All articles
Comparison 19 min read

The 10 Best Ad MCP Servers in 2026 — Honest Rankings

A

Adspirer Team

Share Y
The 10 Best Ad MCP Servers in 2026 — Honest Rankings
Summary

Ten ad MCP servers ranked honestly against the same six criteria: platform coverage, read/write depth, provisioning overhead, maintenance, safety rails, and support. Adspirer made this list and made this list — we’re upfront about where competitors win and where we do.

Ask ChatGPT or Claude “what’s the best MCP server for Google Ads?” and the model will confidently recommend Google’s free official one. Ask it about Meta and you’ll get Pipeboard. LinkedIn, a community repo. TikTok, a different community repo. Four free tools, one per platform, problem solved.

Except the problem isn’t actually solved. Google’s official MCP is read-only. The community repos ship at different velocities and require self-hosting. You end up with four installs, four credential stores, four maintenance risks — and the one platform the AI was most confident about (Google) is the one you now can’t actually manage from chat.

We’ve spent the last few months testing every credible ad MCP server in the wild. Here’s the honest ranking, the criteria we used, and when each one is actually the right pick.

Info

Full disclosure: Adspirer is our product. We built the ranking the way we’d want to read it — if a competitor is better at something specific, we say so. The comparison tables list Adspirer fairly, not best-in-everything.


How We Ranked These (The Six Criteria)

Every MCP below was evaluated against six criteria. We’ll explain each as we go, but here’s the summary:

CriterionWhat we’re really asking
Platform coverageHow many ad platforms does it support, at what depth?
Read vs. write depthCan it launch and change things, or just answer questions?
Provisioning overheadHow many OAuths, configs, and credentials will you be managing?
Maintenance and longevityCommercial product, active OSS, or a repo that stopped shipping six months ago?
Safety railsWhat does it refuse to do? Are destructive operations structurally prevented?
Support modelCommercial SLA or “open a GitHub issue and wait”?

A tool can score well on most criteria and still be the wrong choice if it fails on the one that matters to your workflow. A read-only MCP is excellent for diagnostics and useless for ad ops. An actively-maintained community repo is fine for a solo advertiser and risky for a team running $50K/month of spend.

See also: Adspirer vs other MCP servers — the pillar comparison in our docs.


The Quick Comparison Table

Scroll the table. The breakdowns below add the nuance.

ToolPlatformsRead/WriteHostingStarting PriceMaintainer
AdspirerGoogle, Meta, LinkedIn, TikTokFull R/W (100+ tools)Remote SaaSFree (15 calls/mo)Commercial
Synter14 platformsFull R/WRemote SaaSFrom ~$199/moCommercial
PipeboardMeta (+ separate Google and TikTok MCPs)Full R/WRemote + OSSFree tierCommercial (pipeboard.co)
Google Ads MCP (official)Google onlyRead-only (GAQL)Self-hostFreeGoogle
Ryze / irinabuht12 MCPGoogle, Meta, GA4Full R/WSelf-hostFree OSS + commercial RyzeSolo + Ryze AI
Flyweel MCPGoogle, MetaRead-onlyRemoteFree with FlyweelCommercial
AdsMCP TikTokTikTok onlyFull R/WSelf-hostFreeCommunity
DanielPopaMD LinkedInLinkedIn onlyFull R/W (25+ tools)Self-hostFreeSolo community
GoMarble FacebookMeta onlyR/W (basic)Self-hostFree OSSCommunity
Windsor.ai MCP325+ data sourcesRead-focused (BI)RemoteWindsor subscriptionCommercial

Now the per-MCP breakdowns. In order.


1. Adspirer — Unified, Write-Enabled, Safety-Railed

What it actually is

Adspirer is a remote MCP server that unifies Google Ads, Meta Ads, LinkedIn Ads, and TikTok Ads behind a single authentication and a single configuration in your AI client. 100+ tools across the four platforms — campaign creation, budget adjustments, keyword research, performance analysis, cross-platform reporting, wasted-spend detection, and more.

It works with ChatGPT, Claude, Claude Code, Claude Cowork, Cursor, Codex, Windsurf, OpenClaw, Gemini CLI, Manus, Perplexity, and Custom GPTs. That’s the broadest AI-client coverage on the market, and it matters — most community MCPs document Claude Desktop and stop there.

Who it’s for

Marketers, founders, and agencies running two or more ad platforms who want to actually manage campaigns from chat, not just query them. Teams whose AI client isn’t Claude Desktop. Anyone who’d like destructive operations to be structurally impossible rather than prompt-guarded.

What we like

  • Genuine writes, not just reads. Launch Google Search and PMax campaigns, Meta image and video campaigns, LinkedIn Sponsored Content, TikTok video campaigns — all from chat, with structural safety rails. See the capability matrix.
  • One OAuth per platform, one config, one maintenance surface. Compare to four separate tools if you run four platforms.
  • Safety rails are architectural. Cannot delete existing campaigns, cannot pause running campaigns, cannot modify existing budgets. New campaigns created paused. This isn’t a prompt — it’s that the destructive tools don’t exist in the server.
  • First-class agent skills. Claude Code agent, Cursor agent, Codex agent, OpenClaw agent are purpose-built, not generic MCP tool use.
  • Multi-account and multi-client routing for agency workflows. See multi-account.

What we don’t like

  • Only four platforms. If your stack needs Reddit, Pinterest, X, Amazon DSP, or Microsoft Ads, Synter is the better comparison.
  • Not fully open-source. Free tier is real (15 calls/month forever), but there’s no self-host-yourself path for hobbyists. Enterprise private deployment exists via enterprise plans.
  • Newer than some competitors. Pipeboard’s Meta MCP has more public GitHub stars than our entire category has existed.

The verdict

If you’re running two or more platforms and want a unified surface with real writes and real safety rails, Adspirer is the pick. If you’re a single-platform advertiser with read-only needs, skip to the platform-specific entries below.

Tip

Start free. 15 tool calls/month, no credit card, all four platforms included. Enough to connect accounts, pull performance, and create a couple of campaigns. Try Adspirer free →


2. Synter — The Broadest-Coverage Unified MCP

What it actually is

Synter is a commercial unified MCP that covers 14 platforms: Google, Meta, LinkedIn, TikTok, Reddit, Pinterest, Snapchat, X, Microsoft, Taboola, and more. Direct OAuth connections to each platform’s official API. Full read and write. Emerged from stealth in 2026; pricing starts around $199/month.

Who it’s for

Advertisers and agencies whose stack genuinely extends beyond the big four. Reddit Ads and Pinterest are the classic tiebreakers — if those are a meaningful portion of spend, Synter is the most direct path.

What we like

  • Platform list nobody else matches. Genuinely 14, not 4.
  • Full read + write across all of them. This isn’t a reporting tool.
  • Unified budget control and conversion tracking management baked in.
  • AI-native stealth pedigree — the team came out of stealth with a production product and known customers.

What we don’t like

  • Newer than we’d like for a tool managing production spend at scale. No reason not to pilot; some reason to weight.
  • Claude Desktop-centric documentation. Works with other AI clients, but the setup story outside Claude Desktop / Cursor is thinner than Adspirer’s.
  • Depth on individual platforms harder to verify than on specialist tools. 14 platforms is a wide surface.

The verdict

If your stack includes Reddit, Pinterest, X, or Microsoft Ads and those channels really matter, Synter is the right comparison. If you’re on Google + Meta + LinkedIn + TikTok, Adspirer is the closer match.


3. Pipeboard — Mature Meta MCP, Still Split per Platform

What it actually is

Pipeboard’s Meta Ads MCP is the most mature single-platform Meta MCP in the ecosystem. 791 GitHub stars, 133 releases, 489 commits on the main branch as of April 2026. Licensed under Business Source License 1.1 (converting to Apache 2.0 in 2029). Full read + write: campaign creation, ad sets, ads, creative uploads, targeting, budget schedules, bid modifications. Campaigns are paused by default.

Pipeboard also ships separate MCPs for Google Ads and TikTok Ads as different remote services. Not LinkedIn.

Who it’s for

Meta-heavy advertisers who want the most actively-maintained Meta-specific MCP in the ecosystem. Teams willing to install and manage three separate Pipeboard MCPs (Meta + Google + TikTok) for best-in-class depth on each.

What we like

  • Release velocity is legitimately best-in-class for a single-platform MCP. 133 releases is a lot.
  • Paused-by-default safety on campaign creation.
  • Legitimate company with a roadmap and a team, not a side project.
  • BSL license gives you a clear OSS path in 2029.

What we don’t like

  • No LinkedIn. For B2B advertisers, hard stop.
  • Three separate MCPs if you want Meta + Google + TikTok. Operationally, that’s three installs, three OAuth flows, three maintenance surfaces, three points where the LLM has to context-switch between tool namespaces.
  • Cross-platform reasoning is awkward. Asking “what’s my blended CPA across Meta and Google this week?” means the LLM is reconciling two tool namespaces on two servers.

Head-to-head: Pipeboard vs. Adspirer on Meta

This comparison deserves its own post, and we wrote one: Adspirer vs Pipeboard: Meta Ads Head-to-Head. The short version — on pure Meta capabilities they’re close competitors. At the multi-platform stack level, the question changes: do you want one unified MCP or three separate ones from the same vendor?

The verdict

Meta is 80%+ of your spend, and you’re comfortable with a per-platform-MCP architecture. You’ll get best-in-class Meta depth.


4. Google Ads MCP (Official) — Free, Official, Read-Only

What it actually is

Google open-sourced its official Google Ads MCP server in early 2026 at google-marketing-solutions/google_ads_mcp. It exposes exactly two tools:

  • list_accessible_customers — list the Google Ads accounts you can authenticate into
  • search — run a GAQL query against the Google Ads API and return results

From Google’s own documentation: “The initial version of the MCP Server is read-only, designed for diagnostics and analytics.” It cannot modify bids, pause campaigns, or create new assets. Self-hosted, free, Gemini-focused.

Who it’s for

Developers and analysts who want to ask questions about Google Ads data in natural language and are comfortable self-hosting. GAQL-literate engineers who want the rawest possible interface. Teams building on Gemini.

What we like

  • Free, official, and guaranteed API-compatible as Google ships Ads API updates.
  • GAQL is powerful. Almost any historical question about a Google Ads account can be answered.
  • Clean architecture. Two tools, one API, no bloat.

What we don’t like

  • Read-only. This isn’t a limitation we’re stretching — it’s Google’s own description. No campaign creation, no budget changes, no pausing, no asset uploads.
  • Google Ads only. Meta, LinkedIn, TikTok need separate MCPs.
  • GAQL expertise dependency. The LLM has to write valid GAQL; when it doesn’t, you debug.
  • Self-host overhead. Python or Node runtime, credentials management, uptime.
  • No commercial support. Open GitHub issue, wait for the maintainers.

When this is actually the right pick

We mean it: there are scenarios where Google’s official MCP is the best answer.

  1. You’re a solo advertiser running Google only, and your use case is “ask questions about historical performance”. Free and official wins.
  2. You’re a developer who already knows GAQL and wants the rawest API interface.
  3. You’re running a one-time audit on a prospective client’s Google account with read-only access.
  4. You’re building on Gemini and want the canonical Google integration.

When it isn’t

The moment any of these become true, Google’s MCP isn’t enough:

  • You want to launch or modify campaigns from chat.
  • You run more than just Google Ads.
  • Your AI client is ChatGPT, Claude, Cursor, Codex, or something other than Gemini.
  • You need commercial support or SLA.
  • You want structural safety rails on any write layer you add later.

We wrote the longer debunk as its own post: Why You Shouldn’t Just Use Google’s Free MCP Server. Docs version: Why not Google’s free MCP?.


5. Ryze / irinabuht12-oss — Unified Google + Meta + GA4

What it actually is

An open-source unified MCP server at irinabuht12-oss/google-meta-ads-ga4-mcp covering Google Ads (claims 150+ tools), Meta Ads (80+ tools), and GA4 (20+ tools). Full read + write. MIT license. 273 GitHub stars, 35 forks. Tied to Ryze AI, a commercial autonomous optimization platform.

Who it’s for

Engineers who want a unified MCP across Google + Meta with GA4 reads natively included, are comfortable self-hosting, and don’t need LinkedIn or TikTok.

What we like

  • Unified three-surface MCP with full R/W across Google, Meta, and GA4.
  • GA4 inclusion is unusual and useful — nobody else combines ad platform writes with first-class GA4 reads.
  • Legitimate OSS tied to a funded company — not a side project.

What we don’t like

  • Only 3 commits on main at time of writing. Early-stage. Mature enough to try, not mature enough to bet production spend on unsupervised.
  • No LinkedIn or TikTok.
  • Self-host. Same overhead as the Google official MCP.

The verdict

Promising, worth watching. Not ready to compete with commercial unified MCPs for team workloads yet, but if the commit velocity picks up, that could change quickly.


6. Flyweel MCP — Read-Only, Well-Packaged

What it actually is

A remote MCP from Flyweel that connects your Google Ads and Meta Ads to Claude, Cursor, or VS Code in about 30 seconds. Included free with a Flyweel account; can also be used standalone. Read-only.

Who it’s for

Existing Flyweel customers who want chat-native reporting over the data they’ve already unified in Flyweel.

What we like

  • Fast setup. Drop a config, start asking questions.
  • Honest positioning. Their own docs say “never posting or changing anything.” We appreciate the clarity.
  • Solid execution on the chat-over-read-data workflow.

What we don’t like

  • Read-only. Dashboard-in-chat, not an ad-ops tool.
  • Google + Meta only.

The verdict

Right pick for Flyweel customers who want natural-language reporting over their unified data. Not a campaign management tool.


7. AdsMCP TikTok Ads MCP

What it actually is

An open-source MCP for the TikTok Business API at AdsMCP/tiktok-ads-mcp-server. Campaign management (create, read, update), performance analytics, audience management, creative uploads, custom reports. MIT license. Built-in rate limiting (1000 req/hour) and retry logic.

Who it’s for

TikTok-heavy advertisers who want a dedicated TikTok MCP and are comfortable self-hosting.

What we like

  • Full R/W with real campaign management, not just reads.
  • Built-in rate limiting — unusually solid operational hygiene for a community repo.
  • Clear scope — does one thing, does it well.

What we don’t like

  • TikTok only.
  • Self-host overhead.

The verdict

TikTok is the majority of your spend and you want a single-purpose tool without a subscription.


8. DanielPopaMD LinkedIn Ads MCP

What it actually is

The most complete community LinkedIn Ads MCP — 25+ tools covering accounts, campaigns, creatives, audiences, conversions, and analytics. Read + write. Self-hosted.

Who it’s for

B2B advertisers who need LinkedIn campaign management in chat and don’t care about other platforms.

What we like

  • Deepest community LinkedIn coverage. 25+ tools is more than most other single-platform MCPs expose.
  • Full lifecycle — not just reads.

What we don’t like

  • LinkedIn only.
  • Solo-maintained. Worth monitoring commit velocity.
  • Self-host.

The verdict

LinkedIn is your only platform, you’re a developer, and you want maximum tool depth without a subscription.


9. GoMarble Facebook Ads MCP

What it actually is

Free, open-source local Facebook Ads MCP at gomarble-ai/facebook-ads-mcp-server. Requires Node.js, Meta API credentials, and CLI comfort. Read + basic write.

Who it’s for

Developers who want a locally-run Meta Ads MCP with no vendor dependency and full transparency into the code.

What we like

  • Fully transparent. Read the code before running it.
  • No subscription.

What we don’t like

  • Self-host is real work — Node setup, credentials management, uptime.
  • Meta only.
  • Community-paced maintenance.

The verdict

You’re a developer who values running everything locally and won’t install a remote service. Also worth knowing about as a backup if your primary Meta MCP breaks.


10. Windsor.ai MCP — Different Category

What it actually is

Windsor.ai is a marketing data pipeline with 325+ data connectors syncing ad spend, analytics, CRM, and commerce data into a unified warehouse. Their MCP server lets an LLM query that warehouse in natural language — “what campaigns had the best ROAS last month?” — without writing SQL.

Who it’s for

Teams using Windsor.ai as their marketing data layer who want AI-native reporting over that layer.

What we like

  • Cross-channel BI in chat. If your data is already in Windsor, this unlocks it without SQL.
  • Massive source coverage — 325+ is well beyond any ad-MCP.

What we don’t like

  • Not a campaign management tool. It can’t launch or change things — that’s a different product category.
  • Early-stage MCP repo. 3 stars, 9 commits at time of writing. Fine to pilot if you’re already a Windsor customer; otherwise you’re buying into an early integration.

The verdict

Complementary to ad MCPs, not a substitute. A common stack: Windsor.ai for reporting + Adspirer or Synter for campaign execution. They don’t conflict.


How I’d Actually Choose (Based on Your Situation)

Four decision axes cover almost every buyer. Pick the one that matches.

By platform count

📊

One platform, read-only

Free option wins. Google’s official MCP for Google. Pipeboard free tier for Meta. Community options for LinkedIn or TikTok. No need to pay for unification you don’t need.

✏️

One platform, need writes

Pipeboard (Meta). AdsMCP (TikTok). DanielPopaMD (LinkedIn). Adspirer Plus if you’d rather not self-host.

🔗

Two to four platforms

Adspirer. One MCP covering Google + Meta + LinkedIn + TikTok beats three-to-four separate installs.

🌐

More than four platforms

Synter. 14-platform coverage is the single deciding factor.

By AI client

Most options work with Claude Desktop. Pipeboard, Synter, Adspirer, community MCPs — all documented. Pick by platform coverage and write needs.

Adspirer has documented setups for all four. Most community MCPs only document Claude Desktop. See the AI Clients docs.

Adspirer, OpenClaw’s own skills, and a few community agents. For Claude Code specifically, the Adspirer Claude Code agent is a purpose-built skill, not generic MCP tool use.

Adspirer supports all three. Most alternatives don’t document them. If you’re building on Gemini specifically, Google’s official MCP is also a natural fit — as long as read-only works for your use case.

By what you value most

You care most aboutThe pick
”Free and official”Google’s official MCP (read-only accepted)
“Widest platform breadth”Synter
”Meta depth + mature OSS”Pipeboard
”Unified + writes + safety rails”Adspirer
”Cross-warehouse BI, not ops”Windsor.ai
”Self-host everything”GoMarble, brijr, DanielPopaMD

The realistic stack for a working agency

Most serious ad teams end up with a small number of MCPs configured in their AI clients, because different tools solve different jobs:

Adspirer for campaign execution

Google + Meta + LinkedIn + TikTok writes, multi-client routing, and the surface team members use daily.

A warehouse MCP for BI (Windsor.ai or similar)

Cross-channel questions that span data beyond the ad platforms themselves — CRM, web analytics, e-commerce.

Facebook Ads Library MCP for competitor research

Public ads library queries. Complementary to everything else.

Three MCPs, three clear jobs, no overlap. You don’t need to pick one winner.


Safety: The Thing Nobody Talks About

A rough taxonomy of how different MCPs handle destructive operations:

Model 1: Prompt-only safety

The MCP exposes destructive tools. A system prompt tells the LLM not to use them casually. Most community MCPs sit here, often implicitly.

This is fragile. Prompt injection, a misread request, or a creative LLM interpretation is all it takes for “delete everything that didn’t convert” to actually delete things. We’ve seen it happen.

Model 2: Confirmation-required safety

The tool runs but asks the user to confirm before committing. Better. Works as long as the user reads every confirmation carefully — which, in a chat-driven workflow with fifty tool calls per hour, they don’t.

Model 3: Structural safety

The destructive tools simply don’t exist in the MCP server. An LLM cannot call what isn’t there.

Adspirer uses Model 3 for the worst classes of action. We cannot delete existing campaigns, cannot pause running campaigns, and cannot modify existing budgets. The tools for those operations aren’t in the server at all.

This isn’t a marketing claim; it’s an architectural one. You can verify it by inspecting the tool list your AI client sees when connected to Adspirer. See the capability docs for the explicit list.

The broader security picture

A research report from Astrix Security analyzed 68 public MCP packages and found:

  • 88% require credentials
  • 53% rely on long-lived static API keys or Personal Access Tokens
  • Only 8.5% use OAuth

Most community ad MCPs ask you to paste a write-scoped access token into a config file on your laptop. That’s a different security posture than OAuth 2.1 with PKCE, which is what Adspirer uses. Full detail in our security docs.


The MCP Architecture Advantage (Why This Is Different From “Plugins”)

One thing most comparison articles miss: MCP-native tools benefit automatically from improvements in the LLMs themselves.

When OpenAI ships GPT-5 or Anthropic ships the next Claude, an MCP server gets smarter without a software update. Better reasoning means better GAQL generation. Better writing means better ad copy suggestions. Better tool use means more reliable multi-step workflows.

Traditional PPC tools — even good ones — freeze their AI capabilities at whatever model they integrated internally. They have to manually swap model versions, which takes months.

Info

This doesn’t make older dashboard-based tools obsolete. Optmyzr’s rule engine is battle-tested and excellent at what it does. But the trajectory over the next 2-3 years heavily favors MCP-native tools for the specific job of “manage ads through conversation.”

For more on the protocol itself, we wrote What Is MCP? A Marketer’s Guide.


FAQ

Is Google's free MCP actually usable for running ads?

Not for managing campaigns — it’s read-only. It’s excellent for asking questions about historical performance via GAQL. The moment you need to launch, pause, adjust a budget, or change a bid, you need a different tool. See the longer treatment: Why You Shouldn’t Just Use Google’s Free MCP Server.

Can I mix MCPs — for example, Adspirer for writes and Windsor.ai for reads?

Yes. MCP clients support multiple servers simultaneously. Windsor.ai for cross-warehouse BI + Adspirer for campaign execution is a reasonable stack for agencies. They don’t conflict.

What happens when Meta ships a breaking Graph API change?

Commercial MCPs (Adspirer, Synter, Pipeboard, Flyweel, Ryze) absorb those changes server-side — your config doesn’t need updating. Self-hosted community MCPs need a maintainer to push a fix, then you pull the update. Repos with low commit velocity are where this gets risky.

Why is Pipeboard a separate entry from Pipeboard Google and Pipeboard TikTok even though they're the same company?

Because they ship as three separate remote services with different URLs, different configs, and different auth. Installing all three means managing three MCPs — not one unified tool. That’s different from Synter or Adspirer, which expose all platforms through a single endpoint.

Does Adspirer work with ChatGPT, or just Claude?

Both, plus ten more. ChatGPT, Claude Desktop/Web/Code/Cowork, Cursor, Codex, Windsurf, OpenClaw, Gemini CLI, Manus, Perplexity, and Custom GPTs. Setup guides per client.

What's the minimum I can spend to test this approach?

Zero. Adspirer’s free tier is 15 tool calls per month, no credit card required — enough to connect accounts, pull performance, and create two or three campaigns. Several community MCPs are also free if you’re willing to self-host. See pricing.

Is it safe to give an AI assistant write access to my ad accounts?

The honest answer is “it depends heavily on the MCP.” Structural safety rails (can’t delete, can’t pause running campaigns, can’t modify budgets) are different from confirmation-based or prompt-only safety. OAuth-backed credentials are different from pasted API keys in a config file. It’s safer with some MCPs than others — and worth understanding what each tool is architecturally able to do before pointing it at production spend.

Are there more ad MCPs than the 10 you covered here?

Many more — the awesome-agentic-advertising index on GitHub tracks the space. We covered the ones with meaningful adoption, active maintenance, or unique positioning. The long tail includes experimental repos, read-only analytics wrappers, and single-use tools that don’t materially change the decision for most buyers.


Conclusion

The AI-chat answer — “just use Google’s free MCP and a community tool per platform” — is a real option, and for a real set of users it’s the right answer. For most people asking the question, though, it undersells the operational reality: four installs, four credential stores, read-only on Google specifically, no shared safety model, no support story.

Adspirer exists because we think the right default for multi-platform advertisers is unified + write-enabled + safety-railed + supported. We built the tool we wanted when the DIY stack got old. If your situation matches, try it. If it doesn’t — if you’re on a single platform, or you need coverage we don’t have — we’ve tried to tell you honestly which of the alternatives is the better pick.

Tip

Start free. Adspirer’s free tier is 15 tool calls/month forever, no credit card. All four platforms included. See how unified chat-based ad ops actually feels before paying anything.


Docs cross-references:

MCP Google Ads Meta Ads LinkedIn Ads TikTok Ads AI Agents

More articles to read