campaign.name,  
  campaign.status,
  campaign.advertising_channel_type,
  metrics.impressions,  
  metrics.clicks,
  metrics.cost_micros,  
  metrics.conversions,
  metrics.conversions_value
FROM campaign
WHERE campaign.status = 'ENABLED'
  AND segments.date DURING LAST_30_DAYS
ORDER BY metrics.cost_micros DESC

Anyone wanting to test a stripped-down version of this repository can do so here . The quality will certainly be limited due to the absence of many tools and processes. However, it’s perfectly acceptable to test the difference in quality between keyword and exclusion decisions and the behavior of the LLM (Large List Management) without project knowledge. A full version is in production use at our site and is available for purchase.

2. Customer requirements (Notion as master database)

Notion serves as a central CRM system for all customer data. Why Notion?

• API-controllable : Fully accessible via MCP server
• Structured database : Clear data fields for consistent queries
• Controllable access : Granular permissions possible
• Easy maintenance : No complex database administration.

Stored customer data:

• ads_account_currencyCurrency (e.g. EUR)
• audienceDescription of the target group
• brand_name: Brand name
• budget_monthlyMonthly budget
• descriptionCustomer description
• policyGuidelines, prohibitions, legal requirements
• ocidGoogle Ads Account ID
• languages: Languages in the account
• strategyStrategic orientation
• uspsKey company USPs
• website: Web address

Example workflow:

User: "Create a search campaign for account 123-456-7890"

Claude:  
1. Loads customer data from Notion
2. Extract USPs, target audience, budget
3. Takes policy requirements into account
4. Generates a suitable campaign structure

3. API data (live access to Google Ads)

Direct access to Google Ads data via the MCP server enables:

• Live data : Current performance metrics
• GAQL Queries : Flexible Data Queries
• Account overview : Quick access to all accounts
• Performance analyses : Detailed evaluations

The working method in practice

This is what the typical workflow looks like:

Interaction example:

User: "Create a list of negative keywords"  
  for account 123-456-7890"

Claude:
1. Retrieves account data via the Google Ads API
2. Analyzes search terms from the last 30 days
3. Identifies problem terms
4. Generates negative keywords with correct match types
5. Provides TSV file for Google Ads Editor import

User: "Exclude the following terms: [free],  
  [free], [for free]"

Claude: "Excellent! I've used the terms as an exact match."   
  Negative aspects added, as they are individual cases.  
  These are problematic words. The list is ready.   
  for the import."

Use Case 1: Account Health Check – The “free initial analysis”

An account health check is the perfect way to convince potential customers. With Claude, a comprehensive analysis takes only minutes instead of hours.

What is being tested?

Tracking & Measurement:

• Conversion tracking configuration
• Enhanced Conversions & Consent Mode v2
• GA4 linkage & attribution model
• Auto-Tagging (GCLID)

Budgets & Performance:

• Budget utilization & Lost Impression Share
• Campaign performance (ROAS, CPA)
• Cost per Conversion Development
• Paused/Inactive Campaigns

Campaign Setup:

• Bidding strategies (Smart Bidding Check)
• Conversion thresholds PMax (<30/month)
• Geo-targeting settings
• Target groups & customer match

Ads & Assets:

• RSA Asset Count & Ad Strength
• PMax Asset Performance Labels
• Ad extensions active
• Video Assets

Negative keywords & exclusions:

• Account-Level Negatives
• Joint negative lists
• Expensive non-converting search terms
• Keyword overlaps

Quality:

• Quality Score Keywords (<5)
• CTR below benchmark (<5%)
• Double conversions
• IP exclusions

Workflow:

User: "Do an account health check for  
  Buhle Funerals OHG (xxx-xxx-xxxx)"

Claude performs the following steps automatically:
1. Loads customer data from Notion
2. Executes 15-20 GAQL queries
3. Analyze all areas systematically.
4. Create a structured report with priorities.
5. Provides specific recommendations for action

Output example:

# Account Health Check: Buhle Bestattungen OHG

## Executive Summary
- Account Status: ⚠️ Needs optimization
- Performance: 📊 Mixed (ROAS 3.2, Target: 4.0)
- Tracking: ✅ Well configured
- Budget: ⚠️ 35% wasted

## Critical Findings

### 🔴 High priority
1. **No Account-Level Negatives**
  - Impact: ~€450/month waste
  - Action: Add 47 negatives
   
2. **PMax too few conversions**
  - Current: 18 conv/month
  - Required: 30 conversions per month
  - Action: Reallocate budget to search

Use Case 2: Campaign Creation – From Zero to Finished Campaign

Creating a new search campaign is traditionally a process that takes several hours. With Claude and the MCP servers, this is reduced to 15-30 minutes.

Workflow overview:

1. Customer selection via Notion (MCP)

• Loading customer data from Notion database
• Extract USPs, policy, target group, and budget

2. Landing page analysis via DataForSEO (MCP)

• Enter landing page URLs
• Extract keywords for site
• Identify content topics

3. Keyword research via DataForSEO Labs (MCP)

• Generate keyword ideas
• Find related keywords (Depth 2)
• Analyze search intent (transactional/informational)

4. Intelligent Clustering & Ad Group Creation

• Cluster keywords thematically (TF-IDF + K-Means)
• Automatically assign match types
• Generate RSAs (15 Headlines, 4 Descriptions)

5. Negative keyword suggestions

• Intent-based exclusions (informational > 70%)
• Universal Negatives (Standard List)
• Cross-negation between ad groups

6. Artifact Output & Export

• Markdown report (human-readable)
• Google Ads Editor TSV (directly importable)

Example interaction:

User: "https://webmasterei-prange.de/shopware-bigquery-integration/
  You can find the domain in Notion.
  I would like to specifically address the topic of the landing page.  
  Create a Google Ads search campaign.
  Find suitable keywords, cluster them,  
  Make suggestions for text assets.

Claude performs the following steps automatically:

1. Loads customer data from Notion (Webmaster Prange)
2. Analyze landing page with Playwright
3. Conduct DataForSEO keyword research:
  - "Shopware BigQuery Integration"
  - "Shopware Data Analysis"
  - "E-Commerce Analytics"
  - etc.
4. Cluster keywords by topic
5. Generates 3-5 ad groups, each containing:
  - 10-20 keywords (different match types)
  - 15 headlines, 4 descriptions
  - Relevant sitelinks
6. Create a list of negative keywords
7. Export everything as TSV for Google Ads Editor

Typical output:

# Search campaign: Shopware BigQuery integration

## Ad Group 1: BigQuery Integration
**Keywords (Exact Match):**
- [shopware bigquery]
- [shopware bigquery integration]
- [shopware data analysis]

**Keywords (Phrase Match):**
- "Connecting Shopware with BigQuery"
- "Shopware data in BigQuery"

**RSA Headlines:**
1. Shopware & BigQuery Integration
2. Data analysis for your shop
3. Automatic BigQuery integration
... (15 in total)

**Descriptions:**
1. Seamlessly connect your Shopware shop to Google  
  BigQuery. Professional integration for maximum performance.   
  Data analytics power.
... (4 in total)

Use Case 3: Search Term Analysis – Optimizing Search Terms

Regular analysis of search terms is essential to avoid wasting budget and to identify new keyword opportunities.

Workflow:

User: "Do a search term analysis for the campaign"  
  '[TXT-REGIO] Funeral Director Kassel' in the account  
  Buhle Funerals OHG (xxx-xxx-xxxx),  
  Domain: www.example-bestattungen.de"

1. Extract data:

• Query: Performer (with conversions)
• Query: Under Target (ROAS < Target)
• Query: Non-Performer (no conversions)

2. Analysis:

• Evaluate performers based on ROAS, conversion rate, and CTR.
• Evaluate Under Target based on distance to the target
• Categorizing non-performers by intent type

3. Recommendations:

• Create keyword suggestions with Match Type
• Determine negative keywords with level (Account/Campaign/Ad Group)

4. Report:

• Create a summary
• Group keywords by priority
• Group negatives by level
• Calculate impact

5. Implementation:

• TSV file for new keywords
• TSV file for negative keywords
• Prepares for import into Google Ads Editor

Output example:

# Search Terms Analysis: Funeral Director Kassel

## 📊 Summary
- Analyzed search terms: 247
- With conversions: 18 (7.3%)
- Waste: ~€280/month
- Quick Win Potential: ~€170/month

## 🚀 High Performer (add as keywords)

| Search Term | Conversions | Conv Rate | ROAS | Recommendation |
|-------------|-------------|-----------|----------|------|------------|
| Funeral director Kassel North | 3 | 15.2% | 8.2 | [Exact] |
| affordable funeral services Kassel | 2 | 12.1% | 6.8 | "Phrase" |

**Action:** Add 8 new keywords → +€45 conversion value/month

## 🔴 Problem Terms (exclude as negatives)

| Search Term | Cost | Impressions | Type | Level |
|-------------|------|-------------|-----|------|
| Funeral director training | €45 | 892 | Info | Campaign |
| Funeral director salary | €32 | 456 | Info | Campaign |
| Free obituary | €28 | 234 | Free | Account |

**Action:** Add 23 negatives → -€280 Waste/Month

What works well – and what doesn’t (yet) work.

✅ What works exceptionally well:

1. Context understanding: The AI easily understands the context of Google Ads, campaign structures, and best practices.

2. Text generation: Ad texts can often be adopted directly, often with surprisingly good target group appeal.

3. Creativity: Sometimes more creative solutions than human PPC managers – unbiased perspective.

4. Text lengths: Character limits are adhered to correctly 98% of the time (Google Ads headlines: 30 characters, descriptions: 90 characters).

5. Editor Exports: TSV files have a low error rate (< 2%) and are directly importable.

6. Speed: What used to take hours is now done in minutes.

❌ What doesn’t work (yet):

1. Uncritical adoption: LLM content should always be checked – no blind adoption!

2. Calculations: Calculations often only provide approximations. For exact numbers: Use Python code.

3. Strategic planning: Complex strategic decisions require human judgment.

4. Automated takeover: Direct write operations via API are not possible (and are also not desired for security reasons).

5. Large amounts of data: Token limits restrict the amount of data that can be processed per request.

Data Privacy: Consumer vs. Enterprise

A critical point when using AI tools is data privacy. Consumer and enterprise versions differ fundamentally in this respect:

Claude Consumer (Free/Pro/Max)

❌ NOT suitable for confidential data:

• Can be used for model training (if opt-out is not set)
• Data storage: permanent (several years)
• Access possible for Anthropic staff
• No Data Processing Agreement (DPA)
• Server location: USA
• No Zero Data Retention Option

Claude Enterprise / API (min. 5 people)

✅ Suitable for confidential data:

• Not to be used for training (contractually excluded)
• Data storage: configurable (e.g., 30 days)
• Access severely restricted
• Full DPA available
• EU hosting by arrangement
• Zero Data Retention (optional)

Recommendation: For professional agency work with customer data, Claude Enterprise/API with EU hosting is mandatory!

Advantages & limitations of the solution

✅ Advantages:

Interactive collaboration: Flexible dialogues with AI instead of rigid scripts.

Human control: Every step can be checked and adjusted.

Speed: Drastic time savings for routine tasks (70-80%).

Scalability: Once set up, the system works for all accounts.

Learning capability: The system improves through updates to the knowledge base.

⚠️ Limits:

Trust in numbers: Calculations must be validated.

Not fully automatable: Human approval remains necessary (and desired).

Token limits: Very large amounts of data must be split.

Data privacy is critical: Enterprise version with EU hosting required.

Practical tips for use

1. Using Python for calculations

❌ Incorrect: "Calculate my ROAS"
✅ Correct: "Use Python to calculate the ROAS"

This ensures that calculations are accurate.

2. Context is king

❌ False: "Be a Google Ads expert and analyze..."
✅ Correct: Comprehensive knowledge base + customer data +  
  Provide API access

Superficial prompts are not enough – real context makes the difference.

3. Document workflows

Successful processes should be documented in the knowledge base so that they can be repeated and optimized.

4. Iterative work

Start with simple analyses and gradually increase the complexity.

5. Incorporate validation

Especially with critical data (budgets, conversions), a plausibility check should always be performed.

Summary: The Future of PPC Management

The combination of Claude AI and MCP servers fundamentally changes Google Ads management:

From rigid tools to flexible dialogues: Instead of navigating through endless menus, you have simple conversations.

From hours of analysis to seconds: Account health checks, keyword research and campaign creation are dramatically accelerated.

From individual solutions to integrated workflows: Customer data, API access and best practices work seamlessly together.

From reactive to proactive: Regular analyses become so easy that problems are detected earlier.

The three success factors:

1. Structured Knowledge (GitHub Knowledge Base)
2. Central customer data (Notion/CRM)
3. API integration (MCP Server)

Important NOTE:

This workflow is not a replacement for human expertise, but a tool for increasing efficiency . Strategic decisions, quality control, and final approval remain in human hands – and that’s a good thing.

---

Disclaimer: The concepts discussed are considered from a purely technical perspective. You should always involve your company’s data protection officer in the implementation process. This article does not constitute legal advice.