This article is part of our MCP Guide: what the Model Context Protocol is, how to use it safely, and how to choose servers, tools and help. Want a second opinion on your setup? Request a free MCP readiness review.
MCP lets a marketing team ask an AI assistant about GA4, Search Console, HubSpot and Google Ads in plain English, using live data. You skip exporting four reports into a spreadsheet. You ask "Which campaigns drove the most demo requests last month?" and the AI pulls the numbers through each tool's MCP server.
MCP (Model Context Protocol) is an open standard for connecting AI apps to other software. Each connection runs through an "MCP server," a small program that tells the AI what it can read or do in that tool. If that is new to you, start with What Is MCP?.
This guide covers the four servers most marketing teams ask about. We checked each one against the vendor's own docs on 28 October 2026 and note who publishes it.
Key Takeaways
- Google publishes MCP servers for GA4 and Google Ads, and both are mostly or fully read-only as of October 2026.
- We found no Google-published Search Console MCP server; the options we found are community-built and need your own review.
- HubSpot runs a hosted MCP server with read and write access to CRM, content and marketing email data, signed in with OAuth.
- Semrush and Ahrefs both offer hosted MCP servers, and calls draw on your plan's API units.
- Start with read-only reporting questions, then add write actions one at a time with a named owner.
What changes for a marketing team
Most marketing reporting is copy and paste. Someone pulls GA4 numbers, someone else pulls ad spend, and a third person checks the CRM for lead quality. MCP lets one AI conversation reach all three.
Typical uses:
- Weekly reporting. "Summarize sessions, conversions and spend by channel for last week, and flag anything that moved more than 20%."
- Campaign questions. "Which Google Ads campaigns had the highest cost per conversion in September?"
- Lead quality. "How many HubSpot contacts from the October webinar became deals?"
- Content gaps. "Which pages get impressions in Search Console but few clicks?"
The AI does the joining and the first draft. A person still checks the numbers and decides what to do. For more department ideas, see MCP Examples: 25 Real Business Uses.
The four servers at a glance
| Tool | Who publishes it | Access | How it signs in |
|---|---|---|---|
| GA4 | Google (googleanalytics GitHub org), labeled experimental | Mostly read; one optional write tool | Google Application Default Credentials |
| Search Console | Community projects only, in our search | Varies by project | Usually a Google service account |
| HubSpot | HubSpot (hosted at mcp.hubspot.com) | Read and write | OAuth with PKCE |
| Google Ads | Google (documented on Google for Developers) | Read-only | OAuth client or default credentials |
Each row is explained below.
GA4: the Google Analytics MCP server
The Google Analytics MCP server lives in the googleanalytics GitHub organization. Its title still says "Experimental," so expect changes.
What it can do, per the repo:
- List your accounts and properties, and show property details.
- Run standard reports, funnel reports and realtime reports.
- List custom dimensions and metrics, and linked Google Ads accounts.
- List property annotations, and optionally create them.
Creating annotations is the only write action. It needs a separate --allow-write flag and the analytics.edit scope. Everything else works with the read-only scope.
Setup is technical. You enable the Analytics Admin and Data APIs in a Google Cloud project, set up credentials with gcloud, and run the server locally with pipx. The repo documents setup for Gemini CLI and Claude Code. Google has since moved free and Google AI Pro and Ultra users from Gemini CLI to Antigravity CLI; paid API-key and enterprise users keep Gemini CLI. Plan for a developer or a technical marketer to do the first install.
Search Console: community servers only
As of 28 October 2026, we could not find a Search Console MCP server published by Google. The options we found are community projects that call the public Search Console API.
One widely listed example is ahonn/mcp-server-gsc. Its README documents a search analytics tool with up to 25,000 rows per request, filters by query and page, and a "quick wins" option. It signs in with a Google service account that you add to Search Console.
Because it is community-built, treat it like any open-source code:
- Have someone read the code before you connect a client's property.
- Give the service account the lowest Search Console permission that works.
- Pin a version so an update doesn't change behavior without warning.
Our MCP for SEO guide goes deeper on search workflows.
HubSpot: the remote MCP server
HubSpot runs a hosted MCP server at https://mcp.hubspot.com. Its developer overview describes "secure read and write access to your HubSpot CRM data."
According to HubSpot's integration guide, the server can:
- Read CRM records, activities, content, content analytics, conversations, marketing email data and AEO data.
- Write CRM records and activities, landing pages, website pages, blog posts, campaigns, marketing email drafts, segments and more.
- Respect each user's existing HubSpot permissions.
A few limits matter for marketers:
- AEO features need Marketing Hub Professional or Enterprise.
- If Sensitive Data is turned on, conversation and activity data are blocked from the MCP server.
- Some content tools, such as website page previews, are only available in select AI clients. HubSpot names Claude, ChatGPT and Gemini.
Sign-in uses OAuth with PKCE. You create an MCP connector in your HubSpot account and use its details in your AI app. HubSpot also runs a separate local Developer MCP server for building HubSpot apps; that one is for developers, not campaign work.
Google Ads: the official read-only server
Google documents the Google Ads MCP server on its developer site and links the code at github.com/googleads/google-ads-mcp.
Google says the current release is "strictly read-only." It cannot change bids, pause campaigns or create assets. Its tools are:
list_accessible_customers, to list the ad accounts you can reach.search, to run Google Ads Query Language (GAQL) queries for metrics, budgets and status.get_resource_metadata, to describe a resource such as a campaign.
You need a Google Cloud project with Google Ads API access and an OAuth client or default credentials. Google's page also says developer tokens were sunset on 9 September 2026 and are no longer required in the latest version. A hosted version is in a pilot for allowlisted participants.
Read-only is a good thing here. Your AI can explain spend and performance without any risk of changing a live budget.
Semrush and Ahrefs
Many teams also want competitor and keyword data in the same conversation. Both major SEO platforms run hosted servers.
- Semrush. Its MCP page says the server covers the SEO API, the Trends API and read-only Projects API methods. It signs in with OAuth 2.1 by default. Access depends on your plan, and each request uses API units.
- Ahrefs. Its MCP docs say the hosted server is available on Lite plans and up. Calls use API units from your monthly allowance, and new OAuth keys default to a 200,000-unit monthly cap per key.
We compare these and others in Best MCP Servers for Marketing and SEO Teams.
How to roll this out without surprises
A good first month looks like this:
- Pick one question. For example, a weekly channel report. Write down which numbers it needs.
- Connect read-only first. GA4 with the read scope and Google Ads are natural starting points.
- Use a test property or a sandbox CRM if you can. Check that the AI's numbers match the tool's own reports.
- Add HubSpot reads. Keep write tools off until the reports are trusted.
- Turn on one write action at a time. Drafting a marketing email is lower risk than editing a live landing page.
- Review access monthly. Remove people and servers that no longer need it.
Use our checklist for choosing MCP servers to score each server before step 2. If your company has an IT team, loop them in early. Our page on MCP governance for IT leaders covers what they will ask.
Common mistakes
- Trusting the first number. AI summaries can misread a date range or a filter. Spot-check against the source tool.
- Mixing client accounts. Agencies should keep each client's servers in separate workspaces or profiles.
- Leaving write access on "just in case." If nobody needs it this week, turn it off.
- Ignoring API costs. Semrush and Ahrefs calls use plan units, and an agent can make many calls fast.
Next step
Ready to compare specific tools? Read our roundup of the best MCP servers for marketing and SEO teams. If you want help wiring these into a reporting workflow, see our Claude Code consulting service.
Frequently asked questions
Can MCP replace my marketing dashboard?
Not fully. MCP is best for questions and first drafts, such as "why did conversions drop last week?" A dashboard is still better for numbers everyone checks daily. Many teams keep both and use the AI for the follow-up questions a dashboard can't answer.
Is there an official Google Search Console MCP server?
As of 28 October 2026, we did not find one published by Google. The servers we found are community projects that use the public Search Console API. Review the code and use a service account with limited access before connecting real properties.
Can the Google Ads MCP server change my campaigns?
No. Google says the current release is strictly read-only. It cannot change bids, pause campaigns or create assets. That makes it a safe way to let an AI explain performance.
Does HubSpot's MCP server work on free plans?
The HubSpot docs we checked note some free-tool limits, such as not creating custom pipelines, and say AEO features need Marketing Hub Professional or Enterprise. Check HubSpot's integration guide for your exact tier before you plan around a feature.
Which AI apps can use these servers?
Most work with any MCP-compatible client. The GA4 repo documents Gemini CLI and Claude Code. HubSpot names Claude, ChatGPT and Gemini for some content tools. See Which AI Apps Support MCP? for the full picture.
