Buyer guide
Best MCP servers for Meta Ads and Google Ads in 2026
Meta and Google now run their own MCP servers, and four hosted servers connect the same assistants to ad accounts with more platforms or more work done. The best one depends on the job: Meta's connector for catalogs and signals, Google's for read-only GAQL, Adrails for agents, automations and store data on Meta and Google Ads, Pipeboard and Adspirer for breadth across platforms, GoMarble for monitoring agents with many connectors, Ryze AI for an ads service with a connector included, Windsor.ai for many data sources with platform writes, and Polar Analytics for Shopify analysis that changes nothing.
At a glance
Nine MCP servers for ad accounts
Every server below speaks remote MCP, so Claude, ChatGPT and Claude Code add it by address. The rows are what differs.
| Server | Platforms | Runs | Writes | Price |
|---|---|---|---|---|
| Meta Ads AI Connectors | Meta | Hosted by Meta | Yes, created objects land paused | Free in the open beta |
| Google Ads MCP server | Google Ads | You run it, locally or on Cloud Run | No, read-only | Free, open source |
| Adrails | Meta Ads, Google Ads | Hosted | Prepared, applied in Adrails | From $49 a month, unlimited ad accounts |
| Pipeboard | Meta, Google, TikTok, LinkedIn and more | Hosted | Yes, scoped by token | Free, then $29.90 a month |
| GoMarble | Meta, Google, TikTok, Microsoft Ads | Hosted, with an open source server | Monitoring agents, MCP reads | Published on its pricing page |
| Adspirer | Six ad platforms | Hosted | Yes, as agent tasks | Free tier, then $40 a month |
| Ryze AI | Google Ads, Meta, GA4, Search Console, TikTok, Shopify | Hosted | Yes, write tools gated in the client | Connecting free; included in its Paid Ads service from $89 a month |
| Windsor.ai | Meta, Google Ads, TikTok, LinkedIn, Microsoft Ads; 350+ sources, Shopify read only | Hosted | Yes, created paused; spend increases confirmed by token | Free tier, then from $19 a month billed yearly |
| Polar Analytics | Shopify, Meta, Google Ads, Klaviyo, Amazon; 45+ connectors | Hosted | No writes to ad platforms or the store | Sized on store GMV; Core plan $750 a month on the Shopify App Store |
Checked on the vendors' pages and documentation on 4 October 2026, and on 8 November 2026 for Ryze AI, Windsor.ai and Polar Analytics. Prices are the entry price each vendor publishes; beta and free tiers may change.
01
1. Meta Ads AI Connectors, for catalogs, signals and native writes
Meta's own MCP server at mcp.facebook.com/ads launched on 29 April 2026 as an open beta. Its 29 tools cover campaign creation and updates, account and Page lookups, ten catalog tools, four dataset quality tools and seven insight tools with benchmarks and an anomaly signal. You sign in with your Facebook Business login and choose a consent level; everything it creates lands paused.
Pick it when the job is inside Meta and nowhere else: a feed error, a product set, pixel and dataset quality, or a change you want written straight away. It has no schedule, no cross-platform view and no store data.
02
2. Google's Google Ads MCP server, for read-only GAQL
Google's open source server exposes three tools: list the accounts your credentials reach, run a GAQL query, and read a resource's fields. It is read-only in its current release, written in Python, and runs over stdio on your machine or on a Cloud Run service you deploy. Developer tokens are no longer required according to Google's documentation.
Pick it when a developer can set it up and the team only needs to read, especially fields no packaged tool covers. It builds nothing, edits nothing and looks up no keyword volumes.
03
3. Adrails, for agents, automations and store data on Meta and Google Ads
Adrails runs one hosted server at adrails.ai/api/mcp for Meta Ads and Google Ads, with 78 tools: the agents' own reports, Keyword Planner volumes, campaign builds and edits, pauses and budget moves. Shopify, Stripe, Klaviyo, Triple Whale, PostHog, RevenueCat and Google Drive are read next to spend. Every change is prepared as a proposal and applied in Adrails after it reads the platform again, with an undo. Scheduled and threshold automations keep running when no chat is open.
Pick it when the team runs several accounts for clients or stores, wants revenue next to spend, and wants changes reviewed before they happen. It covers two platforms, not six, and has no catalog tools.
04
4. Pipeboard, for breadth across ad platforms
Pipeboard connects Claude, ChatGPT, Cursor and other clients to Meta, Google, TikTok, LinkedIn and more platforms through one hosted MCP server, with a free plan and tokens scoped per permission. It is the widest platform list of the hosted servers here.
Pick it when the team buys media on platforms Adrails does not cover and wants one connector for all of them. It brings no agents or automations of its own, so the work happens in the chat.
05
5. GoMarble, for monitoring agents with many connectors
GoMarble runs AI agents that watch paid media accounts across Meta, Google, TikTok and Microsoft Ads, with more than 80 connectors, and exposes MCP access from Claude and ChatGPT alongside an open source server. Its strength is the number of sources its agents read from.
Pick it when monitoring across many sources matters more than building campaigns from the chat, and check on its pricing page which plan includes MCP access.
06
6. Adspirer, for an agent that lives in the assistant
Adspirer is an MCP server for Claude, ChatGPT, Cursor and other clients across six ad platforms, priced by tasks with a free tier, then $40 a month. It positions itself as the agent inside your assistant rather than a workspace.
Pick it when you want to stay in the chat and pay per task. There is no workspace of campaign tables, split tests or automations around it.
07
7. Ryze AI, for an ads service with a connector included
Ryze AI describes itself as "AI runs your ads, SEO, and website", a service sold for a flat monthly fee, and includes a hosted MCP connector at connector.get-ryze.ai/mcp with an OAuth sign-in through Google or Meta. Its developers page lists tools for Google Ads, Meta Ads, GA4 and Search Console, says the same endpoint covers TikTok, Shopify and more, and counts over 150 tools in total: 8 read and 6 write tools on Google Ads, 9 read and 3 write on Meta. It names Claude, ChatGPT in Developer mode, Claude Code, Cursor, Windsurf and VS Code as clients.
Write tools are gated in your client before anything changes. Its pricing page says connecting is free and lists the Paid Ads plan at $89 a month with a 7-day trial, the Claude and ChatGPT integration included on every plan. Pick it when you want a service that runs ads and SEO with a connector in the same subscription. The pages we read do not say whether what it creates on a platform starts paused.
08
8. Windsor.ai, for many data sources with platform writes
Windsor.ai connects more than 350 data sources and runs a remote MCP server at mcp.windsor.ai, signed in with OAuth or an API key. Its documentation lists 18 tools, and writes go through one action tool across Meta, Google Ads, TikTok, LinkedIn, Microsoft Ads, Instagram, Google Business Profile, Klaviyo and Amazon Seller Central: enabling and pausing campaigns, ad groups and ads, setting budgets, and on Meta and Google Ads creating campaigns, ad sets and ads. Shopify is a source it reads, with no write actions.
Campaigns, ad sets and ads it creates on Google Ads and Meta start paused, and an action that starts or raises spend first returns a preview with a confirmation token, run only on a second call. Its pricing page has a free plan, a 30-day trial of the full features and then one user and one source, and Basic at $23 a month, or $19 a month billed yearly, with MCP and write actions on every plan. Pick it when the team wants reporting data from many sources in the assistant and the same server to make platform changes.
09
9. Polar Analytics, for Shopify analysis that changes nothing
Polar Analytics calls itself "the first headless BI built for commerce". Its MCP server at api.polaranalytics.com/mcp is an official Claude connector and also works in Claude Code, Cursor, ChatGPT and other clients, over 45 connectors including Shopify, Meta, Google Ads, Klaviyo and Amazon. Its page lists 37 tools, among them generate_report, get_metrics and fetch_live_connector_data.
Polar's page is plain about writes: "Nothing the agent does reaches your sources". It changes no campaign, budget or store; what it can create, update and delete are custom metrics, dimensions and labels inside Polar. The pricing page sizes plans on the store's gross merchandise value without listing amounts, and the Shopify App Store lists a Core plan at $750 a month. Pick it when the job is analysis of a Shopify brand's data across sources and nobody needs the assistant to change the ads.
10
How we compared the servers
Each server was checked on four questions: which platforms it reaches, who runs it, what happens to a write, and what it costs to start. The answers come from each vendor's own pages and documentation on the review date, listed in the sources. Where a vendor does not publish a figure, the table says so instead of estimating one.
Adrails is our product, so the recommendation for it is ours. The other eight are described by what their pages state, and the jobs we point them to are the ones their tools cover and ours do not. Ryze AI, Windsor.ai and Polar Analytics were added on 8 November 2026 from their own pages, listed in the sources with that date.
Read more
- Adrails vs Pipeboard vs AdspirerCompare documented platform support, access controls and advertising review workflows.
- ChatGPT for Meta Ads through MCPCheck ChatGPT plan access, Meta reporting tools and how advertising proposals are reviewed.
- Meta Ads MCP serverRead Meta results and prepare campaign changes for review in Adrails.
- Google Ads MCP serverResearch keywords, read Google Ads results and prepare campaign builds and edits.
- Meta's connector or AdrailsMeta's official connector and Adrails, tool by tool, and when to add both.
- Google's MCP server or a hosted oneGoogle's read-only server and a hosted one that researches, builds and edits.
- Connect an AI clientClaude, Claude Code, ChatGPT and Codex, step by step.
Sources checked
- Meta: Introducing Meta Ads AI Connectors
- mcp.directory: the 29 tools of Meta's Ads MCP and CLI
- Google Ads API: MCP server
- googleads/google-ads-mcp on GitHub
- Pipeboard pricing
- GoMarble MCP
- Adspirer pricing
- Ryze AI: developers and MCP connector, 8 November 2026
- Ryze AI pricing, 8 November 2026
- Windsor.ai MCP server documentation, 8 November 2026
- Windsor.ai write capabilities, 8 November 2026
- Windsor.ai pricing, 8 November 2026
- Polar Analytics MCP, 8 November 2026
- Polar Analytics pricing, 8 November 2026
- Polar Analytics on the Shopify App Store, 8 November 2026
- Adrails MCP tools reference
Common questions
What is an MCP server for ads?
A server that speaks the Model Context Protocol and exposes tools on ad accounts, so an AI client such as Claude, ChatGPT or Claude Code can read results and prepare or make changes from a conversation. You add it by address and sign in.
Should I use the official Meta and Google servers or a hosted one?
Both, often. The official servers reach what their platform exposes and nothing else: Meta's writes and opens catalogs, Google's only reads. A hosted server adds platforms, store data, Keyword Planner, agents or a review step. Claude and ChatGPT accept several at once.
Can an MCP server change my ads without me?
Depends on the server. Meta's connector writes through the Marketing API as asked, with the client's permission prompt as the review. Adrails returns a proposal that an owner or admin applies in Adrails. Windsor.ai asks for a second call with a confirmation token before it starts or raises spend. Google's server and Polar Analytics change nothing.
Do these servers need API keys or developer tokens?
The servers here sign in with OAuth or a vendor login; Windsor.ai and Polar Analytics also accept an API key. Google's server needs an OAuth client or a service account you create in Google Cloud, and no developer token since 9 September 2026. Meta's connector and the hosted servers need no developer app.