Connect Claude to Google Ads: full MCP setup guide for 2026
Claude is the most capable AI for campaign analysis today — it handles large context windows well, writes structured reasoning about attribution gaps, and follows multi-step instructions reliably. SpendSignoff connects it to your live Google Ads data in about 10 minutes.
Why Claude works well for ad operations
Three things make Claude a strong fit: its 200K-token context window can hold an entire account snapshot including all campaigns, ad groups, and 30-day metrics without truncation; its structured output is reliable enough to drive typed MCP tool calls with low hallucination on field names; and Claude's extended thinking mode (available in Claude 3.7 Sonnet and Opus) can reason through attribution conflicts before proposing a budget reallocation.
That said, none of those properties matter if the model can spend money unilaterally. SpendSignoff's safety contract is client-agnostic — the same server-side scope restriction applies whether you are using Claude Desktop, Claude Code, Cursor, or ChatGPT.
Setup: Claude Desktop
In Claude Desktop (macOS or Windows), open Settings → Integrations → MCP Servers → Add. Enter your SpendSignoff server URL and complete the OAuth handshake. Claude will list the discovered tools in the sidebar. You are now ready to query your account.
Claude Desktop MCP config (claude_desktop_config.json)
{
"mcpServers": {
"spendsignoff": {
"url": "https://mcp.spendsignoff.com/mcp",
"transport": "streamable-http"
}
}
}Setup: Claude Code (CLI)
Claude Code picks up MCP servers from the same config file. Add the spendsignoff entry shown above to ~/.claude/claude_desktop_config.json, restart Claude Code, and run /mcp to verify the server handshake. You can then issue campaign commands directly from your terminal alongside your normal coding workflow.
This is particularly useful for performance marketers who also manage tracking code or conversion tag deployments — you can update a Google Ads campaign and push a related GTM change in the same Claude Code session.
Use Claude Code for attribution debugging
Making your first campaign change
After connecting, try this: "Show me campaigns where CPA is above $80 and ROAS is below 2, then draft a 20% budget reduction on each." Claude will call get_account_snapshot, filter to your criteria, and call propose_change for each qualifying campaign. Each draft lands in the SpendSignoff approval queue as a separate before→after diff. You review all of them at once and approve or reject individually.
The entire interaction — tool calls, model reasoning, proposed payloads — is logged in the append-only audit trail with Claude's session ID and your user ID.
FAQ
- Does this work with Claude.ai (the web app)?
- Claude.ai added MCP support via Integrations in mid-2026. The setup is the same — paste your SpendSignoff server URL and complete OAuth. The safety contract is identical regardless of which Claude surface you use.
- Can Claude propose cross-platform changes (Google Ads + Meta in one prompt)?
- Yes. If you have both platforms connected in SpendSignoff, Claude can read from both and propose changes to both in a single conversation. Each platform's changes are separate drafts in the approval queue, so you can approve Google and reject Meta independently.
- What model version does SpendSignoff recommend for ad ops?
- Claude Sonnet 3.7 or Opus 3 for anything involving budget decisions or cross-platform attribution. Haiku works fine for read-only queries and report generation where speed matters more than reasoning depth.
Connect an account read-only and watch the operator work.
Reads are free on every plan. Nothing spends without your two-step approval.
Related reading