Google Ads & GA4 MCP server for Claude Code and Claude Desktop
Server has 18 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, critical gaps exist: (1) Output schemas are undocumented, tools return markdown strings with no structured schema definition, forcing LLMs to parse unstructured text; (2) Parameter descriptions lack format constraints (e.g., date format 'YYYY-MM-DD' stated in description but not in schema); (3) No error handling guidance, errors return raw exception messages instead of actionable recovery steps; (4) Write tools (set_campaign_status, set_keyword_bid, etc.) lack explicit dry_run documentation in descriptions despite having the parameter; (5) Safety/plan tools (create_plan, get_plan, execute_plan) have minimal descriptions (under 50 chars) that don't explain the safety workflow. Naming is strong (list_, get_, set_, create_, execute_), but schema completeness and error guidance are weak.
Add negative keywords to a campaign or ad group.
Create a safety plan for a proposed change. Returns a plan ID that can be referenced later.
Execute a previously created and reviewed plan.
Get performance metrics at the individual ad level for a Google Ads account. Useful for identifying underperforming ads, comparing creative variations, and finding ads to pause or improve.
Get performance metrics for all campaigns in a Google Ads account. Returns impressions, clicks, cost (USD), conversions, CTR, and average CPC. Results are sorted by spend (highest first), limited to 100 campaigns.
List all event names and their 30-day count in a GA4 property.
Output schemas undocumented. All tools return markdown strings with no structured schema definition. LLMs cannot parse results reliably or chain downstream calls. Example: get_campaign_performance returns formatted table string, not JSON with campaign_id, cost_usd, conversions fields.
Error handling returns raw exception messages instead of actionable recovery guidance. Example: 'Exception: Invalid customer_id' tells LLM nothing about what to do next. Should return 'Invalid customer_id format. Use list_google_ads_accounts() to discover valid IDs.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Get real-time active users in Google Analytics 4, optionally broken down by dimensions.
Run a custom Google Analytics 4 report with custom dimensions and metrics.
Get performance metrics for all keywords in a Google Ads account. Returns keyword text, match type, Quality Score, bid, clicks, cost, and conversions. Useful for identifying low-QS keywords, high-spend with no conversions, and keywords that are candidates for negative keyword lists.
Retrieve a previously created plan by ID.
Get actual search terms that triggered ads in a Google Ads account. Essential for finding negative keyword opportunities and understanding real user intent. Returns the top terms by spend.
List all accessible Google Analytics 4 accounts and properties.
List all Google Ads accounts accessible with the current credentials. Use this first when the user hasn't specified a customer_id, or to discover which accounts are available.
Execute a custom Google Ads Query Language (GAQL) query. Use this for advanced analysis not covered by the other tools. GAQL syntax: SELECT fields FROM resource WHERE conditions ORDER BY ... LIMIT n Common resources: campaign, ad_group, keyword_view, search_term_view, ad_group_ad, campaign_criterion, ad_group_criterion
Pause, enable, or remove an ad group.
Update the daily budget for a campaign.
Pause, enable, or remove a campaign.
Update the bid (CPC) for a keyword.
Safety/plan tools (create_plan, get_plan, execute_plan) have minimal descriptions (<50 chars). Descriptions do not explain the safety workflow, when to use plans, or what happens on execute. LLMs cannot determine if a plan is needed before calling set_campaign_budget.
Write tools lack explicit dry_run documentation in descriptions. Parameter exists (dry_run: boolean, default true) but descriptions do not mention that changes are previewed by default. LLMs may not realize they need to set dry_run=false to apply changes.
Parameter format constraints not in schema. Dates described as 'YYYY-MM-DD' in text but no JSON Schema format/pattern field. Limit parameters (e.g., limit: integer) lack min/max bounds. LLMs cannot validate inputs before calling.