TL;DR. A Google Ads MCP server lets Claude query and (sometimes) modify Google Ads accounts over the Model Context Protocol. It's great for analysts running ad-hoc reports, painful for production ad operations. The Google Ads API has developer-token approval, per-customer quotas, and OAuth refreshes that MCP setups don't handle gracefully. Sprites uses a Google Ads partner integration — which is why production teams use it instead.
What is Google Ads MCP?
Google Ads MCP is a Model Context Protocol server that wraps the Google Ads API so Claude (or any MCP-compatible client) can call it through natural language. Community servers like cohnen/mcp-google-ads expose a handful of endpoints: list campaigns, fetch reports, edit keyword bids.
MCP is Anthropic's open standard for connecting AI assistants to external systems. A Google Ads MCP server is the adapter layer between Claude and Google's APIs.
How to set up Google Ads MCP with Claude
Apply for a Google Ads developer token. Go to Google Ads → API Center. Basic access review takes a few business days. Production access requires demonstrating usage and can take weeks.
Create a Google Cloud project and enable the Google Ads API. Create OAuth 2.0 credentials and download
credentials.json.Run the OAuth flow to generate a refresh token. Store it alongside your developer token and customer ID.
Install an MCP server. Most teams use a community npm or Python package. Configure it with the developer token, customer ID, and refresh token.
Register the server in Claude's
claude_desktop_config.jsonwith the path to the server binary and the required environment variables.Restart Claude Desktop. The Google Ads tools become available in the tool picker.
What works
Ad-hoc reporting. "What was CPA by campaign last week?" Claude generates the GAQL query, runs it, and summarizes.
Account audits. Read-only sweeps across campaigns and ad groups.
Keyword research sketches. Listing the keywords in a campaign and asking for expansion ideas.
What breaks in production
1. Developer token access
Basic access developer tokens are rate-limited to 15,000 operations per day per token. Production-level access requires Google approval — and the review process expects an actual product, not an MCP experiment.
2. Per-customer API quotas
Google Ads API enforces quota per login-customer-id and per method. Agents fan out requests ("look at every ad in every ad group") and quickly hit limits that trigger RESOURCE_EXHAUSTED errors.
3. OAuth refresh fragility
Refresh tokens can be invalidated when Google detects unusual activity or when scopes change. MCP setups store tokens on disk and rarely auto-recover from invalidation.
4. Narrow tool surface
Most Google Ads MCP servers cover a subset of the API. Performance Max asset-level operations, Smart Bidding strategy changes, and Recommendations ingestion are usually missing.
5. Performance Max is opaque
PMax exposes limited reporting surfaces. Tools without a dedicated ingestion and enrichment layer can't surface asset-level performance. Sprites does that separately; MCP servers generally don't.
6. No approval gates
When Claude calls a write endpoint, the change goes live. There's no "here's the plan, approve?" surface. For production Google Ads spend, that's unacceptable.
Google Ads MCP vs Sprites
| Capability | Google Ads MCP + Claude | Sprites |
|---|---|---|
| Read campaign data | Yes | Yes |
| GAQL query generation | Yes | Yes — plus pre-built reports |
| Developer token management | Manual | Partner-managed |
| API quota headroom | Basic only | Partner-grade |
| Performance Max asset-level insights | Usually not | Yes |
| Smart Bidding adjustments | Often unsupported | Yes |
| Approval gates | None | Required on every write |
| Audit log + rollback | None | Yes |
| Cross-channel (Meta, LinkedIn, SEO) | No | Yes |
When Google Ads MCP makes sense
Solo analysts, engineers prototyping tooling, and anyone who wants ad-hoc reporting from Claude with minimal setup. It's a legitimately good analyst workflow.
When Sprites makes sense
Switch when any of the following apply:
You're running production spend. Approval gates, audit log, and rollback become required.
You need Performance Max coverage. Asset-level insights and creative refresh require surfaces MCP servers don't expose.
Multiple teammates. Shared workspace instead of per-laptop MCP configs.
Cross-channel ops. Meta, LinkedIn, SEO in the same agent.
Reliability matters. Partner-grade quotas and token management instead of manual refreshes.
Takeaway
Google Ads MCP with Claude is the analyst's sandbox. Sprites is the operator's tool. For live spend, Performance Max coverage, and cross-channel work, the purpose-built AI marketing agent wins.
See the full Sprites vs Google Ads MCP comparison →