Best ad MCP servers in 2026: what to know before you connect

The three tiers of ad MCP tooling

Tier 1 — Read-only data connectors. These issue only read scopes, typically to the ad platform's reporting API. Use cases: pulling campaign metrics into a conversation, answering questions about account performance, generating reports. Google's official MCP server (released mid-2025) is in this tier. Zero spend risk; also zero ability to change anything.

Tier 2 — Write-capable connectors without an approval gate. These issue write scopes to the ad platform API and allow the AI client to apply changes directly. Convenience is high; spend risk is also high. If the model misinterprets an instruction or a prompt injection occurs, changes go live without a human review step.

Tier 3 — Write-capable connectors with an enforced approval gate. Write scopes are available but the server issues only mcp.draft to the AI client. The actual apply-to-platform call requires a separate authorization step that only a human can perform. SpendSignoff is in this tier.

Google's official MCP server

Google released a read-only MCP server for Google Ads as part of the Ads API developer tools in 2025. It connects to the Google Ads Query Language (GAQL) endpoint and lets an AI client run reporting queries in natural language.

What it does well: Accurate reporting queries, fully official, no third-party data custody.

What it does not do: Any writes. No campaign creation, no bid changes, no budget modifications. Also no Meta/cross-platform support. If you need read-only Google Ads data access via an AI client and are not ready to connect a third-party tool, this is a safe starting point.

We wrote a dedicated post on this: why-not-google-free-mcp-server covers exactly when it is and is not sufficient.

Open-source GitHub connectors

Dozens of open-source MCP connectors for Google Ads, Meta, and other platforms appeared on GitHub in late 2024 and 2025. Quality varies widely. Some are careful about scope minimization; others request full account admin permissions for simplicity.

If you are evaluating an open-source connector, check three things: the OAuth scope list in the setup instructions, whether there is any approval gate for write operations, and when the repository was last actively maintained. Stale connectors against frequently-updated ad platform APIs produce errors or silently drop fields.

Where SpendSignoff sits in the comparison

SpendSignoff is a Tier 3 commercial MCP server. Write operations go through a draft-and-approve kernel enforced in the policy core — not in a prompt. The model gets mcp.read and mcp.draft; the apply-to-platform call requires a human two-step confirm in the dashboard.

Beyond the safety model, the differentiating features are: the always-on autonomy loop (propose-only V1) that monitors accounts and queues optimization drafts, the KMS-signed audit log, multi-tenant agency support, and the normalized cross-platform data layer covering Google Ads and Meta V1 (LinkedIn and TikTok in active development).

Pricing is competitive versus both agency retainers and comparable commercial tools: Free (1 account, 15 calls), Solo ($485/year, 1,800 calls), Pro ($999/year, 7,200 calls), Max ($2,000/year, 50,000 calls). These are per-organization prices, not per-seat.

How to evaluate any ad MCP tool

• Scope request — What OAuth permissions does it request? Admin/write access to an ad account is broad. Prefer tools that scope minimally.
• Approval gate — Is spend-moving authorization enforced server-side or by prompt? Prompt-only enforcement can be overridden by a sufficiently adversarial prompt.
• Audit trail — Does the tool log every AI action against the account? Essential for any account where multiple people or systems can make changes.
• Platform coverage — Does it support the platforms you actually run? Multi-platform normalized reporting matters if you run both Google and Meta.
• Maintenance signal — Is the connector actively maintained against current API versions? Ad platform APIs change frequently.