The 10 Best Ad MCP Servers in 2026 — Honest Comparison

Model Context Protocol (MCP) has quietly become the connective tissue between AI assistants and the systems that actually run businesses. Nowhere is that more visible than in advertising — there are now dozens of MCP servers promising to let you manage Google Ads, Meta, LinkedIn, or TikTok through ChatGPT or Claude. Most of them don’t do what you think they do. Some are read-only when you need writes. Some are single-platform when you need four. Some are solo-maintained side projects that will break the week Meta ships its next Graph API change. This guide ranks every credible ad MCP server we’ve tested, with honest strengths and limitations for each. No superlatives, no fluff — the criteria are below, and Adspirer is included because it’s ours and we think it’s earned a spot. When one of the alternatives is the right answer, we say so.

​

How We Ranked These

Six criteria. Every MCP gets evaluated against the same checklist. A tool that scores well on most criteria but fails on one is still dangerous — a read-only MCP is great for diagnostics and useless for ad ops. A well-maintained but single-platform tool is fine if you run one platform and terrible if you run three.

​

Quick Comparison Table

Now the per-MCP breakdowns.

​

1. Adspirer — Unified Four-Platform Ad Ops

​

What it actually is

Adspirer is a remote MCP server that connects your AI assistant — ChatGPT, Claude, Cursor, Codex, Windsurf, OpenClaw, Gemini CLI, Manus, or Perplexity — to Google Ads, Meta Ads, LinkedIn Ads, and TikTok Ads with full read and write access. 100+ tools across the four platforms, plus cross-platform analysis and agent skills for Claude Code, Cursor, Codex, and OpenClaw.

​

Who it’s for

Marketers and agencies running two or more platforms who want to launch, optimize, and pace campaigns from chat — not just query them. Teams that value safety rails over maximum permissiveness. Anyone whose AI client isn’t Claude Desktop (because most community MCPs only document Claude Desktop).

​

What we like

• Real writes, not just reads. Launch Search, Performance Max, Image, Video, Carousel, and Sponsored Content campaigns. See the capability matrix.
• One OAuth per platform, one config. Add all four platforms to one Adspirer account; configure your AI client once.
• Structural safety rails. Cannot delete existing campaigns, cannot pause running campaigns, cannot modify existing budgets. New campaigns are created paused by default.
• Broadest AI-client coverage on the market. See the AI Clients section for setup across twelve supported clients.
• Agent skills for developer stacks. Claude Code, Cursor, Codex, and OpenClaw have purpose-built agent packages — not just generic MCP tool use.

​

What we don’t like

• Only four platforms. If your stack needs Reddit, Pinterest, X, Amazon DSP, or Microsoft Ads, Synter’s wider list is the better match today.
• Commercial by design. Free tier exists (15 tool calls/month forever), but it’s not free forever at scale. See pricing.
• Remote-only. Self-hosted deployment is available under Enterprise; there’s no OSS-yourself path.

​

When this is the right pick

You run 2+ platforms, you need writes, and your team includes people using Cursor, Claude Code, Codex, OpenClaw, or ChatGPT — not just Claude Desktop.

​

2. Synter — The Broadest-Coverage Unified MCP

​

What it actually is

Synter is a commercial unified MCP server that emerged from stealth in 2026. It covers 14 platforms — Google, Meta, LinkedIn, TikTok, Reddit, Pinterest, Snapchat, X, Microsoft, Taboola, and more — through direct OAuth connections to each platform’s official API. Full read and write. Pricing starts around $199/month.

​

Who it’s for

Advertisers and agencies whose stack goes beyond the “big four.” If Reddit Ads or Pinterest is a meaningful chunk of spend, Synter is the most direct unified option.

​

What we like

• Platform list nobody else matches. 14 is 14.
• Full read+write across all of them. This isn’t a reporting tool.
• Unified budget control and conversion tracking management built in.

​

What we don’t like

• Newer product. Less public case-study depth than older tools. Not a reason to avoid it — just something to weight.
• Claude Desktop-centric positioning. Documentation leans heavily toward Claude Desktop and Cursor. Agent-skill packages for other AI clients are thinner than Adspirer’s.

​

When this is the right pick

You need coverage outside Google/Meta/LinkedIn/TikTok. Reddit Ads and Pinterest are the classic tiebreakers.

​

3. Pipeboard — Mature, but Still Split per Platform

​

What it actually is

Pipeboard is the most mature single-platform Meta Ads MCP in the ecosystem — 791 stars, 133 releases, 489 commits on GitHub as of April 2026. Licensed under Business Source License 1.1 (converts to Apache 2.0 in 2029). Read + write: campaign creation, ad set and ad creation, creative upload, 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 at different URLs. Not LinkedIn.

​

Who it’s for

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

​

What we like

• Genuinely active maintenance. Release velocity on the Meta MCP is the highest in the category.
• Paused-by-default safety on campaign creation.
• Legitimate company with a real product roadmap, not a side project.

​

What we don’t like

• No LinkedIn support. Hard stop for B2B.
• Three separate installs if you want Meta + Google + TikTok. Same operational shape as any DIY stack, just with one vendor.
• Cross-platform reasoning is awkward. Asking a single question spanning Meta and Google means the LLM reconciles two tool namespaces across two MCPs.

​

When this is the right pick

Meta is 80%+ of your spend and you’re comfortable installing two or three separate MCPs from the same vendor.

​

4. Google Ads MCP (Official, google-marketing-solutions/google_ads_mcp)

​

What it actually is

Google’s own open-source MCP server for the Google Ads API, released in early 2026. Exposes exactly two tools: list_accessible_customers (list your accounts) and search (run raw GAQL queries). Free. Self-hosted. Gemini-focused by design.

​

Who it’s for

Developers and analysts who want to ask diagnostic questions about Google Ads performance in natural language, and who are comfortable writing GAQL (or letting an LLM write it).

​

What we like

• Free and official. No subscription, no OAuth intermediary.
• GAQL is powerful. If your LLM writes valid GAQL, you can query almost anything about historical performance.
• Google-maintained. API compatibility is guaranteed.

​

What we don’t like

• Read-only. Full stop. 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.
• Google Ads only. No Meta, LinkedIn, TikTok.
• GAQL expertise dependency. When the LLM writes invalid GAQL, you debug it.
• Self-host overhead. You’re running and maintaining it.

​

When this is the right pick

You only need to ask questions about Google Ads in chat, you’re fine with self-hosting, and you’ll never need writes. See our full breakdown of what this MCP does and doesn’t do.

​

5. Ryze / irinabuht12-oss/google-meta-ads-ga4-mcp

​

What it actually is

A unified open-source MCP server covering Google Ads (150+ tools), Meta Ads (80+ tools), and GA4 (20+ tools). Full read + write. MIT licensed. 273 GitHub stars, 35 forks. Tied to Ryze AI, a commercial autonomous optimization platform.

​

Who it’s for

Advertisers who want a single unified MCP across Google + Meta + GA4, 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 — nobody else on this list combines ad platform writes with analytics reads natively.
• Legitimate OSS foundation tied to a funded commercial platform.

​

What we don’t like

• Early-stage. Only 3 commits on the main branch at time of writing. Mature enough to try, not mature enough to bet mission-critical spend on without monitoring.
• No LinkedIn or TikTok.
• Self-host overhead.

​

When this is the right pick

You’re an engineer, you run Google + Meta only, and GA4 reads inside the same MCP sound useful.

​

6. Flyweel MCP

​

What it actually is

A remote MCP from Flyweel that links your Google Ads and Meta Ads to Claude, Cursor, or VS Code. Included free with a Flyweel account (optional; can be used standalone). Takes about 30 seconds to set up.

​

Who it’s for

Flyweel customers who want to ask questions in chat about data that’s already flowing through Flyweel.

​

What we like

• Fast setup. Drop a config, start asking questions.
• Clear positioning. “Never posting or changing anything” is honest.

​

What we don’t like

• Read-only. It’s a chat-native dashboard, not an ad-ops tool.
• Two platforms only. Google + Meta.

​

When this is the right pick

You’re already a Flyweel customer and you just want performance-query chat over the data you’ve already unified.

​

7. AdsMCP TikTok Ads MCP

​

What it actually is

An open-source MCP server for the TikTok Business API — campaign management, performance analytics, audience management, creative uploads, and custom reports. MIT licensed, built-in rate limiting (1000 req/hour).

​

Who it’s for

TikTok-heavy advertisers who want a standalone MCP specifically for TikTok Ads.

​

What we like

• Read + write with real campaign management (not just diagnostics).
• Built-in rate limiting and retry logic — surprisingly solid operational hygiene for a community repo.

​

What we don’t like

• TikTok only.
• Self-host.

​

When this is the right pick

TikTok is the majority of your spend and you want a dedicated tool.

​

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.
• Full campaign lifecycle, not just reads.

​

What we don’t like

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

​

When this is the right pick

LinkedIn is your only platform 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. 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.

​

What we like

• Fully transparent — read the code before running it.
• No subscription.

​

What we don’t like

• Self-host is real work.
• Meta only.
• Community-paced maintenance.

​

When this is the right pick

You’re a dev who values running everything locally and won’t install a remote service.

​

10. Windsor.ai MCP (Different Category)

​

What it actually is

Windsor.ai is a marketing data pipeline — 325+ data connectors syncing ad spend, analytics, CRM, and commerce data into a unified warehouse. Their MCP lets an LLM query that warehouse in natural language (“what campaigns had the best ROAS last month?”).

​

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 makes it queryable without SQL.
• Massive source coverage (325+).

​

What we don’t like

• Not a campaign management tool. Don’t come here to launch campaigns or change budgets — that’s a different product category.
• Early-stage MCP repo (3 stars, 9 commits).

​

When this is the right pick

Your job is cross-warehouse reporting, not campaign execution. Pair Windsor (for reporting) with Adspirer or Synter (for execution). They’re complementary, not substitutes.

​

Honorable Mentions

Not every MCP merited its own breakdown, but these are worth knowing about:

• cohnen/mcp-google-ads — Nicer developer experience than Google’s official, same read-only ceiling.
• brijr/meta-mcp — Thorough Meta lifecycle, solo-maintained.
• Facebook Ads Library MCP (trypeggy) — Public ads library research, not account management. Complementary to any of the above.
• ysntony/tiktok-ads-mcp — TikTok read-only.
• radiateb2b/mcp-linkedin-ads — LinkedIn analysis and benchmarking focus (not full write).
• promobase/unified-ads-mcp — Early-stage unified Google + Meta experiment.

The awesome-agentic-advertising index on GitHub is worth bookmarking if you want to watch the space move.

​

How to Choose

Four decision axes cover almost every buyer.

​

By how many platforms you run

​

By AI client

​

By what you value most

​

The combo many teams end up running

Multiple MCPs is a legitimate pattern. A common agency setup:

1. Adspirer for Google + Meta + LinkedIn + TikTok writes, multi-client routing, and the conversational surface team members already use.
2. Windsor.ai MCP (or similar) for warehouse-BI queries across more data sources than any single ad MCP covers.
3. Facebook Ads Library MCP for competitor research, which is read-public-data and fits nowhere else.

These three solve different jobs. They don’t conflict.

​

Safety: the Thing No One Talks About

Most MCP comparisons stop at features. Safety is where the interesting differences show up. An LLM with write access to a production ad account will, eventually, be asked something like “delete everything that didn’t perform last week.” The question is whether the tool layer can say no. Three models exist in the ecosystem:

1. Prompt-only safety. The MCP exposes destructive tools, and a system prompt tells the LLM not to use them. This is fragile. Prompt injection or a misread intent is all it takes.
2. Confirmation-required safety. The tool runs, but asks the user to confirm. Better — but depends on the user reading the confirmation carefully every time.
3. Structural safety. The destructive tools simply don’t exist in the MCP server. An LLM can’t call what isn’t there.

Adspirer uses model 3 for the worst classes of action. We cannot delete existing campaigns. We cannot pause running campaigns. We cannot modify existing budgets. The tools for those operations aren’t in the server at all. Full detail in the capabilities doc. Most community MCPs use model 1 or 2 (when they document safety at all). This isn’t a condemnation — it’s fine for personal accounts and read-heavy workflows. It’s a real risk for agency or team workloads running real spend. A broader security note: Astrix Security’s research found that 53% of public MCP servers use long-lived static credentials (API keys, PATs), and only 8.5% use OAuth. Many community ad MCPs ask you to paste a write-scoped access token into a config file. That token, sitting in plaintext on a laptop, is a different security posture than OAuth 2.1 with PKCE. For the full Adspirer security model, see security.

​

FAQ

​

Related Reading

Prefer the opinion-driven version? Read the blog post: The 10 Best Ad MCP Servers in 2026 — Honest Rankings.