Just stumbled upon this GA4 MCP tutorial video from Google. What puzzles me is to expect marketers to set up a full MCP project locally using python. No GA4 remote MCP server that you can install in one click. Damn, you could even use Gemini to bootstrap it (even though it would be more reliable and faster with Claude Code 😘 ) The Google analytics team must be cooking something here hopefully or maybe I missed it in the docs. It's still telling of how GA4 is treated as a second-class citizen in today's AI world.
tagstack.io
Technology, Information and Internet
The observability platform for your marketing data.
About us
One’s capability to analyze customers’ needs, tailor and optimize their experience on an website is today indispensable in order to win. The world of Analytics is in constant motion with fast innovations driven by more than 40K+ different marketing technologies that shape its landscape, rendering it challenging for all stakeholders to pick the right ones to optimize their ROAS. In this fast moving world, we, TagStack, will make it easier for you by giving you access to 2M+ GTM container contents, that will allow you to identify your prospects, size your missions and opportunities and in the end cut your sales cycle by 80%! Not only that, we will also allow you to perform fast audits or your websites in order to identify the areas of improvements and to be quick on the fixes. And guess what? Our product is online and we're opened to feedbacks! Any comment or repost will get you a 50% discount on our offers, so that you can give it a full test, while helping us make the product better.
- Website
-
https://tagstack.io
External link for tagstack.io
- Industry
- Technology, Information and Internet
- Company size
- 1 employee
- Headquarters
- Saint-Mande
- Type
- Self-Employed
- Founded
- 2024
Locations
-
Primary
Get directions
Saint-Mande, FR
Employees at tagstack.io
Updates
-
It's out, Tagstack MCP Server is now live, making it easier for all to audit and retrieve marketing technologies used by sites straight from Claude. It was an enlightening experience and I've condensed some takeaways I've gleaned along the way. It's available on the free plan if you want to give it a try. The cool thing is that it takes less than 5 minutes to set up. Excited to see what you'll build with it and I am also going to share some cool use cases and demos in the coming weeks. You'll find the blog post here if you want to learn more about what it took to build it: https://lnkd.in/es44utqE
-
I don't see it mentioned a lot here but when building your AI agent setup, keep in mind that using multiple cooperative agents will usually offer better performance than using a single agent. The most effective choice is to use different models and model providers. Use best performance model as the orchestrator and smaller models for performing tasks. I see three reasons for this: 1/ It's financially more efficient to evaluate the task complexity and dispatching it to a smaller model that might perform well at 1/10 the cost. 2/ If latency is an issue, using smaller models will definitely help. 3/ If done well, using smaller models with a bigger model can approach the performance of using only bigger models. What's your thoughts? Genuinely interested Anthropic, in its recently released System Card for Opus 4.5, confirmed those results in their conclusion of the multi-agent search scenarios (though they used their own test and benchmark for this task)
-
-
tagstack.io reposted this
MCP is nothing more than a new distribution channel for your app. Think of it as the App Store or a Facebook App. It's not a replacement for your API. What makes MCP great is not the protocol but the orchestration. You can call a bunch of MCP servers in the same chat session, giving them instructions so they coordinate to reach your goal. Like, you could call Tagstack's MCP server to audit a site, rework the JSON to extract the opportunity, then call the Gmail MCP to send an email to that customer. MCP as a channel is worth exploring for your app, even if your users aren't nerds—you can still build a great guide or onboarding flow to help them get started with your app in Claude. Maybe we'll even see the Zynga of 2025 emerge as an MCP-first startup, giving up on the frontend entirely to become infrastructure for the AI layer. Time will tell.
-
Budgets are tight and so companies are even more scared of hiring the wrong partner to manage their analytics pipelines. At least, that's the new use case that seems to emerge with tagstack: lots of companies are now checking their own site but also sites from agencies and freelancers. I guess, to assess their skills. Or maybe just to use tagstack as a ranking factor for the final decision. In any case, that's pretty enlightening and so if you work in the field, make sure your site reflects your expertise with a clean setup (no UA leftovers, no custom Javascript code that's reinventing how to read and write cookies for the hundredth time, good core web vitals). Because if you don't trust your own tracking, why should they?
-
I was skeptical about lifetime deals. One payment, forever support. Forever usage. Taking a step back, many products in the physical world work that way. The only difference, that matters, is the cost of the ongoing usage, something your refrigerator does not have. But at the end of the day, it's all economics. It depends on your customer usage patterns and the associated cost. For SaaS, that's where having the right infra makes this feasible, and potentially lucrative, versus a marketing gimmick that eats your margins. We launched lifetime plan for tagstack.io and saw some good success with it because it fits most user needs, in terms of limits, at the best possible price, one-time. What makes this possible for us is our use of Cloudflare as back-end. The generous limits coupled with ease of development and deployment make it feels like cheating. Having your economics right is the first step towards offering value to your customers, at a fair price. Don't take my words for it, check our lifetime plan for a real life example: https://lnkd.in/eyjuYnjd
-
Very revealing of the agency industry. We've collected a list of the top 2500 analytics agencies in the world. Among which, obviously, lots of Google Marketing Partners. Going through the list, I was surprised to see that maybe 1 site out of 20 is either: 1/ dead (domain name parking page 2/ has expired SSL certificate 3/ has a server issue preventing from loading the page (or maybe the site is just gone...) 4 Or the infamous wordpress login page Running agencies is harder than some might think and is definitely not an easy path. So... Kudos to all the default-alive agencies and freelancers out there helping companies with their data!
-
-
Surprise Wednesday! Now tagstack audits reports include a Core Web Vitals scorecard! We've also released a new mechanism to improve GTM container detection on page. Meaning that if you faced an issue in the past scanning a URL, give it a new try today! The scan may take some more time to load, working on making this faster. Share your feedback! PS: guess the website from the audit screenshot, 50% on your subscription for the winners, no cap.
-
-
Before starting tagstack, I thought losing access to your container was quite marginal among companies. I mean, I had experienced a couple times but not more really. Boy, how wrong was I. Dozen of customers actually come to tagstack to get back access to GTM containers they lost access to. I've tried to recap when that happens, how you can mitigate it and when tagstack can help. Let me know your thoughts :)
-