How to Use Claude with Google Ads: CSV, API and MCP Workflows Explained

Claude can work with Google Ads at three levels: a manual CSV review, a controlled read-only API pipeline, or an MCP tool layer. Start with CSV. Use API or MCP only when access, logging, data retention and human approval are explicit.

1. Level 1 - CSV/manual: lowest technical risk, easy to audit and suitable for an initial diagnosis.
2. Level 2 - Google Ads API read-only pipeline: repeatable reporting, but requires OAuth, developer-token governance and a controlled backend.
3. Level 3 - MCP/tool layer: useful for conversational access to approved functions, but the connector still needs authentication, allowlisted tools and audit logs.

Create one Claude Project per account or controlled account group. Add business context, conversion definitions, naming conventions and approved diagnostic rules. Upload only the exports needed for the question. Anthropic documents CSV support; spreadsheet support depends on the analysis environment. Its Google Drive integration covers Google Docs, not Google Sheets.

Anthropic: file uploads, Projects, Google Docs integration.

You are reviewing a Google Ads account as a senior acquisition and measurement specialist. Work in evidence-first mode. Do not confuse correlation with causation and do not infer missing tracking events.

Inputs: campaign export, search-term export, conversion-action export, change history and business context.

Protocol:
A. Inventory each file, date range, currency, attribution fields and missing columns.
B. Reconcile campaign totals across files; report discrepancies before diagnosis.
C. Build a diagnostic tree covering traffic quality, auction/creative, landing page, conversion tracking, Consent Mode, value import and bidding constraints.
D. For every finding provide: evidence, confidence (high/medium/low), business impact, verification step and reversible action.
E. Produce a read-only priority plan: fix measurement first, then waste, then growth.
F. Never publish, pause, change bids or add negatives. Return an explicit proposed-change file for human approval.

The Google Ads API can reproduce UI reports with GAQL, including search-term views. Access inherits Google Ads account roles, and developer tokens are governed by access levels and permissible-use rules. Build the backend so Claude receives only the fields and functions required for the current task.

Google: access model, access levels, UI reports and GAQL, GAQL structure.

Google does not publish an official Claude MCP server in the Google Ads documentation reviewed for this guide. A community or vendor MCP server is therefore a third-party integration, not an official Google connector. Inspect its code, scopes, logging and retention before use.

• Default to read-only reporting and deny campaign mutation tools.
• Keep OAuth tokens and developer credentials on the server, never inside prompts or browser code.
• Allowlist accounts, customer IDs, GAQL resources and tools.
• Require a proposed-change artifact and human approval before any write path is enabled.
• Log every query, tool call, result, approval and rollback reference.

Use Claude for long-context reconciliation, diagnostic trees, documentation review and proposed change plans. Do not use it as an unattended campaign operator when tracking, attribution or business constraints are unresolved.

For a simpler manual workflow, see the ChatGPT Google Ads search-term audit.

If you want this workflow adapted to a real account, start from the CreatikLab Google Ads service platform. Lia gathers the problem, preserves the evidence and routes the brief to a senior specialist.

Sources reviewed: August 12, 2026.