Best MCP Servers for Marketing in 2026, Checked and Compared

The best MCP servers for marketing, in short

The best MCP servers for marketing are mostly the official ones from platforms you already pay for: Google’s Google Analytics and Google Ads servers, Meta’s Ads MCP server, HubSpot’s remote MCP server and Klaviyo’s server. Organic search is the gap, because Google doesn’t publish a Search Console MCP server, so add a third-party Search Console and Bing server. Connect three or four per task, start read-only, and give write access to ad accounts last.

We make GSC MCP Server, a hosted, read-only Search Console and Bing MCP server, so we have a side in this. It is one row in the table below; everything else here is another company’s product. This is a comparison of documented capabilities, not a hands-on test: every vendor detail comes from the vendor’s own documentation, repository or announcement, checked on 30 October 2026, and we left out anything we couldn’t verify there.

Many lists of MCP servers for marketers are already stale: HubSpot’s server left beta with write access in April 2026, Meta launched an official ads server the same month, and Google’s Ads server is still read-only. The quick picks:

• Organic search: a Search Console and Bing MCP server (ours, or one of the alternatives we compare elsewhere).
• Web analytics: Google’s Google Analytics MCP server (official, experimental, runs locally).
• Paid search: Google’s Google Ads MCP server (official, read-only).
• Paid social: Meta’s Ads MCP server (official, open beta, reads and writes). LinkedIn Ads only through third parties.
• SEO research: Ahrefs, Semrush or DataForSEO.
• CRM and email: HubSpot and Klaviyo, both official.
• Site and team tools: WordPress.com, Webflow, Shopify’s AI connectors, Notion, Slack and Zapier.

New to MCP? Start with our plain-English guide to MCP servers.

Search and SEO data: Search Console, Bing and the suites

Unlike analytics and paid search, organic search has no official MCP server from the platforms. We’re not aware of a Search Console MCP server from Google or a Bing Webmaster Tools server from Microsoft, so every option is third-party code calling the public APIs.

GSC MCP Server (ours)

GSC MCP Server, a Search Console and Bing MCP server, is hosted and reads Google Search Console and Bing Webmaster Tools. You sign in to each with OAuth (Google’s read-only webmasters.readonly scope), with no Google Cloud project or API key. Its 12 tools are all read-only: query_search_analytics returns clicks, impressions, CTR and position by query, page, country, device, date or search appearance, including Discover and Google News; compare_search_periods explains gains and drops; inspect_url returns Google’s index verdict and chosen canonical; get_bing_crawl_stats covers Bing crawling; and multicall runs up to 10 tools in one request. Plans are Free (one Google or Bing account, 30 tool calls a month), Pro at $9 a month (5 accounts, 10,000 calls) and Agency at $29 (20 accounts, 50,000 calls).

What it doesn’t do: no GA4, ads, Core Web Vitals or Google links data, and it can’t read Search Console’s generative AI or Page indexing reports, which are part of what the Search Console API doesn’t include. Those gaps are why the rest of this list exists. Open-source and other hosted alternatives are compared in the best Google Search Console MCP servers.

SEO data suites

If you want an MCP for SEO research beyond your own site, the suites answer outward-looking questions (competitors, backlinks, search volume, difficulty) that first-party data can’t. Two sentences each, checked on 30 October 2026:

• Ahrefs MCP: official and remote only; Ahrefs’ getting-started guide says local setup is no longer supported and access starts on the Lite plan. Every call costs at least 50 API units from the same pool as the Ahrefs API.
• Semrush MCP: a remote server at mcp.semrush.com that signs in with OAuth 2.1 or an API key and exposes the SEO API, the Trends API and read-only Projects data, per the Semrush MCP docs. Requests use API units, included with some Semrush One and SEO Classic plans or bought as a package.
• DataForSEO MCP: the official version 3 server runs hosted with OAuth or locally with your API login. There’s no subscription: the minimum top-up is $50, new accounts get a $1 trial credit, and you pay per request.

How these compare with first-party data, what they cost in practice and how to use them together is covered in Search Console MCP vs Ahrefs and Semrush MCP.

Web analytics: the Google Analytics MCP server

Google publishes an official Google Analytics MCP server. The README is titled “Google Analytics MCP Server (Experimental)”, and on 30 October 2026 the repository had about 3,600 GitHub stars, an Apache 2.0 licence and a 0.8.0 release dated 22 October 2026. It installs from the analytics-mcp package and runs locally through pipx.

What it can do, according to the README and Google’s guide:

• List accounts and properties, property details, Google Ads links and annotations.
• Run standard reports through the Data API (run_report), plus funnel and realtime reports.
• Read a property’s custom dimensions and metrics.

Setup is the developer kind. You need a Google Cloud project with the Google Analytics Admin API and Data API enabled, Application Default Credentials (your own Google login through gcloud, or a service account) with the analytics.readonly scope, and at least Viewer access on the property. Google’s MCP guide for Analytics gives configurations for Claude Desktop, Claude Code, Cursor and Antigravity. Because it runs over stdio on your machine, it isn’t a URL you can paste into Claude on the web or ChatGPT.

The guide says the server is read-only. The README adds one opt-in exception, create_property_annotation, which needs the --allow-write flag and the analytics.edit scope. The server is free; normal Data API quotas apply.

GA4 and Search Console measure different things, which we explain in Search Console vs Google Analytics 4.

Paid search: the Google Ads MCP server

Yes, there is an official Google Ads MCP server. Google documents it on the Google Ads API developer site and points to the googleads/google-ads-mcp repository (about 1,100 stars, Apache 2.0, release v0.0.5 on 9 October 2026). Older guides point to a repository under google-marketing-solutions; Google’s current docs don’t.

The key fact for marketers: Google’s page says “This implementation is strictly read-only.” It can’t change bids, pause campaigns or create assets. It has three tools: list_accessible_customers, search (which runs Google Ads Query Language, GAQL, for metrics, budgets and status) and get_resource_metadata (which describes the fields a report can use).

• Setup: a Google Cloud project with the Google Ads API enabled and Explorer, Basic or Standard API access; an OAuth client or Application Default Credentials with the adwords scope; and GOOGLE_ADS_LOGIN_CUSTOMER_ID if you work through a manager account. Google’s guide says the latest version no longer needs a developer token.
• Where it runs: locally over stdio through pipx, or on Cloud Run that you deploy yourself, which gives you an OAuth-protected URL to add as a custom connector in Claude or ChatGPT. A Google-hosted endpoint exists only as a pilot for allowlisted participants.
• Cost: free; you pay for Cloud Run if you deploy there.

The model writes GAQL for you. Ask it to show the query it ran, so you can check the date range and filters before you trust a number.

Paid social: Meta Ads and LinkedIn

Meta Ads MCP server (official)

Meta introduced Meta Ads AI connectors in open beta on 29 April 2026: an MCP server for advertisers and a command-line tool for developers. According to Meta’s Ads MCP server overview, the server is Meta-hosted at https://mcp.facebook.com/ads, works with any MCP-compatible agent, and needs no developer credentials, API setup or code.

Its tools fall into seven groups: reporting, ad creation and management, catalogue creation and management, signals and datasets, help and troubleshooting, A/B tests and conversion lift studies, and activity logs. That means it writes: it can create and edit campaigns, ad sets and ads. On 16 July 2026 Meta added MCP server rules, which let anyone with full control of a business portfolio govern what AI agents can do on an ad account. Set those rules before you connect, not after. Meta’s pages don’t mention a fee. Community Meta Ads servers still fill older roundups, but the official one signs in through Meta rather than a token you create and store.

LinkedIn Ads: no official server

LinkedIn’s Marketing API documentation doesn’t document an MCP server, and its Marketing APIs require approved developer access and member (three-legged) OAuth. So every LinkedIn Ads MCP route is third-party:

• Windsor.ai offers a LinkedIn Ads connector for Claude that signs in with LinkedIn OAuth, reads campaign, lead and demographic fields, and has 12 write actions, including pausing campaigns and setting budgets, with a confirmation step before each change. Its free plan covers one data source, one account and the last 30 days; blending sources starts at $19 a month.
• Open-source servers exist, but you need your own approved LinkedIn developer app to run one. Before installing any of them, check that it uses LinkedIn’s official API rather than a logged-in browser session.

CRM and email: HubSpot and Klaviyo

HubSpot remote MCP server

HubSpot’s changelog says its remote MCP server became generally available on 13 April 2026 for all HubSpot accounts, with write access added. It lives at https://mcp.hubspot.com and requires OAuth with PKCE. Per the integration docs:

• Reads CRM records, activities (calls, emails, meetings, notes, tasks), blog posts, landing pages, campaigns, marketing email drafts and analytics, and conversations.
• Writes contacts, companies, deals, tickets, activities, and some content and campaign objects.
• Respects HubSpot user permissions, so a user can only see and change what they can in HubSpot. With Sensitive Data enabled, activities and conversations are blocked from the MCP server.
• Some tools are plan-gated: the AEO tools need Marketing Hub Professional or Enterprise.

HubSpot’s separate, local Developer MCP server is for building apps; marketers want the remote one.

Klaviyo MCP server

Klaviyo’s connection guide recommends the remote server at https://mcp.klaviyo.com/mcp (OAuth with dynamic client registration), with a local option that uses a private API key. It is only available to users with an Owner, Admin or Manager role, so you can’t limit it by giving the agent a low-privilege login. Use the switch instead: adding read-only=true to the URL (or READ_ONLY locally) disables every tool that writes to your account.

CMS, store and team tools

WordPress

WordPress.com runs an official MCP server at https://public-api.wordpress.com/wpcom/v2/mcp/v1 with OAuth 2.1 sign-in. It works on every WordPress.com site, free or paid, and on self-hosted sites connected through Jetpack. You enable it in your account settings and choose Full access or Read only. For other self-hosted sites, the WordPress project’s MCP Adapter (v0.7.1, October 2026) exposes the Abilities API that plugins and themes register, so what an agent can do depends on what you install.

Webflow

Webflow’s official MCP server is hosted and reads and writes broadly: elements, styles, components, CMS collections, assets, custom code and site analytics. Each authorisation covers one workspace, and an agent can only do what your user can do in the Designer. A few tasks need the Designer open with the Bridge App; it can’t create new site projects.

Shopify

Shopify publishes official AI connectors rather than a single server URL: a Shopify connector for Claude, a ChatGPT plugin, and connectors for Gemini, Grok, Perplexity and Copilot. In Claude it can create and update products, collections, inventory and percentage discounts; orders, customers and analytics are read-only, it can’t edit your live theme or store settings, and refunds stay blocked even with write access approved. There’s no extra Shopify fee. Our ecommerce SEO guide covers the search side of a store.

Notion, Slack and Google Drive

• Notion hosts an official server at https://mcp.notion.com/mcp with OAuth. It reads, creates and updates pages and databases, and works on every plan. Good for briefs, content calendars and campaign docs.
• Slack hosts an official server at https://mcp.slack.com/mcp. It searches and reads messages and canvases and can send messages. Only Marketplace and internal Slack apps can use it, and workspace admins approve them.
• Google Drive and Sheets have remote MCP servers from Google, but Google lists them as Developer Preview, which requires membership of the Google Workspace Developer Preview Program. Not one for most marketing teams yet.

Zapier for everything else

Zapier MCP reaches apps that have no MCP server of their own. You choose which apps and actions the agent may use, it’s included in every Zapier plan, and each tool call uses two tasks from your quota.

Marketing MCP servers compared

The side-by-side view, from each vendor’s own pages, checked on 30 October 2026. “Writes” means the server can change something in the platform.

Combining servers without tool overload

Every server adds tool definitions the model has to read and choose between, so ten at once wastes context and invites wrong tool choices. Three or four servers per task is plenty.

• One job, one set of servers. In Claude you switch connectors on per conversation from the + menu. In Cursor you can disable whole servers. In Claude Code, a project-scoped .mcp.json keeps ads servers in the ads repository only, and tool search is on by default, so full tool definitions load only when needed.
• Name the source in every step. “Search Console for clicks, Google Analytics for key events” beats hoping the model guesses.
• Give the join key. Landing page path, campaign name or UTM value: tell the model which field links the two sources.
• Keep outputs small. Ask for 20–100 rows, because clients cap tool output; our guide to Claude Code for SEO lists its limits.
• Batch where you can. Our multicall runs up to 10 Search Console and Bing calls in one request.

Which AI clients can connect remote servers?

Most servers on this list are remote (hosted at an HTTPS URL). The local ones, such as Google Analytics and the default Google Ads setup, only run in desktop and terminal clients.

Sources: Claude custom connectors, ChatGPT custom MCP servers, Cursor MCP and VS Code MCP servers. Step-by-step setup for each client is in how to connect Search Console to Claude, ChatGPT and Cursor; the same steps work for any remote server.

Security: scopes, read-only first and prompt injection

Marketing stacks hold customer records and spend money, so the bar is higher than for a docs assistant.

• Start read-only. Most servers have a switch: Klaviyo’s read-only=true, WordPress.com’s Read only level, Shopify’s separate write approval, Google Analytics without --allow-write. Google Ads and our server are read-only by design.
• Read the consent screen. analytics.readonly and webmasters.readonly can’t change anything. analytics.edit can. Google Ads has a single adwords scope, so the read-only guarantee comes from the server’s code, not the token. Don’t run a fork that adds write tools without reading it.
• Use the least-privileged login. HubSpot and Webflow act within your user’s permissions, so connect as a limited user, not a super admin.
• Treat marketing data as untrusted text. Form submissions, CRM notes, product reviews, ad comments, Slack messages and even search queries are written by strangers. Any of them can contain instructions aimed at your assistant. The risky mix is one session that reads private data, reads untrusted text and can send data or take actions. Claude Code’s docs warn that servers that fetch external content expose you to prompt injection.
• Don’t click “Allow always” on write tools. Anthropic’s connector guide warns against it, and Cursor asks for approval by default for good reason.
• Revoke cleanly. Disconnect in the AI client and in the platform (HubSpot connected apps, your Google Account’s third-party connections, Meta business settings).

Giving an agent write access to an ad account

Meta’s server can create and edit campaigns, so set its MCP server rules first: decide what the agent may change and what stays manual, such as budgets. Ask for a plan before any change, approve each write, and check the activity log afterwards. A prompt that keeps the first sessions read-only:

Starter stacks for founders, in-house teams and agencies

Solo founder

Keep it to the questions you ask every week: what changed in search, which page to improve next, and what to publish. A Search Console and Bing server answers the first two, a read-only CMS server shows what already exists, and nothing on this stack can spend money. Our weekly SEO routine for founders shows the prompts.

In-house team

The value is in joining sources: organic clicks against key events, paid terms against organic rankings, CRM leads against landing pages. Keep the ads server read-only and share to Slack only after a human has read the findings.

Agency

Multi-account access matters more than tool count. Google Ads works through a manager account with GOOGLE_ADS_LOGIN_CUSTOMER_ID, Meta through business portfolios, and Search Console and Bing through whichever logins hold client access. Our agency reporting workflow covers the monthly cycle.

If a read-only Search Console and Bing server fits your stack, the Free plan covers one account and 30 tool calls, enough to test the prompts above, and paid plans start at $9 a month.

Frequently asked questions