Author: brendonrowe

The No Module Named ‘mcp.server.fastmcp’ – Issue and Fix in Simple Terms

If you use Claude with a data connector — Google Analytics 4, Google Search Console, or any of the growing list of “MCP” tools that let AI pull your marketing data directly — you may have opened your app one day at the end of July 2026 and found a connector that worked perfectly yesterday now refuses to start.

You didn’t break anything. A behind-the-scenes update did. Here’s what happened, how to confirm it’s the same problem you’re hitting, and the one-line fix.

The error name

The technical error looks like this:

ModuleNotFoundError: No module named 'mcp.server.fastmcp'

You might not see that exact line inside your AI app — connectors often fail silently there, showing only a vague “failed to connect” or a red dot next to the tool’s name. The full message only appears when you test the connector directly (more on how to do that below).

The Issue Explained

Most AI data connectors are built on top of a shared piece of plumbing called the MCP SDK (Model Context Protocol Software Development Kit). Think of it as the standard set of wiring that lets AI tools talk to services like Google Analytics.

On July 28, 2026, that plumbing released a major new version — version 2.0. As part of the upgrade, it reorganized itself and removed a component (mcp.server.fastmcp) that a large number of connectors were quietly depending on.

Here’s the catch: many connectors don’t lock themselves to a specific version of that plumbing. So the next time they start up, they automatically grab the newest version available — the 2.0 one — and immediately crash, because the piece they were built to use is gone.

The important takeaway for you as a marketer: nothing in your setup is wrong. Your credentials, your property ID, your file paths — all fine. Your connector just got handed a newer, incompatible version of a dependency it didn’t choose. It’s a timing problem, not a you problem.

How to verify this is your issue

Before you spend an hour re-checking your Google credentials, confirm the problem is actually this one. Two quick signals:

1. Check the timing. Did the connector work fine until late July 2026, then suddenly stop — often after a restart, reinstall, or update? That is the most likely sign you have this issue.

2. See the real error. Instead of relying on the vague failure in your AI app, run the connector through a free testing tool called the MCP Inspector. Your command prompt tool can do this easily. This surfaces the true error message.

On Windows, open Command Prompt (press the Windows key, type cmd, hit Enter).

Do not run this in an alternative program like PowerShell, which can block these commands with a “running scripts is disabled” security message. Then run a command in this shape, swapping in your connector’s own start command:

npx @modelcontextprotocol/inspector [your connector's command here]

For my issue, with Google Analytics 4 connector, for example, that might look like:

npx @modelcontextprotocol/inspector uvx --from google-analytics-mcp ga4-mcp-server

A local web page opens (usually at an address starting with localhost:6274). Along with a slide bar activation with the word Connect in the prompt box. If the error that appears mentions No module named 'mcp.server.fastmcp', you’ve confirmed it — this article’s fix is for you.

How to fix it

The solution is to tell your connector: use the older, working version of the plumbing, not the broken new one. You pin it below version 2.0.

Step 1 — Test the fix first. In Command Prompt, run your connector’s command again, but add the instruction --with "mcp<2". Using the GA4 example:

npx @modelcontextprotocol/inspector uvx --from google-analytics-mcp --with "mcp<2" ga4-mcp-server

Open the localhost link, click Connect, and this time the connector’s tools should appear. That confirms the fix works before you change anything permanent.

Step 2 — Make it permanent in your config. Open the configuration file where your connectors are defined and add two lines to the connector’s args list: "--with" and "mcp<2". For a uvx-based connector, the result looks like this:

json

"google-analytics": {
  "command": "uvx",
  "args": [
    "--from",
    "google-analytics-mcp",
    "--with",
    "mcp<2",
    "ga4-mcp-server"
  ],
  "env": {
    "GOOGLE_APPLICATION_CREDENTIALS": "path\\to\\your-key.json",
    "GA4_PROPERTY_ID": "your-property-id"
  }
}

The only change is those two new lines. Everything else stays exactly as you had it.

Step 3 — Fully restart your AI app. Quit it completely (actually quit, don’t just close the window), then reopen it so it reloads the updated configuration. Your connector should now come up healthy.

A couple of things worth knowing

Different connectors, slightly different commands. The --with "mcp<2" approach works for connectors launched with uvx (a common Python tool). If your connector uses a different launcher, the exact wording may vary, but the goal is always the same: pin the MCP plumbing below version 2.0. The connector’s own documentation or GitHub page usually spells out the equivalent.

This is a temporary bridge. The connector makers are patching their tools to lock the version themselves. Once a connector releases an update that does that, you’ll be able to remove the --with "mcp<2" line and things will just work. Until then, the pin keeps you running.

The PowerShell gotcha (Windows). If you try these commands in PowerShell and see a “running scripts is disabled on this system” error, that’s a Windows security setting, not a real failure. Use Command Prompt instead, or add .cmd to the command (for example, npx.cmd ...). Note this only affects you running tests by hand — it doesn’t affect how your AI app launches connectors day to day.

Why this matters for marketing teams

As more of us wire AI directly into our analytics, search, and ad platforms, we inherit a bit of the software world’s fragility — including the occasional dependency break that has nothing to do with our own work. The good news is that these breaks tend to be diagnosable in minutes and fixable in one line, if you know where to look. Bookmark the MCP Inspector trick above; it turns a mysterious “connector failed” into a plain-English answer nearly every time.

How to Connect Claude and Google Analytics for Free

Agentic AI is shifting marketing from running campaigns to supervising autonomous systems.

Search is becoming an answer engine — discovery now happens inside AI, not just on results pages.

Real-time personalization is now the baseline expectation, especially for B2B buyers.

In healthcare, roughly 70% of organizations already use AI, and patients increasingly research care through AI summaries before they ever reach your site.

There are a lot of ways to connect Claude AI to Google Analytics, and most of the ones you’ll see advertised cost money. Paid connectors like Zapier, Windsor.ai, and Supermetrics all work, and if you want a no-code, someone-else-handles-the-plumbing setup, they’re fine. But you don’t need them. You can connect Claude and Google Analytics for free with three things: Python, a free Google Analytics account, and a free Google Cloud project.

I want to be upfront about who’s writing this: I’m Brendon Rowe, a marketer, not a coder. I’ve spent 14+ years in digital acquisition, SEO, and paid search, but I don’t write software. If the idea of installing Python and touching a command prompt makes you a little nervous, I get it — it made me nervous too. I still got the whole thing working in an hour, and this article walks through exactly how, plus the small snags nobody warned me about.

Where I Found the Method

I initially got the core instructions from a Rank Math SEO tutorial on YouTube. It’s a clear, follow-along walkthrough, and I’d recommend watching it before or alongside this article. I’ve embedded it here:

Rather than repeat the video step for step, I want to give you the tips I found useful that aren’t covered in it — the security prompts that stopped me cold, the one-line shortcut for the trickiest install, the naming confusion around “Google Cloud Storage,” and how to point Claude at more than one Google Analytics account once you’ve got the first one running.

Installing Python and UVX: The Part That Made Me Hesitate

The connection relies on a small local program — an MCP server (Model Context Protocol) — that acts as the bridge between Claude and the Google Analytics API. To run it, you need Python and a lightweight tool called uv (its runner command is uvx). Here’s what actually happened when I installed them.

You May Have to Give Your Security System Permission

Both Python and uv were initially blocked by my security software. My first instinct was to stop — an unfamiliar download getting flagged is exactly the kind of thing we all teach each other to avoid. So I want to be clear about what I found: after granting permission to install them, there were no issues with viruses or malware. Python and uv are mainstream, widely used developer tools, and the block was my security system being cautious about unrecognized installers, not a sign of anything harmful. If you hit the same wall, that’s normal. (As always, only download these from their official sources, which I’ve linked below.)

Installing uv the Easy Way

Installing uv turned out to be the easiest part, because you can do it entirely from the command prompt with a single line:
winget install astral-sh.uv
That’s it. Windows’ built-in package manager (winget) fetches and installs it for you — no browser downloads, no installer wizard, no clicking through options. If you’re on a Mac, the equivalent one-liner uses Homebrew (brew install uv). You can grab Python the same way, or from the official Python download page.

Why the Python Package Matters Here

It’s fair to ask why you’re installing Python at all if you’re not going to write any code, I was curious about it as well and did some research. The answer: the official Google Analytics MCP server is a Python program. Python is the engine it runs on, and uv is the tool that downloads and launches that program on demand. You never edit the code — you just need the runtime present so the server can start when Claude asks for it. Think of Python as the electricity and the MCP server as the appliance you’re plugging in.

Google Cloud Console, Not Google Cloud Storage: Clearing Up the Confusion

This tripped me up, and I’ve seen it trip up other people searching for the same thing, so it’s worth stating plainly. You’ll see this setup described as needing “Google Cloud Storage” or the “Google Cloud Store.” That’s a naming mix-up. Google Cloud Storage (GCS) is a specific product for storing files in buckets, and it has nothing to do with this. What you actually need is a free Google Cloud Console — the umbrella account inside the Google Cloud console that lets you switch on Google’s APIs.

Once you have that mental model straight, the connection is four moves:

  1. Create a free Google Cloud project. If you have a Google account, you already have access to the console. Spin up a new project — no billing or paid tier required for this.
  2. Enable the Google Analytics Data API. This is the switch that lets your project read GA4 reporting data. Search for it in the console’s API library and turn it on.
  3. Set up credentials. You have two paths. The simplest for a solo site owner is Application Default Credentials using your own Google login. For client work the cleaner path is a service account with a downloadable JSON key — a dedicated, non-human identity that isn’t tied to your personal login. Make sure to grab the email produced during the end of this step and save it in a notepad or word doc.
  4. Grant access inside Google Analytics — the step most guides skip. Creating a service account and downloading its key does not, by itself, give it any access to your data. You have to add the service account’s email address, which is provided when you set up the Data API, as a user on the GA4 property, under Admin → Property Access Management, with at least Viewer permission. Add the email produced in the last step as a marketer role. Miss this and every request fails, even though everything else looks correct.

After that, you add a short entry to Claude Desktop’s configuration file pointing at the MCP server and your credentials, restart the app, and Claude can talk to your GA4 data directly. The Rank Math video shows the config edit on screen if you want to follow along visually.

Using Claude AI and Google MCP Across Multiple Analytics Accounts

I didn’t want this just for my own site — I manage Google Analytics for clients, and the real payoff is being able to analyze and monitor their properties from the same setup. Adding another Google Analytics account comes down to two separate layers, and you usually have to touch both.

Layer 1: The MCP Config (the “Which Property” Layer)

If you’re running a local GA4 MCP server through Claude Desktop’s config file, the property ID lives there — usually as an environment variable or argument. To add another property, you add or adjust that value and restart Claude. If instead you’re on a managed or remote connector where you pick the property in a menu, there may be no file to edit at all — you just select or add the property in the interface. Make sure to save all code in the Config file before closing it and restarting Claude. If you have a bad syntax or line, then the entire config file will go back to a default setup.

Layer 2: Credential Access in Google Analytics (the “Is It Allowed” Layer)

This is the step people miss. Whatever identity your MCP server authenticates as has to be granted access to each new property:

  • Service account setup (a JSON key referenced in your config): copy the service account’s email and add it as a user with at least Viewer on the new property, under GA Admin → Property Access Management. That property can live under a completely different Google Analytics account — access is granted per property, so you don’t need a new credential for each client.
  • OAuth setup (you authorize with your own Google login): simpler — any GA property your Google account already has access to is reachable. You just point the config or connector at the new property ID.

In short: the config decides which property Claude looks at, and Google Analytics decides whether it’s allowed to. One without the other won’t work.

What Claude Can Do With Just Google Analytics

The Rank Math video also demonstrates connecting Google Search Console alongside Google Analytics, and pairing the two is genuinely useful for organic-search questions. But there are plenty of tasks Claude handles faster with Google Analytics alone — no second connection to set up or manage. Once you’re connected, you can ask Claude to pull and interpret:

  • Traffic & acquisition — sessions, users, channels (organic, paid, direct, referral), and campaign performance.
  • Content performance — top pages, engagement time, and bounce/engagement rates by page.
  • Conversions & events — goal completions, form submissions, and key events.
  • Audience — geography, device and platform breakdowns, and demographics.
  • Ecommerce — if configured: revenue, transactions, and product performance.
  • Attribution — which channels and campaigns actually drive conversions.

Instead of exporting rows into a spreadsheet and building the pivot yourself, you ask a plain-English question — “which landing pages lost the most organic traffic month over month?” — and Claude queries the data and answers. For anyone who does this reporting week after week, that shift alone is worth the afternoon of setup.

Helpful Resources for Using Claude With Google Analytics

If you want to go deeper, these are the credible, first-party sources I’d start with:

AI Marketing Trends 2026: What Healthcare Marketers Need to Know

Agentic AI is shifting marketing from running campaigns to supervising autonomous systems.

Search is becoming an answer engine — discovery now happens inside AI, not just on results pages.

Real-time personalization is now the baseline expectation, especially for B2B buyers.

In healthcare, roughly 70% of organizations already use AI, and patients increasingly research care through AI summaries before they ever reach your site.

The AI marketing trends 2026 conversation has finally moved past the hype. After two years of pilots and experimentation, artificial intelligence is now the operating layer underneath modern marketing — planning campaigns, personalizing outreach, and increasingly answering the questions your customers used to type into a search bar. For healthcare and behavioral health organizations, where trust and accuracy carry more importance in the decision making than in almost any other category, understanding these shifts isn’t optional. This guide breaks down the trends that matter, the tools driving them, and what they mean for the way patients and buyers actually find you.

How AI Has Reshaped Marketing — From Experiment to Operating System

A few years ago, AI in marketing meant a chatbot on a landing page or an auto-generated subject line. Today it sits at the core of strategy and execution. Industry analysts frame 2026 as the year the “pilot era” ends and enterprise-wide transformation begins. As Gartner’s marketing predictions put it, AI agents are taking over many routine customer engagements — reorders, notifications, personalized guidance — which collapses traditional martech stacks and moves marketers into roles focused on supervising intelligent systems rather than running discrete campaigns.

The adoption numbers make the shift concrete. According to the 2026 B2B Trends Research Report, an overwhelming 96% of marketers now report using AI in their roles, and nearly half rank it as the single trend they’re most excited about. AI is no longer a differentiator you adopt early — it’s the ground floor everyone is building on.

New Marketing Roles Created by AI

As AI absorbs mechanical work — manual reporting, basic content formatting, routine A/B variations — entirely new job titles have emerged. Guidance from HubSpot’s 2026 marketing predictions and hiring data point to a cluster of roles that barely existed two years ago:

  • AI Marketing Specialist — owns the AI stack that runs personalization and analytics, making sure the tools actually deliver.
  • Prompt Engineer — crafts and refines the inputs that produce accurate, on-brand AI outputs.
  • Marketing Automation Manager — integrates AI workflows across platforms so the whole engine runs coherently.
  • GEO / AEO Specialist — a brand-new discipline focused on getting content surfaced inside AI answers rather than ranked on a results page. Every SEO on LinkedIn added “GEO” or “AEO” to their headline this year.
  • Data Storyteller — translates AI-generated insight into narratives that drive real business decisions.

How AI Keeps Evolving the Marketer’s Role

The through-line: none of these are “operate the AI” jobs, and none are “ignore the AI” jobs. They’re hybrid roles that pair machine speed with human judgment. Research reinforces that the winners aren’t the teams handing employees the most tools — Microsoft’s 2026 Work Trend Index found that organizational factors account for roughly two-thirds of AI’s real impact, more than twice the influence of individual behavior. In other words, redesigning how work happens beats simply buying more software. For marketing leaders, that means fewer people running discrete campaigns and more people orchestrating systems, guarding brand voice, and owning the strategic calls AI can’t make.

How AI Is Changing Digital Marketing From 2025 to 2026

If one theme defines AI in digital marketing trends for 2026, it’s the collapse of the old click-based model. Discovery is migrating away from the ten blue links and into conversational interfaces. Zero-click search is no longer confined to Google — as WordStream reports, it now stretches across ChatGPT, Perplexity, Gemini, Bing, and Meta AI, where users ask a question, get an instant synthesized answer, and often never visit a website at all.

That has consequences for trust, too. OpenAI’s early-2026 rollout of ads inside ChatGPT, flagged by Adweek, marked a pivotal shift — forcing consumers to distinguish organic AI recommendations from sponsored ones, and putting marketing on the front line of that ethical exposure.

New Capabilities in Today’s Top AI Marketing Tools

The AI marketing tools powering this era have grown from single-task helpers into orchestration platforms. Recent tooling analysis highlights how Adobe’s Experience Platform Agent Orchestrator now coordinates specialized AI agents to personalize website content, experimentation, and offers at scale, while HubSpot’s AI email tools dynamically adjust content, subject lines, and send times around each recipient’s engagement patterns. Predictive AI forecasts who’s likely to convert or churn; generative AI drafts the copy, images, and variations to act on it. The frontier now is agentic AI — systems that don’t just recommend but execute.

The Shift From Search Engines to Answer Engines

For content teams, this is the single biggest operational change. Search engines are becoming answer engines that extract and attribute rather than simply rank. Winning visibility now means writing clearly structured, factually verifiable, well-attributed content that AI systems can confidently cite. Refusing to optimize for AI-mediated search, as one widely shared case study showed, is like refusing to optimize for Google in 2010 — a slow, expensive way to lose share.

AI Personalization Trends B2B Marketers Should Watch in 2026

Among the AI personalization trends B2B marketers 2026 should track, the headline is simple: personalization has moved from a competitive edge to table stakes. Buyers now expect relevant, individualized touches at every stage of the journey. The 2026 B2B research found that 23% of marketers are using AI specifically to hone messaging and build campaigns that meet buyers where they are — moving beyond one-to-many blasts toward experiences that feel genuinely individual.

The mechanics are changing as well. Instead of starting with demographics and guessing at behavior, AI-driven segmentation lets clusters emerge from actual intent signals, content consumption, and buying-committee activity. That behavioral approach routinely surfaces buying patterns that firmographic-only models miss entirely — a meaningful upgrade for any organization with long, multi-stakeholder sales cycles.

How B2B Marketing Is Evolving From AI Innovation

The next step is autonomy. Per Adobe’s B2B digital trends research, most B2B organizations now expect agentic AI to manage at least half of their customer interactions in the near future. But there’s a hard prerequisite: unified data. Agents can’t orchestrate a buying-group campaign if marketing and sales data live in separate silos, and you can’t personalize for a group you can’t identify. In 2026, most “strategy” problems turn out to be data problems in disguise — fix the foundation before layering on advanced tactics.

AI Marketing Automation Trends 2026: Easier and More Sophisticated

The most practical of the AI marketing automation trends 2026 is that automation has quietly become both simpler to set up and dramatically more capable. What used to require a dedicated ops team stitching together triggers and rules is increasingly handled by agents that plan, schedule, launch, and optimize campaigns on their own. The scale of that shift is striking: one analysis cited by industry career research projects that roughly 84% of campaign setup and execution tasks will be automated in 2026.

That doesn’t erase the marketer — it relocates the work. Reports are automated; interpretation isn’t. Campaign optimization is automated; deciding where to invest remains a human responsibility. The teams that benefit most treat AI as an operating system with structured workflows, keeping humans firmly in the loop for judgment, accountability, and quality control.

How AI Is Transforming Healthcare Marketing

Healthcare has always had more at stake than most industries, and AI is raising those stakes fast. According to NVIDIA’s 2026 State of AI in Healthcare report, 70% of healthcare organizations now use AI — up from 63% the prior year — and up to 85% of them say it has already increased annual revenue, as summarized in this healthcare AI marketing analysis. The market backdrop is just as dramatic: Grand View Research valued the global AI-in-healthcare market at roughly $36.67 billion in 2025, projecting it toward $505 billion by 2033.

The bigger change is how patients behave. Nearly 70% of patients now begin their care journeys online, per data cited from the National Institutes of Health, and a majority of U.S. adults — around 65%, per Pew Research Center — now at least sometimes encounter AI summaries in their search results. For a treatment center or multi-location provider, that means the first impression is increasingly made inside an AI answer, before a prospective patient or their family ever clicks through to your website.

What This Means for Behavioral Health and Treatment Providers

Three priorities rise to the top for behavioral health, addiction treatment, and specialty providers:

  • Structured, credible content for AI discovery. Answer engines favor clear headings, structured data, and verifiable, expertly-reviewed information. E-E-A-T signals — real author credentials, clinical review, citations — are now visibility levers, not just quality checks.
  • Precision patient acquisition. Predictive models can identify who is likely to need a specific service and time outreach accordingly, improving both effectiveness and cost efficiency.
  • Conversational, always-on access. AI voice agents and chat are becoming standard patient-facing infrastructure. Clinics deploying AI-powered chat and appointment automation report a 20–35% reduction in no-shows and more after-hours inquiries.

The organizations pulling ahead treat AI as foundational infrastructure while keeping human oversight non-negotiable — a balance that matters far more in healthcare, where accuracy and trust are the whole game.

Putting the Latest AI Marketing Trends 2026 to Work

Pulling the latest AI marketing trends 2026 together, the roadmap for the year is clear enough to act on now:

  • Unify your data first — it’s the prerequisite for personalization, automation, and agentic AI alike.
  • Optimize for answer engines — structure content so AI systems can extract, trust, and cite it.
  • Build AI fluency into your team — the salary and performance gap between AI-fluent marketers and everyone else is already visible at every level.
  • Keep humans on strategy and trust — automate execution, but own the judgment, voice, and accountability.

None of this requires chasing every shiny tool. The most successful teams diagnose their own context, pick two or three high-impact moves, and execute them with discipline. That’s especially true in healthcare, where thoughtful, compliant, expertly-guided marketing beats speed-for-its-own-sake every time.