← All posts
Ad Performance
November 18, 2026

Ad MCP servers compared: Meta, Google and the tools that wrap them

by
Daria Bochkareva
Yellow curly Subscribe to Blog button

You already run Claude or ChatGPT next to Meta Ads Manager, and you want it inside the ad accounts. The question on the table is which MCP server to connect, what it will be allowed to do once it is in, and what keeps working after you close the chat. On demo calls the same question arrives as which of these do we use for what: Meta's own server, a third-party connector, a rules tool with a server of its own, sometimes all three at once.

We read the lists that rank ad MCP servers before writing this one. Every one of them ranks the same job: an assistant that makes a direct change, such as creating a campaign, moving a budget, or pausing an ad. None of them separates a change a model decides, which can come out differently each time, from a change a rule decides, which comes out the same every time. The two jobs need different servers, and the routing at the end assumes you may want one of each. Three things worth settling before you read the entries:

  • Decide which changes a model may make and which must follow a rule, and pick a server for each.
  • Connect read-only first where the server offers it, and add write access only for the changes you have decided to hand over.
  • Write down what runs when you are not in the chat, and where it is logged.

How we compared

We read each server from its vendor's own site and documentation as they stood on Nov. 17, 2026, with page dates in each entry where the vendor publishes one. We tested nothing hands-on. Every entry answers the same eight questions, from platforms to pricing, and each answer is what the vendor's own pages state. Pricing is quoted as published on the snapshot date. Birch is read on the same terms, from its own pages, and its entry carries the same lines as the others. The Meta entry is read from the text versions of Meta's developer pages, with our own Meta ads MCP guide for the reporting tools.

If you're after always-on rules that an agent writes and runs for you, the setup for Birch MCP takes under a minute.

Who decides the change

The first job is the change a model decides. You ask for a budget move, a pause, or a new campaign built from a brief, and the model in your client turns the request into calls; a scheduled agent does the same without a prompt. The result depends on how the model reads the account that day, so a person checks it before it goes live. Every platform MCP and every connector in this comparison is built for that job, and the differences between them are platform coverage, write depth, and what stands between the request and the account.

The second job is the change a rule decides. A rule that pauses ad sets above a cost per result, checked every hour on four platforms, takes the same action on the same condition every time and logs every check, so you know what it will do before it runs and can read afterward why it did. A server built for the first job has no place to keep such a rule.

Whatever runs when nobody is in the chat is one of three kinds, and the entries say which kind each server uses:

  • Rules decide on a threshold a person wrote down. The same condition, checked on the same schedule, with the same action every time. A rule can be read, audited, and dry-run, so the decision sits where a person can read it, with the agent drafting the rule and a person approving the spend.
  • Agents that propose read the account on a schedule and queue changes for approval. The model decides what to suggest; a person decides what goes live.
  • Agents that act change the account inside caps a person set. In the entries below, most of the agents’ changes wait for a person’s approval. 

A team that wants both jobs connects two servers: the platform's own or a connector for the changes a model makes, and a rules engine with an MCP for the changes that must come out the same every time. That is what the routing section at the end assumes.

The 13 at a glance

Server Platforms (writes) Reads Writes Own AI Safety default Clients Pricing (Nov. 17, 2026)
Google Ads MCP Google (assets and labels only) Any GAQL query Creates assets, manages labels None Propose-and-confirm; read-only config Claude, ChatGPT, Codex, Cursor, Gemini CLI Free, open source, self-hosted
Meta Ads MCP Meta Metrics, benchmarks, auction, log, audiences, pixels Creates campaigns, ad sets, ads, audiences, tests None Created paused; activation confirmed Claude Desktop, Claude Code, ChatGPT, Codex, Cursor Free
TikTok for Business MCP TikTok Performance reports Campaigns, audiences, creatives, budgets None None stated Any compatible agent Free
Snapchat Ads MCP None (reads Snapchat) Ads data within your role None (read-only release) None Read-only; admin approves each agent Claude, ChatGPT, Codex, Gemini, Antigravity Free
Adspirer Google, Meta, TikTok, LinkedIn, Amazon, ChatGPT Ads Pacing, tracking, performance Builds campaigns; agents monitor Yes, agents Created paused under a spend cap; approval ChatGPT, Claude, Claude Code, Cursor, Codex Free 15 tasks/mo; $40, $83, $167/mo
AdKit Meta, Google, TikTok, Reddit (campaign drafts) Account data Drafts campaigns, ads, keywords None Everything a draft until approved Claude, ChatGPT, Codex, Cursor, Perplexity $29/mo, $49/mo; 7-day trial
GoMarble Meta, Google Meta, Google, analytics, commerce Approved changes across connected accounts (Operator) Yes, agents Approval gates per workspace; Copilot read-only Claude, ChatGPT Work, Claude Code, Cursor $139/mo read-only; $399/mo writes
Markifact 14 ad platforms Reports across 50+ platforms Creates campaigns, budgets, creatives, workflows Yes, agents Read & write default; read-only per server; client confirms writes Claude, ChatGPT, Gemini, Cursor, Codex $79/mo, $249/mo; trial credits
PaidSync Builds on Google, Meta, LinkedIn, TikTok, Microsoft, ChatGPT Ads; edits 4 more Across 14 platforms Creates campaigns, keywords, ads, audiences; budgets, bids, status Yes, chat workspace Dry run on most creates, not uploads; confirm flag on destructive Claude, ChatGPT, Gemini, Cursor, Codex, n8n Free 100 tasks/mo; $99–$199/mo; $249+/mo teams
Pipeboard Meta, Google, TikTok, Snap, Reddit, LinkedIn, Pinterest Reporting across 11 platforms Creates campaigns, uploads creatives, budgets, bids None Scoped tokens from Pro, per account on Agency; Google creates paused Claude, ChatGPT, Claude Code, Cursor, Codex, n8n Free; $29.90, $99, $199/mo
Ryze AI Google, Meta (connector tools) Google, Meta, GA4 at once Pauses, updates bids, adds keywords; schedules actions Yes, agents Approval before any write Claude (official connector), ChatGPT, Cursor Included on every plan; from $89/mo
Birch Rules on Meta, Google, TikTok, Snapchat Rules, every run, bulk launches from Stage Creates, edits, duplicates, runs rules; webhooks Yes, Birch AI Write opt-in; preview and dry run; never toggles on edit Claude, ChatGPT, Claude Code, Codex All plans; 14-day trial
Madgicx Meta (Google read-only) Meta and Google performance Pauses, restarts, budgets, renames Yes, AI Marketer Shows the change, waits for go-ahead Claude, ChatGPT, Grok, Cursor, Codex, Gemini, Perplexity, n8n From $45/mo by spend; 7-day trial

Official platform MCPs

The platforms run these four servers themselves. Each is tied to one platform, and each stops where a rule would start.

Google Ads MCP

Platforms: Google Ads only; writes limited to assets and labels.

Reads: anything a GAQL query reaches, through three tools: list accessible customers, search, and resource metadata.

Writes: creates text, image, video, and extension assets, uploads a video to YouTube, and manages account labels. The docs state it cannot create, modify, pause, or enable campaigns, ad groups, ads, budgets, or bids.

AI: none of its own, and nothing runs on a schedule. The AI is the client you connect it to, such as Claude or ChatGPT.

Agent access: self-hosted, locally over stdio or on Cloud Run; named for Gemini CLI, Claude, Claude Code, Codex, and Cursor, with a custom connector for Claude and ChatGPT. Needs a Google Cloud project with the Google Ads API enabled and production access for live accounts. Setup is command-line and config-file work, and video uploads work only when the server runs locally, not on Cloud Run.

Safety default: every write is propose-and-confirm with a dry run first; a config switch turns labels and assets off for a read-only server.

Pricing: free and open source under Apache-2.0; you pay for your own hosting. Version v0.0.6 on Nov. 17, 2026.

Best for: pulling Google Ads data into any client you already run, on your own credentials, with nothing writing to campaigns.

Meta Ads MCP

Platforms: Meta only.

Reads: campaign, ad set, and ad metrics with breakdowns, trends and anomalies, the account opportunity score, benchmarks, auction ranking, the activity log, audiences, pixels and event quality, A/B tests, catalogs, and the Ad Library.

Writes: creates campaigns, ad sets, and ads, each created paused; updates existing ones; activates a paused entity, which Meta's docs say starts spending budget; and creates creatives, boosts Instagram posts, and creates and edits custom audiences, A/B tests, lift studies, and catalogs. The tool list has no scheduling or condition-watching tool.

AI: none of its own, and nothing runs on a schedule; Meta has validated Claude Desktop, Claude Code, ChatGPT, the Codex app and CLI, and the Cursor app and CLI, and says clients not on that list are not supported.

Agent access: you sign in through Facebook Login for Business with your Facebook account or a Meta Managed Account and approve the permissions the server requests, at minimum ads_mcp_management plus ads_read or ads_management. Permissions are revoked from your Meta account settings.

Safety default: created paused; your client confirms activation; the activity log mirrors the Ads Manager campaign history page and can be filtered by object, time window, event category, or user.

Pricing: free.

Best for: making one-off changes and audits on Meta from the client you already have.

TikTok for Business MCP Server

Platforms: TikTok only.

Reads: performance reporting for your TikTok ad accounts.

Writes: campaign management, audience configuration, and creative operations; TikTok's examples include creating a campaign, updating audience targeting, pausing an ad group, and adjusting budgets.

AI: none of its own; TikTok's page describes no scheduler, and the AI is the agent you connect.

Agent access: any compatible agent, with one server URL recommended for Claude; TikTok says no developer credential, API key management, or custom coding is needed.

Safety default: none stated on TikTok's overview page (updated Aug. 2026).

Pricing: free.

Best for: making one-off changes on TikTok from the client you already have.

Snapchat Ads MCP

Platforms: Snapchat only, read-only in this release.

Reads: your Snapchat Ads data, within the Ads permissions your role already has.

Writes: none; Snap's docs say an agent "cannot create, update, or delete anything," and write access is planned for a future release.

AI: none of its own, and nothing runs on a schedule; Snap lists Claude, Codex, ChatGPT, Antigravity, and Gemini, and other clients are not supported.

Agent access: you sign in with your Ads Manager business login after an Admin or Business Admin approves the agent for the organization; each member then authorizes their own access. Members revoke it in their Snapchat account settings, and an admin can revoke an agent for the whole organization.

Safety default: read-only, and connecting an agent grants nothing beyond the member's existing Ads permissions.

Pricing: free.

Best for: reading Snapchat Ads data in the client you already have, on the logins your team already uses.

Rules engines with an MCP

Two servers sit on top of a rules engine, so the rules run inside the product, the same way every time, and the MCP reaches them.

Birch

Platforms: Meta, Google, TikTok, and Snapchat, with writes on all four. Birch is an official partner of Meta, Google, TikTok, and Snapchat.

Reads: every rule as its conditions and actions, every run with the values each check saw, which rules fire and which never do, and the ads launched through Stage, Birch's tool for creating ads in bulk from Google Sheets or Google Drive.

Writes: creates a rule from a plain description with a preview before saving, and edits its conditions, actions, settings, schedule, active dates, filters, and ad accounts; it also activates, pauses, renames, duplicates, moves, and runs rules, and manages webhooks. Outside the rule tools it stays read-only on campaigns, ad sets, and ads. It has 34 tools: 18 read-only and 16 write.

AI: three parts. Rules run on their own schedule on all four platforms. Birch AI, the chat inside Birch, analyzes performance across them and drafts rules; with Meta's ads MCP connected, it also makes one-off changes on Meta, created paused, and its Routines run reviews and Meta changes on a schedule. Birch MCP, the server in this entry, brings the rules into the AI client you already use.

Agent access: Claude, ChatGPT, Claude Code, Codex, and any MCP client, connected in under a minute per the setup article; write access is opt-in on the consent screen, and access is revoked from Settings. Prompts to start with are in the Birch MCP guide.

Safety default: edits are checked like changes in the rule editor and can be dry-run first; an edit is refused if the rule changed since your AI read it; editing never switches a rule on or off; the rule's history shows a change made through Birch MCP under the user who made it.

Pricing: included on all plans, with a 14-day free trial and no card; plans are priced by monthly ad spend, and the rule tools need a plan with automated rules (Nov. 17, 2026).

Best for: writing, auditing, and running always-on rules on four platforms from your agent, with Birch AI inside Birch for analysis and rule drafts on all four and one-off changes on Meta.

Madgicx

Platforms: Meta for writes; a second server reads Google Ads. Madgicx states it is an official Meta Partner.

Reads: Meta ad account performance through 48 tools; Google customers, campaigns, ad groups, and ads through six.

Writes: pauses or restarts campaigns, ad sets, and ads, changes budgets by an amount or a percentage, changes bids, and edits names, including bulk renames. Madgicx's Facebook Ads MCP guide states it "does not create campaigns or ads."

AI: rules, plus an agent: the AI Marketer audits daily and acts with your approval, and custom automation rules run on schedules inside Madgicx.

Agent access: Claude on web and desktop, Claude Code, ChatGPT on paid plans, Grok, Cursor, Codex, Gemini, Perplexity, n8n, and agent frameworks, over OAuth; the team you pick at sign-in decides which ad accounts the assistant can use.

Safety default: it shows you the exact change first and waits for your go-ahead.

Pricing: included in a Madgicx subscription and the 7-day trial; Pro Complete is priced by monthly ad spend, $45 a month under $1k to $329 a month at $20k to $30k, with higher spend quoted in the app (pricing page Nov. 17, 2026; Facebook Ads MCP guide dated Nov. 12, 2026; MCP page updated Nov. 11, 2026).

Best for: opening a Claude or ChatGPT window onto a Meta account already running Madgicx rules.

Connectors and agent platforms

These seven servers wrap several platforms' APIs behind one connection. The connectors expose tools and leave the thinking to your client; the agent platforms add agents of their own that watch the account and propose.

Adspirer

Platforms: Google Ads and Shopping, Meta and Meta Catalog, LinkedIn, TikTok, Amazon Ads, and ChatGPT Ads.

Reads: pacing, tracking, and performance across every connected account.

Writes: builds campaigns from a brief; its agents watch pacing and cost per acquisition and queue what they would change.

AI: yes; agents that propose.

Agent access: ChatGPT, Claude, Claude Code, Cursor, Codex, Windsurf, Perplexity, and Manus; listed in the ChatGPT app store and the Claude directory. Platforms are authorized through their own OAuth, so Adspirer never holds your passwords.

Safety default: new campaigns stay paused under a hard spend cap, and every high-impact change is logged for your approval.

Pricing: Free Forever at 15 tasks a month; Plus $40, Pro $83, and Max $167 a month at 150, 600, and 3,000 tasks; agencies custom.

Best for: running agents that propose across six platforms.

AdKit

Platforms: drafts full campaigns on Google, Meta, TikTok, and Reddit. AdKit states it is an officially approved Meta and Google tech partner.

Reads: the connected accounts, one per platform per project.

Writes: drafts campaigns with targeting and budget set, clones and generates ad variants, and drafts keyword variants into a Google ad group.

AI: none of its own, and nothing runs on a schedule; "the agent is an accelerator, not a requirement." AdKit's own AI features run on an API key you bring, with no credit system.

Agent access: Claude, ChatGPT, Perplexity, Claude Code, Codex, Cursor, Gemini CLI, n8n, and any MCP agent; everything also works from the dashboard.

Safety default: every change your agent makes is a draft inside AdKit, and nothing touches a live account until you approve it.

Pricing: Single Project $29 a month for one brand and one ad account per platform; Multiple Projects $49 a month; 7-day trial; unlimited tool calls on every plan (Nov. 17, 2026).

Best for: preparing launches on four platforms with your agent, while a person presses publish.

GoMarble

Platforms: Meta and Google; write actions from the Operator plan.

Reads: Meta, Google, analytics, commerce, and creative signals through 10 connectors on Copilot and 80 or more from Operator.

Writes: approved changes across connected ad accounts, from the Operator plan; the jobs its site names are campaign builds, budget shifts, and waste cleanup, each prepared for your approval.

AI: yes; agents that propose. They watch every account around the clock, explain what changed, and queue the next move; a ready-made creative fatigue agent watches active ads and flags fatigue in Slack.

Agent access: MCP is included on every plan; Claude, ChatGPT Work, Claude Code, Cursor, n8n, and other clients. Copilot connects read-only; Operator adds write actions.

Safety default: consequential actions go through the permissions and approval gates set for your workspace, you can limit which accounts and tools an agent uses, and GoMarble keeps a record of what was proposed, approved, rejected, and completed.

Pricing: Copilot $139 a month, read-only, up to $50k spend and five accounts; Operator $399 a month with writes, up to $500k and 15 accounts; Scale $799; Enterprise custom; 7-day trial; MCP tools and agent runs use credits (page dated Nov. 17, 2026).

Best for: running an agent that proposes by account on Meta and Google, with the approval path set per client.

Markifact

Platforms: 14 ad platforms including Meta, Google, TikTok, Microsoft, and Snapchat, within 50 or more integrations and 1,000 or more operations.

Reads: reports and audits across every connection.

Writes: creates campaigns, updates budgets, generates creatives, and runs workflows from your client.

AI: yes; agents that propose, running on demand or on a schedule, plus workflows on a schedule or a trigger and a copilot that builds them from a description; you approve before anything goes live.

Agent access: Claude, ChatGPT, Gemini, Cursor, Codex, Gemini CLI, and any MCP client, over OAuth or a token.

Safety default: each server defaults to Read & write; a Read only setting blocks every change, connections and accounts can be scoped, and your client is instructed to describe the change and ask for confirmation before the write tool runs. As with PaidSync, the check is your client's prompt.

Pricing: Pro $79 a month with 2,500 credits and one seat; Team $249 a month with 10,000 credits and 10 seats; Managed custom; trial credits without a card; a data operation costs one credit (Nov. 17, 2026).

Best for: running one server across every channel, with workflows on a schedule next to it.

PaidSync

Platforms: builds new campaigns on Google, Meta, LinkedIn, ChatGPT Ads, TikTok, and Microsoft; edits existing campaigns only on Snapchat, Reddit, Pinterest, and X; configures GA4, Tag Manager, and Merchant Center. PaidSync states it is a Google Partner and a TikTok API Platform Partner.

Reads: across the 14 platforms, with Search Console read-only.

Writes: creates campaigns, ad groups, keywords, ads, extensions, audiences, and experiments on Google; campaigns, ad sets, ads, creatives, audiences, lead forms, and automated rules on Meta; pauses, enables, and changes budgets wherever it writes. 630 or more tools.

AI: a chat workspace that applies changes through approval cards; nothing runs on a schedule.

Agent access: Claude, ChatGPT, Gemini, Grok, Microsoft Copilot, Perplexity, Cursor, Windsurf, n8n, and Codex, or any client with custom connectors.

Safety default: Google Ads creates and most other creates default to a dry run, while some GA4, Tag Manager, and creative-upload tools, including Meta image and video uploads, apply on the first call. Delete, pause, disable, and budget tools need a confirm flag, and over MCP the check is your client's confirmation prompt.

Pricing: Free at 100 tasks a month with no time limit; Pro $99, $149, or $199 a month at 1,000, 2,500, or 6,000 tasks; Team from $249 a month for 20 seats; a task is one read or one change.

Best for: getting the deepest write surface per platform, with the client as the gate.

Pipeboard

Platforms: 11 connectable ad platforms; its own guide names campaign changes on Meta, Google, TikTok, Snap, Reddit, LinkedIn, and Pinterest. Pipeboard is a badged Meta Business Partner.

Reads: reporting and analysis across the connected accounts, including GA4, Search Console, and Merchant Center.

Writes: creates campaigns, uploads creatives, and updates budgets; on Google, creates and edits campaigns, keywords, and bids. Duplication and bulk operations arrive on the paid plans.

AI: none of its own, and nothing runs on a schedule; MCP tools only, with reusable prompts and n8n as an assistant option.

Agent access: Claude, ChatGPT, Claude Code, Cursor, n8n, and Codex. Platform credentials stay server-side; a token can be deleted or the platform connection revoked at any time.

Safety default: a read-only token for reporting and a write token scoped to specific tools from the Pro plan, with scoping by ad account from the Agency plan; on Google, the campaigns it creates start paused.

Pricing: Free with two account slots and 30 AI executions a week; Pro $29.90 a month for three slots; Premium $99 for 10 slots with bulk operations; Agency $199 for 25 slots with per-account token scoping.

Best for: making one-off changes across many platforms from one connection, with the scope set on the token.

Ryze AI

Platforms: Google Ads and Meta through the connector, with GA4 reads; Microsoft, TikTok, LinkedIn, Reddit, Snapchat, and ChatGPT Ads as platforms it manages.

Reads: Google Ads, Meta Ads, and GA4 at once, deduplicated across accounts.

Writes: pauses campaigns, updates bids, and adds keywords from chat; schedules actions to run later.

AI: yes; agents that propose. Ryze's own agents monitor campaigns around the clock, file suggestions you approve or deny, and launch campaigns from a chat or a brief.

Agent access: Claude as an official connector, added in one click; ChatGPT and Cursor shown as clients.

Safety default: write actions need your approval before anything changes.

Pricing: the connector is included on every plan; Paid Ads is $89 a month with unlimited ad accounts, Agency $279 a month; 7-day trial; prices do not move with ad spend.

Best for: connecting Claude to Google and Meta in one click, then pausing, bidding, and adding keywords with an approval each time.

Which server fits which situation

The situation Where to look
Make one-off changes on one platform The platform's own server: Meta's for Meta, TikTok's for TikTok, Google's for reads on Google, Snap's for reads on Snapchat
Make one-off changes across many platforms Birch, Pipeboard, PaidSync, Markifact
Run an agent that proposes changes for approval Birch, Adspirer, GoMarble
Run rules on Meta from a rules engine behind the server Birch, Madgicx
Write, audit, and run always-on rules on Meta, Google, TikTok, and Snapchat Birch
Run rules on Google Birch; Google's own server does not change campaigns, budgets, or bids

The situation: Make one-off changes on one platform

Where to look: The platform's own server: Meta's for Meta, TikTok's for TikTok, Google's for reads on Google, Snap's for reads on Snapchat

The situation: Make one-off changes across many platforms

Where to look: Birch, Pipeboard, PaidSync, Markifact

The situation: Run an agent that proposes changes for approval

Where to look: Birch, Adspirer, GoMarble

The situation: Run rules on Meta from a rules engine behind the server

Where to look: Birch, Madgicx

The situation: Write, audit, and run always-on rules on Meta, Google, TikTok, and Snapchat

Where to look: Birch

The situation: Run rules on Google

Where to look: Birch; Google's own server does not change campaigns, budgets, or bids

Before you connect any of these, check what each one can see and what its vendor keeps. Ad account data safety when you connect an AI compares 11 tools on what they collect, train on, and can touch.

The 6 questions before you connect

Six questions settle most of the choice before a server is connected. Worth sending them to the vendor, and keeping the answers next to the consent screenshot.

  1. What can it write, by platform: campaigns, budgets, bids, status, rules, or nothing?
  2. What does the consent screen expose: one account, selected accounts, or everything under the portfolio or manager account?
  3. What runs when nobody is in the chat, and is it a rule a person can read, an agent that proposes, or an agent that acts?
  4. How is a change attributed in the platform's own history and in the vendor's log?
  5. How do you revoke access, and does revoking stop what is already scheduled?
  6. What does it cost per call, per task, or per account, and does the price move with ad spend?

The answers to the first and third are what separate the two jobs. A server that answers "campaigns and budgets" to the first and "nothing" to the third is a one-off-change server, however many platforms it covers. One that answers "rules" to the first and "the rules" to the third is the automation layer, and how that automation is built is the same whether a person or an agent writes it.

FAQs

Can one MCP server manage Meta, Google, and TikTok together?
Plus sign

For one-off changes, yes: Pipeboard, PaidSync, Markifact, and Adspirer all write to Meta, Google, and TikTok from one connection, with the write depth varying by platform. For standing rules on all three plus Snapchat, Birch MCP is the server that creates and edits them from your client; the platforms' own servers each cover one platform.

Is Meta's Ads MCP official, and is Google's really read-only?
plus sign

Meta's is Meta's own server, connected with your Ads Manager login. Google's is Google's own, open source and self-hosted, and its docs now list a small write surface: assets and labels, each proposed and confirmed. It still cannot create, modify, pause, or enable campaigns, ad groups, ads, budgets, or bids, and a config switch makes it fully read-only.

Is it safe to let an AI change ad spend?
black plus sign

Every write server here puts something between the request and the account: paused creation and a confirmed activation on Meta, drafts on AdKit, token scope on Pipeboard, a dry run on PaidSync, a preview and dry run on Birch. The risk sits in what is allowed always in your client and in what runs unattended. The Meta side is in Meta ads MCP: what AI agents can and can't do in your ad account; the data side per vendor is in Ad account data safety when you connect an AI.

Do ad MCPs charge per tool call, and are there free ones?

Meta's, TikTok's, and Snap's servers are free, and Google's is free to run on your own hosting. Pipeboard, Adspirer, and PaidSync have free tiers metered in executions or tasks a month; AdKit prices flat with unlimited calls; GoMarble meters credits; Ryze AI includes the connector on every paid plan; Madgicx and Birch include the server in a subscription priced by ad spend. Prices as of Nov. 17, 2026, are in each entry.

Do I need Birch MCP if I already have Meta's?

They do different jobs. Meta's server makes one change per prompt and has no scheduling or condition-watching tool; Birch MCP creates and edits rules that keep running on schedule, on four platforms, and logs every run. Teams that want both connect both, and inside Birch AI the two already sit together.

MCP server or API: what's the difference for an ad account?

The API is the platform's interface for code. An MCP server wraps it in tools a chat client can call, with its own consent screen, scope, and confirmation steps. Connecting a server is a permission decision: the questions above are about the server, and the API underneath stays whatever the platform allows. For rules your agent writes and Birch runs the same way every time, Birch MCP connects in under a minute.

Daria Bochkareva

Daria works where product, marketing and user insight meet. At Bïrch she turns what people actually need into messaging, positioning and product direction.

Get started with Birch product

14
days free trial
Cancel anytime
No credit card required