Best MCP for Meta Ads in 2026: 20 Servers Compared and Ranked

AdKit is a hosted MCP server built for running ads, not a generic connector. It talks to Meta through the official Marketing API, and every campaign, ad set, ad, and creative your agent touches is saved as a draft you approve before it goes live. You can loosen that once you trust it; the default protects your budget.

It pairs that safety with the deepest toolkit here: a library of 500,000+ real ads, AI creatives, a swipe file, and a free Meta Ads strategy skill. One connection covers Meta, Google, TikTok, Reddit, X, LinkedIn, and Microsoft. AdKit is a reviewed Meta Tech Partner, setup is three clicks with no tokens or terminal, there are no daily request caps, and it works with Claude, ChatGPT, Cursor, and any MCP client.

In 2026 Meta shipped its own Meta-hosted MCP server, plus a CLI. It creates and edits campaigns, ad sets, and ads through natural language, pulls reports, and builds product catalogs, with no developer credentials or coding required. The most direct line to the Marketing API you can get.

The limit is in the name: Meta only, and account management only, with no research or creative tools. It is free for eligible advertisers in open beta within a daily request limit, and there is no draft or approval step, so changes apply straight to your live account. Official and promising, but narrow.

Adspirer is a hosted MCP with 178+ tools across Meta, Google, LinkedIn, and TikTok. LinkedIn Ads with 45 dedicated tools is rare in this set, the free tier needs no card, and new campaigns are created paused by default.

The catch is metering. Adspirer charges per task: the Plus plan gives you 150 a month, and one campaign launch can burn 20 to 50, so active testing climbs the tiers quickly. No dashboard, ad library, competitor research, or creative generation. A clean API wrapper with good coverage that stops at the API.

Synter has the widest platform coverage of any ad MCP here: 27 channels including Meta, Google, TikTok, LinkedIn, Amazon, Pinterest, Walmart Connect, and Instacart, more than AdKit. It is MIT open-source, runs hosted or as a CLI, and its tiered autonomy auto-executes low-risk actions while routing high-impact ones for approval.

For anyone living in Meta, the counterpoint: it emerged from stealth in March 2026, its per-platform tooling is templated and shallow, and there is no ad library, competitor tracking, or swipe file. Paid plans start at $20 a month plus a fixed 3% of managed spend and per-action fees. Breadth is its argument; Meta depth and draft safety are AdKit's.

Pipeboard is the developer's pick. It wraps Meta, Google, TikTok, and Snap in 120+ tools that expose every API parameter, the most flexible raw access here. New campaigns are created paused, and it plugs into n8n, Cursor, and custom Python. Founded in August 2025 by an ex-Google SRE.

It is built for engineers: you work with raw Meta API objects, edits to live campaigns execute with no gate, and there is no ad library, research, or creative tooling. The free tier caps you at 30 executions a week, and paid plans are priced per ad account, $29.90 for three, $99 for ten, $199 for 25, so an agency with 25 clients pays $199 a month against AdKit's $49 for unlimited brands.

Blend is the strongest pick for ecommerce brands on Shopify or WooCommerce. It pulls your product catalog natively, so your agent builds campaigns from live inventory, and it covers Meta, Google, TikTok, Microsoft, and Pinterest. At $29 a month it is the lowest entry price among the purpose-built ad MCPs here, from a company founded in 2020 with $3.6M raised.

The caveats: every tier has an operations cap (500/2,500/10,000 a month) that an active agent burns through quickly, and actions execute on live accounts with no draft or approval step. Built for ecommerce in that stack; for everyone else, the missing draft layer is the thing to weigh.

Zuckerbot is an open-source (MIT) CLI and MCP server for Meta Ads and an approved Meta partner, with 50 tools covering campaigns, audiences, creatives, and conversion tracking. It has an approval step before a campaign goes live, and the free tier is read-only, so you can explore safely. The most credible of the open-source servers here.

It is Meta only, installed from the terminal with an API key. Pricing is metered by API calls: Pro is $49 a month for 50,000, Scale $149 for 500,000, with overage per extra thousand; the earlier percentage-of-spend fee is gone. Choose it if you specifically want open-source code you can self-host. On coverage, setup, and the toolkit around it, AdKit is ahead even if Meta is your only channel.

Ryze is an optimization engine first and an MCP second. It runs 24/7 on your existing campaigns, adjusting bids and budgets between your agent sessions, and its Meta and Google MCP writes go through a per-action approval gate. If you want continuous optimization rather than create-and-launch, that is its distinct value.

The catches: credit pricing is opaque and every plan caps you by ad spend (the entry plan at $25k a month), TikTok and LinkedIn are marketing claims the MCP page does not confirm, and "Schedule mode" removes the approval gate and runs live. No creative or research tools.

Trust is where it fails our check. On 6 September 2026 we could not verify any of the six testimonials on its homepage:

• The tweet-styled review credits @gabri_isleia. That X account does not exist (404).
• The Trustpilot-badged review names "Daniel Roser, GB, 7 reviews". Trustpilot's page for get-ryze.ai lists another business name and 50 reviews, none by him.
• "Mathilde Biggs, CEO MotifDigital Agence" and the company "Rivert" have no trace online, and the same five names carry different titles and companies on Ryze's own checkout page.

None of the platform logos link anywhere. Treat its testimonials and performance claims as unverified until it publishes proof you can follow to a source.

Windsor.ai connects 325+ marketing sources to BI tools and warehouses, and its MCP is reporting-first. Founded in 2017 and acquired by team.blue in January 2026, it is the most mature company in this list, with SOC 2 Type II, GDPR compliance, and 17,000+ businesses.

It recently added Meta writes in Claude, ChatGPT, and Copilot: pause, enable, create, edit, set budgets, boost a post. There is no approval step in front of them, the toolset stays shallow next to a management server, and row overage billing adds cost. Strong trust posture; the core product is data integration, not campaign management.

An open-source Meta Ads MCP (39 tools) that runs as a Cloudflare Worker in your own account: campaigns, ad sets, creatives, audiences, reporting, and batch workflows, with full write access. For a developer who wants total control and no third party in the middle, that is appealing.

It is the most do-it-yourself option here: no hosted control plane, no shared Meta app, no dashboard. You deploy it with your own Meta app credentials and maintain it yourself, and there is no approval gate. A building block for engineers, not a product for marketers.

An open-source (MIT) server on npm with a sprawling 135 tools across campaigns, ad sets, ads, and creatives. It can create, update, pause, and delete, so on raw surface area it does a lot, and it is one of the more active small repos here.

It is a solo project with no hosted product, company, or support, and no approval or draft gate: writes go straight to the Marketing API and your live budget. Plenty of tools, none of the guardrails.

An open-source (MIT) TypeScript server with 35 read tools always on and 19 write tools you enable with an environment variable. New entities default to paused. For a small repo, that is careful design: hard to spend money by accident.

The catch is everything around it: unofficial and not affiliated with Meta, a single maintainer, self-hosted with your own credentials. The most responsibly built of the obscure repos, but still a repo, not a product.

An open-source (MIT) server that exposes Facebook Ads as REST-style endpoints with MCP framing. It can create, update, and delete campaigns, ad sets, and ads, so it covers the basics of writing to an account.

It is a starting point, not a product: a single commit on main, no approval gate, no hosted option, no real documentation. You deploy it yourself and own everything after that.

A small open-source (MIT) Python server with five functions: create full campaigns, then pause, activate, or delete them. Despite the size it is thoughtful about safety: new campaigns are created paused, and create defaults to a dry run.

It is unmistakably a weekend project, a handful of commits with no docs and no company behind it. Five functions is enough to demo, not to run a real account. A clean reference, not a tool for live spend.

An open-source (MIT) Python server built on FastMCP that wraps the Facebook Business API, and the earliest-stage option here, still on version 0.0.9.

It is barely documented, and whether it can write to an account or only read is unclear, so we mark write support as unknown rather than guess. Something to watch, not something to run your ads on.

Composio is a hosted connector platform, closer to a Zapier for AI agents than a Meta Ads tool: Meta Ads is one toolkit on a marketplace of 1,000+ connectors. The 52-tool Meta Ads toolkit creates campaigns, ad sets, ads, and audiences and updates budgets, handy if you are wiring one agent across a CRM, a spreadsheet, and a dozen other tools.

Judged as an ads tool, it is thin: no draft or review system, and you supply your own Meta OAuth app, which has to clear Meta's app review for real write access. Fine for engineers, a wall for everyone else.

Funnel.io is an established marketing data hub, and its MCP lets Claude or Cursor read the data you have already centralized, including Meta performance pulled through one of 600+ connectors. For analysis against clean, unified data, it is credible.

But it is read-only: it cannot create a campaign, change a budget, or launch an ad. Hosted, sold on a custom plan with no public price, no Meta partner badge found. A reporting companion to a management tool, not a replacement.

Triple Whale is a well-known ecommerce analytics platform, and its open-source MCP turns plain-language questions into calls against your Triple Whale data, surfacing profit, revenue, and Meta ad performance together. It is a certified Meta partner, and a natural fit for a Shopify brand already living there.

Like Funnel, it is read-only: it reports on Meta ads, it does not manage them. Self-hosted with an API key and an active subscription, so it only makes sense if you already pay for the platform.

Adzviser pipes data from 45+ sources into AI clients, spreadsheets, and BI tools, with all 12 destinations (Sheets, Excel, Claude, Gemini, Looker Studio, Power BI, BigQuery, and more) on every plan. Founded in 2022 by an ex-Google Ads engineer, it is an established reporting tool.

It reports on Meta ads, it does not run them: no write capability of any kind. Tiers jump from 3 sources at $48.99 to 7 at $148.99 with nothing between, and some users flag accuracy gaps against native dashboards. A reporting companion, not a substitute for management.

Flyweel is an "Ad Spend OS" with a read-only analytics MCP covering 222 metrics across Google (132), Meta (67), and TikTok (23). The MCP is free with unlimited tool calls on both tiers, the most generous free analytics offer here, and its standout is CRM-to-ad-spend attribution across 9 CRMs plus Xero invoice reconciliation.

The constraint is the section's: Flyweel reports on Meta ads, it does not run them. Eight MCP tools, one of which does actual analytics, and Meta data runs 1 to 3 hours behind. Founded in March 2025 with Antler backing, it is very new, but the free MCP and the attribution angle make it the most interesting read-only option for a finance or ops lens.