Ad-Channel MCP Servers Compared: Google, Meta, Microsoft, TikTok

In a six-week window between April and May 2026, every major ad platform either released or formally backed an MCP server. The shape of the ad-platform tooling landscape now is clearer than it’s been in years, and the differences between the four big offerings matter for how you build agent workflows on top of them.

This is the side-by-side comparison, plus a recommendation for where each platform’s MCP fits in a shopping-feed operation.

Quick comparison table

Per-platform deep dive

Google Ads MCP

The deliberately narrow one. Two tools, list_accessible_customers and search (which runs arbitrary GAQL queries), and that’s the whole API surface. No mutations, no catalog write, no campaign management.

This isn’t a gap, it’s a deliberate scope choice. Google’s view is that ad mutations are too high-stakes for autonomous agents in 2026; reads are safer to expose first. The MCP gets you live performance data into a Claude conversation. The writes still happen via the Google Ads UI, scripts, or the REST API.

Setup friction: Medium. Self-hosted (Python), requires developer token application (instant for test, 1-2 weeks for production), OAuth setup. About 30 minutes from clone to first query.

Best for: performance analysis, search-term mining, disapproval reporting, GAQL-driven custom reports inside agent conversations. Pair with a feed manager for the write side.

Meta Ads MCP

The most ambitious official release. 29 tools, hosted at mcp.facebook.com/ads/{business_id}, full read+write across catalogs, product sets, products, feeds, campaigns, ad sets, ads, audiences, creatives, and reporting.

The catalog cluster is the standout, 10 of 29 tools manage Advantage+ Shopping catalog operations, which is exactly the work most likely to bottleneck on manual review. The March 2026 Shoptalk announcement (Product Set Optimization with SKU-level budget control) effectively requires this kind of programmatic catalog management at scale.

Setup friction: Low. Meta hosts the server; you authorize via Business Suite and copy the personalised endpoint. About 10 minutes end-to-end.

Best for: any team running Advantage+ Shopping or DPA at scale. Catalog operations move from Commerce Manager to agent-driven conversations.

TikTok Ads MCP

Officially supported via the Marketing API, with the most-used implementation being a community server (AdsMCP/tiktok-ads-mcp-server). Full read+write, with particular emphasis on DSA (Dynamic Showcase Ads) catalog management and creative iteration.

The DSA cluster is smaller than Meta’s catalog cluster (around 6 tools) but covers the operationally important calls: catalog registration, product set CRUD, feed URL ingestion, dynamic ad creation. Combined with the campaign-side write tools, an agent can launch a complete DSA campaign in one conversation.

Setup friction: Medium. Community server install, TikTok for Business app registration, access-token rotation. About 20 minutes for first read query.

Best for: TikTok DSA operators, brands testing AI-led campaign launch, agencies running multiple TikTok ad accounts. Less useful if you’re only doing organic TikTok work.

Microsoft Ads MCP

The asterisk in the comparison. No official Microsoft Advertising MCP server as of May 2026. Several community implementations exist, duartemartins/microsoft-ads-mcp-server is the cleanest, Synter MCP wraps Microsoft alongside 13 other platforms commercially, CData publishes a read-only JDBC-backed server.

Quality varies wildly. The duartemartins server is well-maintained, the Pipedream one is hosted but covers a subset of the API, and several others are stale.

Setup friction: High. Community server selection, Azure AD app registration, refresh-token rotation handling, no official documentation path.

Best for: teams that genuinely need Microsoft Advertising coverage in their agent workflows. Most others can wait for the official release.

Catalog tooling: where the platforms differ most

For shopping-feed teams the catalog question dominates. Here’s the breakdown:

Meta treats the catalog as a first-class object in the MCP. Product Sets, feeds, products, and feed runs all have CRUD coverage. The catalog cluster is heavier than the campaigns cluster, which signals where Meta thinks the bottleneck is.

TikTok covers DSA catalogs with fewer tools but enough to manage the lifecycle. Catalog registration, product set CRUD, URL-based feed ingestion, dynamic ad creation. Less granular than Meta but functionally complete for DSA workflows.

Google doesn’t expose Merchant Center via the Google Ads MCP. Merchant Center is a separate Google product with its own (non-MCP) API. If you want agent access to Google Shopping catalog data, you need either a feed manager that exposes its own MCP (AI Shopping Feeds does, several others are catching up) or direct Merchant API calls.

Microsoft community servers are inconsistent on Microsoft Merchant Center coverage. Most cover Search/Audience campaigns thoroughly and treat catalog/shopping as an afterthought.

The implication: if catalog operations are your bottleneck (and for most shopping-feed teams they are), Meta’s MCP is the most directly useful. Google requires a separate feed manager. TikTok DSA is covered. Microsoft needs the feed manager pattern too.

Authentication models compared

Meta’s hosted model is the easiest to maintain, no token rotation in your config files. Google’s standard OAuth is well-documented. TikTok requires manual rotation. Microsoft’s rotating-refresh-token quirk catches most naive implementations.

Rate limits and operational ceiling

The numbers below are rough; check current docs for exact limits per your account tier:

• Google Ads, basic-access tokens: ~15,000 operations/day. Standard: ~200,000/day. Search API has its own quota separate from mutations.
• Meta Marketing API, varies by spend tier. Low-spend accounts cap around 200 calls/hour for reads; high-spend accounts have much higher ceilings. Write operations more restrictive than reads.
• TikTok Marketing API, ~10,000 reads/day per advertiser, much lower write tiers (creative_upload caps at ~500/day).
• Microsoft Advertising, 5 requests/second per app by default; reporting API uses async download model with separate quota.

For agent workloads, the operative constraint is usually concurrent requests rather than daily quota. A single user with multiple parallel agents can trip per-app concurrency limits faster than they’ll exhaust daily quota.

Multi-platform agent setup

The pattern that gives the most leverage: register all four MCPs in your Claude Desktop config so a single conversation can read and write across platforms.

{
  "mcpServers": {
    "google-ads": { "..." },
    "meta-ads": { "..." },
    "tiktok-ads": { "..." },
    "microsoft-ads": { "..." },
    "ai-shopping-feeds": { "..." }
  }
}

The fifth entry, your feed manager, is what makes multi-platform actually useful. Without it, your agent can read performance across four platforms but can’t change the underlying catalog that’s serving on all of them. With it, the agent can identify a bottom-quartile SKU on Google, see it’s also bottom-quartile on Meta and TikTok, and queue a title rewrite that affects all three platforms on the next feed refresh.

Where to start

If you’re a shopping-feed team picking your first MCP, the recommended order is:

1. Meta Ads MCP, highest catalog-ops payoff, easiest setup, official+hosted. Particularly important if you run Advantage+ Shopping.
2. Google Ads MCP, strong read-side reporting, useful for performance analysis. Pair with a feed manager (write side) for full loop.
3. TikTok Ads MCP, for teams running TikTok DSA seriously. Skip if TikTok is a test-budget channel.
4. Microsoft Ads MCP, for teams with material Bing Shopping spend. Otherwise wait for the official release.

The cost-to-value ratio strongly favours getting Meta and Google running in week one, then layering TikTok in if relevant. Microsoft is a wait-and-see.

What’s next

Three near-term expansions to watch:

• Google Ads mutations, likely to land within 2-3 releases. Campaign-level writes will be the first tier.
• Microsoft Advertising official MCP, heavily expected but no public timeline. Microsoft’s track record on MCP elsewhere (Azure, Copilot, GitHub) suggests it’s coming.
• Amazon Ads MCP, Amazon has been the conspicuous absence from the May 2026 batch. The Amazon Advertising API supports MCP-style integration already; an official server would be a major shift.

For now, the four-platform agent setup (with a feed manager for the catalog write layer) is the practical ceiling for shopping-feed operations in 2026. It’s also a significant step-change from where the same teams were running 12 months ago.

Sources

• Model Context Protocol specification
• Google Ads MCP, developer documentation
• Meta AI Connectors, official product page
• TikTok Marketing API, documentation
• Microsoft Advertising API, reference
• Related read: Google Ads MCP setup guide · Meta Ads MCP for catalog operations