Best Meta Ads MCP Servers in 2026 (12 Compared)

Disclosure: InsightfulPipe publishes this and is one of the 12 options ranked.

The verdict

The best Meta Ads MCP server in 2026 is InsightfulPipe. Of the 12 best Meta Ads MCP servers compared here, it is the only one that gives an AI agent 53 dedicated Meta Ads actions, lets an admin switch each one on or off per ad account, keeps deleting campaigns, ad sets and ads off until an admin switches it on, and leaves setting ads live to you. That makes it the safest way to let an AI agent change a Meta Ads account.

That last part decides more than it seems to. Eleven of the twelve can write, and on the other ten some tool lets the agent set a campaign, ad set or ad live. On InsightfulPipe the agent builds and pauses, and switching ads on stays with you in Ads Manager.

• Best overall: InsightfulPipe. Its Meta Ads MCP server carries 53 Meta Ads actions, 38 read, 12 write and 3 deletes, each switched on or off per ad account. From $29.99 a month.
• Safest way to let an AI agent change Meta Ads: InsightfulPipe. New campaigns, ad sets and ads start paused for you to set live, and deleting a campaign, ad set or ad stays off until an admin switches it on.
• Best value: InsightfulPipe. $29.99 a month covers 8 connected accounts across 50 platforms, with unlimited tool calls. Pipeboard's $29.90 plan covers 3 ad accounts.
• Best free option: gomarble's open-source server. Free and MIT licensed, but it only reads, and you host it yourself with a Meta access token.

The best Meta Ads MCP servers at a glance

Twelve servers, compared on what they can do in Meta Ads, whether the agent can set an ad live, and what write access costs to start.

How we tested

We read every capability, default and price from each vendor's own documentation, pricing page or public repository, and ran InsightfulPipe on live accounts. We checked every page on 2026-09-29, the same method as our Google Ads comparison.

We compared the twelve on eight criteria:

Where a vendor's own pages contradict each other, we quote both and attribute them to the vendor. Where a capability is not published, we say "not published" rather than guess. We count only features that ship today.

Search results and AI answers name more options than these twelve. We left out the ones that only read, cannot create a campaign yet, manage no campaigns or come from tiny community repositories, and name each one in the FAQ or under the self-hosted servers.

The single question that splits this list

One question splits these twelve: whether you need the assistant to report or to act. If it only reports, a free server can do the job. If it acts, two things decide the rest: what it can reach before anyone narrows it, and whether it can set an ad live.

If you only need reports, gomarble's open-source server is free, but it only reads, and you run it yourself with a Meta access token. Without a developer, a hosted read-only connection skips all of that. On InsightfulPipe, connecting an ad account as read-only locks every write for it until an admin reconnects it with read and write.

If you need the assistant to act, the list collapses to eleven, and the defaults matter most. Meta's own server creates new objects paused, then offers a tool that sets them live once your AI client confirms. Markifact leaves approval to a setting in your AI client. AdKit starts new projects on Draft + Publish. InsightfulPipe decides on the server, per ad account, before the session starts, and setting ads live stays with you.

Hosted servers that can change campaigns

Ten of the twelve servers are hosted and can create or change campaigns. InsightfulPipe leads them, and the other nine follow, Meta's own server first.

Best overall: InsightfulPipe

InsightfulPipe gives an AI agent 53 dedicated Meta Ads actions, and you decide which of them each ad account allows.

Coverage comes first. InsightfulPipe carries 53 Meta Ads actions, 38 read, 12 write and 3 deletes, across the whole life of a campaign:

• Reporting: insights for the ad account and its campaigns, ad sets and ads, with breakdowns such as age, placement and device, plus change history
• Campaigns, ad sets and ads: list, read, create and update each one, including budgets, bids, schedules and targeting, and pause anything that runs
• Creatives and media: create ad creatives, upload images and videos, build a slideshow and preview any ad
• Targeting research: search interests, behaviors, demographics and locations, and estimate audience size before you build an ad set
• Audiences, pixels and conversions: read custom and saved audiences and pixels, and create custom conversions
• Lead generation: list lead forms and create new ones on a Facebook Page you connect
• Account context: ad account details, Meta's minimum budgets and the Pages and Instagram accounts it can advertise with
• Deletes: remove a campaign, ad set or ad, switched off until an admin switches them on

Why it is the safest way to let an agent change Meta Ads

Control comes second, and it makes InsightfulPipe the safest way to let an AI agent change a Meta Ads account. Every limit is a switch, and your admin decides where each one sits.

An owner, admin or operator opens a connected ad account and picks exactly which of its actions the agent can use, across read, write and destructive, and can switch any of them on or off at any time. The agent gets exactly that set. InsightfulPipe's server enforces it, not your AI client, so the same limits hold in Claude, ChatGPT, Cursor, Claude Code and the CLI.

• Deletes start off. Removing a campaign, ad set or ad stays switched off on a new connection until an admin switches it on.
• Read-only means read-only. Connect an ad account as read-only and it locks every write until an admin reconnects it with read and write.
• New campaigns, ad sets and ads start paused, for you to set live in Meta Ads Manager when they are ready. The agent can pause anything that runs.
• Every ID gets checked. Each campaign, ad set, ad, creative, audience or pixel ID in a request must belong to the ad account you picked.
• Roles hold. Members on the user role only view the accounts an owner or admin grants them, and writes need an operator, admin or owner.

Meta's own server takes a different route. It also creates new objects paused, but its activation tool "starts spending budget", and the check before that step is a confirmation in your AI client. Markifact's MCP leaves approval to your client's settings too. That works while someone watches the chat. On a scheduled run, nobody does.

InsightfulPipe draws the boundary on the server instead. The agent knows its limits before it plans, works inside them, and leaves switching ads on to you. Turn on the twelve actions an account needs, and the model works from a twelve-item list instead of fifty-three, with fewer wrong actions to pick from.

What else you get

• 50 platforms on one login, including Facebook Pages, Instagram, Google Ads and GA4, so one agent can put Meta spend next to organic reach and site sessions.
• Meta's public Ad Library: the Ads Libraries server searches competitors' Facebook ads by keyword or by Page.
• Many ad accounts, kept apart: each ad account you connect gets its own action list, grouped under brands.
• 8 free Facebook Ads Claude skills, including an account auditor and a budget allocator, plus a free prompt library.
• A CLI: the insightfulpipe npm package runs the same reads and writes from a terminal, under the same controls.

Pricing: Starter is $29.99 a month for 8 connected accounts, Growth $89.99 for 35 and Agency $299.99 for 150, with every MCP server and the CLI on every plan. Start a trial and connect your first ad account.

Where it is the wrong answer: if you only ever need reports and you have a developer to host a server, gomarble's free one will do.

Best for: agencies and in-house teams who want an assistant that runs Meta Ads within limits they set, ad account by ad account.

Meta's official ads MCP server

Meta hosts its own server, and it can set ads live once your AI client asks you to confirm.

Meta's docs describe "a Meta-hosted Model Context Protocol server that lets you manage Meta ads directly from an AI agent", at https://mcp.facebook.com/ads. Its tool pages cover reporting, ads, catalogs, datasets and experiments. It creates campaigns, ad sets and ads paused, and ads_activate_entity sets them live, a tool Meta says "starts spending budget".

Meta's developer blog, updated on 2026-09-22, adds rules: "Rules define the actions an AI agent can and cannot take on a given ad account or catalog." Its rules guide says they are "in limited availability", and each rule denies one action, such as creating ads, raising a budget past a limit or changing delivery status. The blog suggests a "read-only or approval-required posture", yet the guide lists no approval rule. Meta publishes no price, and its newsroom post calls the connectors an open beta.

Pros

• Tools for catalogs, datasets, A/B tests and lift studies.
• New campaigns, ad sets and ads start paused.

Cons

• Its activation tool "starts spending budget", behind your AI client's confirmation.
• Rules are in limited availability, and one delivery status rule covers pausing and activating alike.
• No price, and Meta is "gradually rolling out access to tools".

Best for: advertisers who need Meta's catalog and experiment tools and will manage its rules themselves.

Markifact

Markifact runs Meta Ads operations behind a handful of MCP tools, and its own pages disagree about who approves a write.

Markifact's homepage says "Nothing goes live without you" and "Human in the loop, always." Its docs call the MCP approval model "Client-controlled (approve writes per your client's settings)". Its MCP docs say a new server reaches all operations and all connected accounts until you narrow it. Its status operations accept active. Pro is $79 a month for one seat, and Team $249.

Pros

• Creates custom and lookalike audiences.
• New campaigns default to paused.

Cons

• A new MCP server reaches all operations and all connected accounts until you narrow it.
• Over MCP, the approval before a write depends on your AI client's settings.
• Status operations accept active, and plans start at $79 a month.

Best for: teams already running automated workflows inside Markifact.

Pipeboard

Pipeboard documents about 30 Meta Ads tools, and what a token can reach depends on the plan you pay for.

Its API reference names 33 tools, 14 of them writes, and its open code creates campaigns, ad sets and ads paused. A caller can still pass active, and the update tools accept it. A standard token calls every tool. Per its token guide, limiting a token to chosen tools starts on Pro, $29.90 a month, and to chosen ad accounts on Agency, $199. The free plan covers 2 ad accounts and 30 tool runs a week.

Pros

• 33 documented tools, 14 of them writes.
• New campaigns, ad sets and ads default to paused.

Cons

• A standard token calls every tool, and per-account limits start at $199 a month.
• Its update tools accept active, so an agent can switch an ad on.
• Its README promises "Explicit confirmation on every write", with no mechanism in its open code.

Best for: small teams testing Meta writes on two ad accounts.

AdKit

AdKit saves changes as drafts, and its own docs say new projects can publish them.

AdKit sits between the agent and Meta: "Every change becomes a draft first." Four access levels decide what happens next: Disabled, Read-only, Draft only and Draft + Publish. Its docs say new projects start on Draft + Publish, where "a misread request can go live with real spend behind it." Its Meta page says "Nothing reaches your live ad account until you review and click publish." Single Project is $29 a month for one ad account per platform.

Pros

• Every change starts as a draft.
• You can set an access level per agent.

Cons

• New projects start on Draft + Publish, per its own docs.
• Access comes in four levels, not action by action.
• No Meta tool count published.

Best for: teams who want to review every change inside AdKit's dashboard.

AdAdvisor

AdAdvisor is a Meta-only MCP that stages changes as drafts, and approving a draft runs through the same MCP.

Its llms.txt counts 42 curated tools for Meta and lists Google, TikTok and Snap as planned. Its MCP page says "Every change comes back as a draft you approve", while the llms.txt counts "approving a draft" as one of your MCP tool calls. Reads and writes share one quota: 20 calls a month free, then $19.99 a month for 40 a day.

Pros

• Changes start as drafts by default, with a full audit log.
• Creates campaigns, audiences and lookalikes from chat.

Cons

• Meta only today. Google, TikTok and Snap are plans, not tools.
• Approving a draft is itself an MCP tool call.
• Reads and writes share a quota of 20 calls a month free and 40 a day on Starter.

Best for: solo Meta advertisers who want each change staged as a draft.

Adspirer

Adspirer covers Meta alongside Google, Amazon and TikTok, and its own docs disagree about what it can change.

Its Meta docs say "Every campaign Adspirer creates starts paused, and so do its ad set and its ad." Its update tool accepts "'ACTIVE', 'PAUSED'" and warns "Changes take effect immediately." Yet its server comparison says it "cannot pause running campaigns", beside a pause tool in the same docs. Approval before a write is your AI client's setting. Plus is $49 a month for 150 tool calls.

Pros

• New campaigns, ad sets and ads start paused.
• It cannot delete existing campaigns.

Cons

• Its update tool accepts active, and changes take effect immediately.
• Its own docs contradict each other on pausing and budget edits.
• Tool calls are metered: 15 a month free, 150 on the $49 plan.

Best for: teams already running Adspirer's own ad agents.

PaidSync

PaidSync previews Meta writes as dry runs, and says its own workspace limits stop at the MCP door.

Its Meta Ads MCP page says Meta writes preview as a dry run by default, and delete and pause tools refuse a live call without a confirm flag. Its security page says its spend ceilings and kill switches "apply to changes made in the workspace, not to calls from your own assistant over MCP."

On paused defaults its pages say two things: "new campaigns and ads start paused or as drafts", and "The campaign lands paused because you said so, not because the tool decided." The free plan has 15 tasks a month, then $99.

Pros

• Writes preview as a dry run by default.
• Delete and pause tools need an explicit confirm flag.

Cons

• Its workspace limits do not cover calls over MCP.
• Enable tools let the agent set a campaign, ad set or ad live.
• Its own pages describe the paused default two ways.

Best for: agencies that already run client accounts in PaidSync's workspace.

Windsor.ai

Windsor.ai is a data pipeline first, with a set of Meta writes on top.

Its write actions page lists 20 Meta actions, including Enable Campaign, Enable Ad Set and Enable Ad. Its campaign, ad set and ad create actions each say "Created paused by default so it does not spend until enabled." There is no delete action. Its Facebook Ads MCP page says writes run in Claude, ChatGPT and Copilot, while Cursor gets reads. Plans are priced by data source: $23 a month covers 3.

Pros

• 20 documented Meta write actions.
• Create actions document a paused default.

Cons

• Enable actions let the agent set a campaign, ad set or ad live.
• Writes do not run in Cursor.
• No approval step documented.

Best for: teams already piping reports through Windsor.ai.

Ryze AI

Ryze AI sells an autonomous AI marketer, and its MCP server leans on raw Graph API calls.

Its Meta Ads MCP page counts 12 tools, 9 read and 3 write. The writes are generic Graph calls, and the page notes "Status ACTIVE/PAUSED lives here." The approval it describes is a setting in Claude: keep the write tools "on 'Needs approval'". Ryze publishes no paused default. Its MCP page says "Free to connect", while its pricing page starts at $89 a month.

Pros

• Ad Library search on the same server.
• Deletes run only on an item-by-item approved list.

Cons

• No paused default published for its raw writes.
• Its approval step is a Claude setting you keep on "Needs approval".
• The free-to-connect claim and the $89 pricing page disagree.

Best for: advertisers who want Ryze's own agent to run the account.

Self-hosted servers

These two run on your own machine with a Meta access token. One only reads, and the other writes with no approval step.

gomarble facebook-ads-mcp-server

gomarble's server has 21 MIT-licensed tools, and every one of them only reads: the code sends GET requests and nothing else. Its last substantive commit is from 2026-07-29, and its README documents Cursor and Claude Desktop.

Best for: free reporting, if you can host it.

mikusnuz/meta-ads-mcp

mikusnuz/meta-ads-mcp is MIT-licensed TypeScript. Its README's headline tool count and its own category counts disagree. The code can create, change and delete campaigns, audiences and business users with no approval step documented. It needs a META_ADS_ACCESS_TOKEN you manage, and its README names no AI client.

Best for: a codebase to read and fork.

Repositories we would not use

More repositories show up in Meta Ads MCP searches. None of them is a server we would point at a live ad account.

• irinabuht12-oss/google-ads-meta-ads-mcp: over 3,000 stars and no server code. Its server.json points at Ryze AI's endpoint.
• mathiaschu/meta-ads-analyzer: one commit, from 2026-02-26, and no server of its own.
• AbnerZx, gabe-almeida and attainmentlabs: community Meta Ads servers with 14 GitHub stars or fewer.
• trypeggy/facebook-ads-mcp-server: returns 404, and nictuku/meta-ads-mcp redirects to Pipeboard.

What an agent can do on a new connection

What an agent can do on day one depends on each server's defaults, and they differ more than the feature lists suggest. Ask four questions before you connect a live ad account.

What can it reach before anyone narrows it? Markifact's docs say a new server reaches all operations and all connected accounts. Pipeboard's standard token calls every tool. AdKit starts new projects on Draft + Publish. On InsightfulPipe, deleting a campaign, ad set or ad starts off until an admin switches it on, and a read-only connection locks every write until an admin reconnects it with read and write.

Who sets the limits, and who enforces them? Over MCP, Markifact and Adspirer leave approval to your AI client, and Ryze asks you to keep its write tools on "Needs approval" in Claude. PaidSync's workspace limits do not apply over MCP. Meta enforces rules on its side, where your business has them. On InsightfulPipe an owner, admin or operator sets the actions per ad account, and the server enforces them in every client.

Can the agent set an ad live? On ten of the twelve, yes: an activation tool on Meta, enable actions on Windsor.ai and PaidSync, status changes on Markifact, Pipeboard, Adspirer, Ryze AI and mikusnuz, and published drafts on AdKit and AdAdvisor. On InsightfulPipe the agent creates campaigns, ad sets and ads paused and can pause anything that runs. Ask it for any other status and the server refuses, because PAUSED is the only status it accepts.

What happens when nobody is watching? An approval prompt on every write stops a scheduled run at its first change. A boundary set on the server lets the agent plan inside it and finish, and switching an ad on waits for you. That is why InsightfulPipe is the safest of the servers that can write: every default starts on the careful side, and every one is a switch your admin controls. Whichever server you pick, set an account spending limit in Meta Ads Manager before any agent changes a budget.

Which should you choose?

Choose InsightfulPipe if you want an assistant that runs Meta Ads within limits you set, and a free read-only server only if reports are the whole job and you can host one yourself.

• You want an assistant to run Meta Ads within limits you set: InsightfulPipe. 53 actions switched on or off per ad account, campaign, ad set and ad deletes off by default, new ads paused, 50 platforms, from $29.99 a month.
• You run an agency with many client ad accounts: InsightfulPipe. Give each client's ad account its own brand and action list.
• You only need reports and have no developer: a read-only InsightfulPipe connection, which locks every write until an admin reconnects it with read and write.
• You only need reports and have a developer to host a server: gomarble's free server, which only reads.
• You need Meta's catalog and experiment tools: Meta's own server, in open beta with no published price.
• You already pay Windsor.ai for reporting: its MCP is included, so check which of its 20 Meta writes you need.

You can run more than one: MCP clients support several servers in one session.

Connect InsightfulPipe to Meta Ads in three steps

Connecting InsightfulPipe to Meta Ads takes a Facebook sign-in and one server URL. You need no Meta developer app, no app review and no token to create or renew.

1. In InsightfulPipe, open Connected accounts, choose Facebook Ads, pick Read-only or Read & write, then sign in with Facebook and select the ad accounts to connect.
2. In Claude, add InsightfulPipe from the connector directory, or add https://facebook-ads.insightfulmcp.com/ as a custom connector and click Connect. In Claude Code, run claude mcp add --transport http facebook-ads-mcp https://facebook-ads.insightfulmcp.com/.
3. Open the ad account's actions, keep the ones you want, and ask your first question.

Watch: Connect Meta Ads to Claude and audit spend and ROAS from one prompt, 2 minutes 39 seconds

What the other vendors do not publish

Several claims here could not be checked, because the vendors publish nothing to check against or say two different things.

• Meta: no price, no end date for its open beta, and a blog that promises to "Create, edit, and delete campaigns, ad sets, and ads" with no such delete tool in its docs.
• Markifact: three counts of its own MCP tools: 7 in its docs, 8 in its architecture notes and ten in its READMEs.
• Pipeboard: no mechanism for the confirmation its README promises, and tool counts of 30+, 33 and 42.
• AdKit: no Meta tool count at all.
• Adspirer: Meta tool counts of 36, 58 and 60 across its own pages.