Ads MCP servers: Google, Facebook and LinkedIn compared
Compare concrete Ads MCP implementations, account prerequisites and read-versus-write risks before connecting advertising data to an AI assistant.
Sources reviewed October 23, 2026. Documentation review, not a live account test or product endorsement.
An Ads MCP server connects an MCP-compatible assistant to advertising APIs. It is an interface, not an advertising strategy, a guarantee of API access, or permission to spend. Choose the implementation and its exposed tools—not just the platform name.
Three implementations, three different access decisions
These are examples with inspectable primary documentation, not an exhaustive market ranking. “Facebook Ads” below includes Instagram advertising through Meta.
| Platform and implementation | Documented work | Setup boundary | Write boundary |
|---|---|---|---|
| Google Ads MCP by googleads | Account discovery, GAQL search and resource metadata | Google Cloud project, credentials and access to the advertiser account | Google’s current integration guide describes the release as read-only |
| Meta Ads MCP by Pipeboard | Reporting, campaign/ad-set/ad operations and creative uploads | Hosted connection through Pipeboard; self-hosting instead requires your own Meta developer app | The maintainer documents confirmation for writes; verify it in your chosen client before allowing changes |
| LinkedIn Ads MCP by dhawalshah | Performance queries, campaign management and audience analysis | A Marketing Developer Platform-approved app; local stdio or a remotely operated service | Treat campaign management as write-capable; only expose needed tools and permissions |
The Google implementation is published in Google’s ads organization. Pipeboard and dhawalshah are separate third-party implementations: their availability does not mean Meta or LinkedIn endorses them. See Google’s integration guide, the Pipeboard tool reference, and the LinkedIn project’s prerequisites. No prices are compared here: hosted subscriptions, API access and model usage are separate costs to verify with their providers.
Which approach fits your job?
- Explain a Google Ads performance change: start with read-only reporting, not a campaign-creation integration. Follow the Google Ads MCP setup guide.
- Prepare Facebook or Instagram campaign changes: inspect the exact tools, destination account and approval mechanism before connecting a write-capable server. A draft response from an assistant is not proof that an API write has not occurred.
- Analyze LinkedIn campaign performance: establish app approval and account access first. A social-posting integration is not equivalent to Marketing API access.
- Export repeatable reports from a terminal: consider a Google Ads CLI rather than adding an MCP host to a deterministic task.
This is our selection guidance, not a claim of measured speed or reliability. We have not executed these connectors against a live advertising account.
A reporting-first acceptance checklist
- Record the advertiser ID, reporting timezone, currency, date window and attribution settings.
- Confirm the connected user can access only the intended accounts. For hosted services, review where tokens and report data are stored.
- Inspect the discovered tool list. Remove write tools if your job is reporting; a prompt saying “do not change anything” is not an authorization boundary.
- Run one small report and compare its totals with the platform UI using the same filters.
- Require a reviewable proposed diff before a budget, bid, status, targeting or creative change. Preserve the request and resulting resource IDs in an audit trail.
- Re-check scopes and tool behavior after upgrades; do not assume all servers sharing an “Ads MCP” label have the same safety model.
Cross-platform numbers need context
Do not blindly add attributed conversions from Google, Meta and LinkedIn: the same purchase may be counted by more than one platform. Normalize currency and reporting windows, and distinguish platform-attributed outcomes from independently measured revenue. When asking an assistant for recommendations, require the query, filters and limitations alongside the answer.
Where AICMOHQ fits
These external servers are not AICMOHQ’s MCP server. AICMOHQ’s MCP overview describes its own marketing workflow; connecting it does not install the third-party servers above or grant their advertising capabilities. Review the agent fleet and integration documentation separately before choosing an execution path.
Questions to ask before connecting
Can any MCP client use every server?
No. Check transport and authentication compatibility in your specific host. Local stdio and remote HTTP configurations are different deployment choices; a sample configuration for one is not interchangeable with the other.
Is read-only enough to protect confidential data?
No. A read-only server can still send advertising data to the connected model or service. Limit account access and report scope, review retention terms, and avoid exporting customer-level data when aggregate reporting suffices.
Does creating a paused campaign make a connector safe?
It reduces one activation risk, but does not replace approval, account checks or an audit trail. Separate creation, activation and later budget changes in your review process.