Claude can analyze live Google Ads and Meta Ads data when an MCP server is connected and the identity has workspace membership with enabled platform Read permission. The practical route described here uses Adstudio's hosted MCP connection: complete OAuth, select workspace, platform, and account on each call, then ask Claude to discover and query them. The public connection is read-only today. One personal connection can reach every workspace its user belongs to, subject to role-bounded per-platform Read permission; Claude can explain performance and compare selected channels, but it cannot pause campaigns or change budgets through these tools.
MCP is a tool-connection protocol, not a promise that every server has the same capabilities. The server controls its tool names, account boundaries, fields, and write permissions. Adstudio's current public surface covers Google Ads and Meta Ads, with bounded reads and provenance. The Claude MCP setup page and MCP hub are the sources for the current endpoint and setup details.
You should know the client identity you will use, the workspaces where that identity is a member, the relevant platform Read permissions, and who can revoke access. Keep the first test to one account and a complete, closed date range. This makes it easier to distinguish an authorization issue from a reporting discrepancy.
Start with this prompt:
“Use account discovery first. List the active workspace integrations reachable through my current workspace memberships and enabled per-platform Read permission. Include platform, account ID, account name, currency if returned, and access limitations. Ask me to choose an account; do not guess.”
After choosing an account:
“For [platform/account ID], fetch campaign-level spend, clicks, conversions, and CPA for the last complete calendar week. Show source, account ID, exact dates, currency, and any attribution or freshness notes before giving an interpretation.”
Claude's conversational follow-ups are useful once the account and period are pinned. Keep asking it to retain those identifiers when moving from campaign to ad set or ad detail.
For an anomaly review:
“Compare the selected Google Ads account's last 7 complete days with the previous 7. Identify campaigns with a meaningful change in spend or conversion rate. Show the daily evidence, then list hypotheses separately from facts.”
For Meta creative analysis:
“Within the selected Meta Ads account, rank ads by purchase value and by CPA for the last complete month. Preserve campaign, ad set, ad ID, currency, and the source fields. Explain where low volume makes the ranking unreliable.”
For a cross-channel weekly review:
“Query the selected Google Ads and Meta Ads accounts separately for the same dates. Return one table per platform with account and currency labels. Identify directional patterns, but do not merge conversion counts or CPA when attribution definitions differ.”
For an audit handoff:
“Write a concise findings log. For each finding include platform, account ID, date range, metric fields, direct values, calculations, and a next verification step. Mark unknowns as unknown.”
The best result is not the most confident paragraph. It is a conclusion that a colleague can trace back to a platform, account, period, and returned field.
Claude shows no tools. Verify the server was added to the intended Claude environment and that the OAuth callback completed. If your organization restricts remote MCP servers, an administrator may need to allow the connection.
Discovery returns the wrong scope. Stop querying and verify the signed-in identity's workspace membership, role, and enabled platform Read permission. Each subsequent call must explicitly select the intended workspace, platform, and account; a client login does not select an account automatically.
A query fails or returns no rows. Ask Claude to repeat the exact account ID, dates, fields, and server error. Reduce the date range and remove an optional breakdown. Never convert an unavailable result into a zero.
Claude's totals differ from the platform UI. Check timezone, date inclusivity, attribution window, currency, conversion action, and data freshness. Ask for source rows and calculations; then validate in Ads Manager.
Claude suggests editing a campaign. The public Adstudio MCP has no write tools. Copy the recommendation into your normal review process; the MCP connection itself cannot execute it. For the security tradeoff, see read-only versus write MCP.
Adstudio's public MCP currently reads Google Ads and Meta Ads only. It is not evidence that TikTok Ads, OpenAI Ads, DV360, billing, policy appeals, creative uploads, or campaign mutation are supported. First-party platform servers may expose a different, vertical surface. Google's documented official Google Ads MCP server is read-only and focused on Google discovery, querying, and metadata; use the comparison guide when deciding between it and a cross-channel connector.
For a Meta-specific evaluation, read the Meta Ads MCP guide. For the conceptual overview, see what an ad-platform MCP server is.
Claude plus a carefully scoped MCP connection is a useful read-only analyst for selected Google Ads and Meta Ads accounts. Explicitly select the account, preserve provenance, validate the first result, and treat recommendations as review material. The connection gives Claude evidence; it does not give Claude permission to spend money.