You’re a marketer trying to connect AI to your tools. Maybe you want Claude to pull campaign data from GA4, or ChatGPT to update your CRM. You start researching and immediately hit a wall of jargon: MCP servers, agent skills, function calling, tool use, custom integrations.
Which approach actually matters? And more importantly, which one should you care about?
We’ve spent months evaluating this. Here’s why we’re going all-in on MCP — and why we think marketers should pay attention.
Table of contents
Open Table of contents
Three ways AI connects to your tools
Right now, there are three main approaches for getting AI to work with your marketing stack. Each has trade-offs.
MCP servers are remote services that your AI connects to via a URL. Think of them like apps in an app store — you add them once, they update automatically, and they work across different AI platforms. HubSpot, Google Analytics, Notion, and dozens of other tools already offer official MCP servers.
Agent skills are locally installed packages that give AI specific capabilities. They live on your machine, need manual updates, and only work with the specific AI tool they were built for. If HubSpot updates their API, you’re waiting for someone to update the skill package too.
Native tool calling is when AI platforms build integrations directly into their product. OpenAI’s plugins were an early version of this. It’s convenient when it works, but you’re locked into whatever that platform decides to support.
Why MCP wins for marketing teams
Here’s what tipped the decision for us.
One standard, many AI tools. MCP works across Claude, ChatGPT, Gemini, and AI-powered IDEs. If you set up a HubSpot MCP server today, it works regardless of which AI tool your team prefers. Agent skills lock you into one platform. Native integrations lock you into another. MCP is the only approach that’s genuinely vendor-neutral.
Vendors own the integration. This is the big one. When HubSpot builds an official MCP server, they maintain it. When their API changes, they update the server. You don’t need to understand JSON schemas, token budgets, or protocol specifications — that’s the vendor’s job.
This matters more than most people realise. Getting MCP tool design right is genuinely difficult. There’s a hidden token cost to every tool an AI loads — poorly designed schemas can waste thousands of tokens before your AI even starts working. One example: a database server with 106 tools consumed 54,600 tokens just to initialise. The people best placed to solve this are the vendors building the tools, not end users trying to wire things together.
Auto-updating, no maintenance. Remote MCP servers update on the vendor’s side. You connect once and get improvements automatically. Agent skills need manual updates, version management, and local troubleshooting. For a marketing team, the less infrastructure to manage the better.
The ecosystem is already massive. We’re tracking official MCP servers from HubSpot, Google Analytics, Notion, Figma, Zapier, PostHog, Canva, and dozens more. The adoption speed has been unprecedented — faster than REST, GraphQL, or any previous protocol.
The honest trade-offs
MCP isn’t perfect. We’ve written about the challenges before, and they’re real.
It’s still early. Some tools don’t have MCP servers yet. The protocol is evolving quickly, which means occasional breaking changes. If you need something working today with a tool that doesn’t support MCP, you’ll need a fallback.
Security needs attention. Giving AI access to your marketing tools creates new considerations. Who can access what? What permissions does the AI have? The major vendors are being cautious here, rolling out gated previews with strict controls.
Not every tool needs AI access. MCP makes sense for data retrieval and automation — pulling analytics, updating records, scheduling posts. It doesn’t make sense for everything. Start with read-only access to your most-used tools before giving AI write permissions.
What this means for your team
You don’t need to understand how MCP works under the hood. What you need to know is this:
When evaluating marketing tools, ask if they have an MCP server. This is becoming a meaningful differentiator. Tools with official MCP servers will work better with AI assistants than those without.
Start with one integration. Pick the tool you query most often — probably your analytics platform or CRM — and connect it via MCP. See how it changes your workflow before adding more.
Don’t build custom integrations yet. If your favourite tool doesn’t have an MCP server today, wait. The ecosystem is moving fast enough that building a custom solution now will likely be wasted effort. The major vendors are all moving in this direction.
Think vendor-neutral. The AI landscape is shifting rapidly. Today’s favourite tool might not be tomorrow’s. MCP is the only integration approach that lets you switch AI platforms without rebuilding your connections.
The bottom line
We’re betting on MCP because it puts the integration burden where it belongs — on the tool vendors, not on marketing teams. It’s vendor-neutral, auto-updating, and already supported by the tools marketers actually use.
The marketing teams that figure out AI-powered workflows early will have a serious advantage. MCP is what makes that practical without needing a developer on staff.
Growth Method is the GrowthOS built for marketing teams focused on pipeline — not projects. Book a call to see how we can help accelerate your results.
“We are on-track to deliver a 43% increase in inbound leads this year. There is no doubt the adoption of Growth Method is the primary driver behind these results.”
Laura Perrott, Colt Technology Services