Get 25 battle-tested prompts
Turn real client conversations into Linkedin content that builds authority and attracts the right clients.
Turn real client conversations into Linkedin content that builds authority and attracts the right clients.
Every agency is going to have an MCP by the end of 2027.
That's a bold bet. I know. Let me try to back it.
I've spent the last two years on calls with founders of consultancies, implementation firms and agencies. Somewhere north of 900 of them. The conversation has shifted hard in the last six months, and the shift has a name most of those founders haven't heard yet. So this is the plain-language version of what's happening, why it matters to a firm that sells expertise, and what the early movers are already doing about it.
Figma, HubSpot, Slack and almost any tool you can think of are all building a way for you to use their software without opening their software.
The thing that makes that work is called an MCP. It stands for Model Context Protocol. It's an open standard that lets another program read a tool's data and take actions in it on your behalf.
My non-technical brain understands it like this: it lets the AI you're sitting in (Claude, ChatGPT, whatever you use) pull context from the other tools you work in, and even do work for you inside those tools.
A concrete example. You open Claude and type: "Create three Meta ads for the Q4 campaign using last month's best-performing copy, then launch them." Claude reads your ad account through the Meta MCP, writes the ads, creates them in Meta, and activates the campaign. You never opened Ads Manager.
If you've used Connectors in Claude, or Apps in ChatGPT, you've used one. Most people have without knowing the name.
One clarification, because I got this wrong the first time I tried to explain it. You don't sit in your email and pull HubSpot data in. You sit in an AI tool, and the AI reaches out to your email, your CRM, your design files, and everything else. The AI is the one doing the calling. That distinction matters later.
The reason is software fatigue. Nobody wants twelve tabs and twelve logins.
This isn't a feeling. A Harvard Business Review study tracked 137 people across three Fortune 500 companies and found the average person switched between apps and websites nearly 1,200 times a day. The switches alone added up to just under four hours a week.
So the tools are getting out of the way. Instead of being a destination you visit, each one is becoming a capability your AI can reach.
And this is no longer one vendor's experiment. In December 2025 Anthropic handed MCP to the Linux Foundation, with OpenAI, Google, Microsoft, AWS and Block backing the new foundation that governs it. ChatGPT, Microsoft Copilot and Gemini have all adopted it. When the companies that normally refuse to agree on anything agree on a plumbing standard, the plumbing wins.
Every tool is getting easier to reach.
Your firm (probably) isn't getting easier to reach. Or I should say, your expertise isn't.
It's still trapped. Trapped in your best consultant's head. Trapped in Slack threads from 2024 and old decks nobody can find. McKinsey estimated that knowledge workers spend nearly 20% of their week just looking for internal information or hunting down the colleague who has it. That was 2012. Inside most agencies I talk to, it hasn't improved. It's gotten worse, because there are more tools for the knowledge to hide in.
And from the client's side, it's worse again. Clients get an hour of your expertise at a time, on a scheduled call. Then not much for the rest of the week. They have a question on Tuesday afternoon. The answer exists. It's in your head, or your senior strategist's head, or a Loom from eight months ago. But the next call is Thursday, so they guess, or they ask ChatGPT, which has never seen a single thing your firm has learned.
That's the gap. Every piece of software a client uses is becoming available on demand inside their AI. Their most expensive advisor is still available by appointment.
Here's the part that took me a while to see.
In everything above, the agency is on the consuming side. You sit in Claude, you pull from Figma and HubSpot and Meta. You're the one calling the tools.
An MCP for agencies reverses that. Same protocol, other direction. Your firm stops being the one pulling context and starts being the context.
Your frameworks, your playbooks, your fifteen years of knowing what works for companies like this one, exposed as an MCP that your clients' AI can call. Not a portal they have to log into. Not a chatbot on your website. Your expertise sitting in their Claude, right next to their Figma data and their HubSpot data, answering questions the moment they come up.
This is the future we see coming, and it's not far off. Every agency builds an MCP that plugs into their expertise.
Say you're a GTM consultancy with a point of view on outbound. A client's marketing lead is in Claude on a Tuesday afternoon, drafting a cold email sequence.
Without your MCP, Claude writes a generic sequence. Decent. Forgettable. Nothing you'd have signed off on.
With your MCP, the client's Claude asks your firm's knowledge first. It pulls your sequencing rules, your subject-line principles, the three objections you've seen kill this exact type of campaign, and the thing you told this specific client in March about their ICP. Then it drafts. The output sounds like it came from your senior strategist, because in every way that matters, it did.
Nobody booked a call. Nobody waited until Thursday. The client got your judgment at the moment they needed it, and your firm's name was on the answer.
A few agencies are building this now. Some call it an MCP. Others call it an always-on version of themselves. Functionally it's the same bet: expertise available on demand, through the tools the client already lives in.
Both are harder than they sound, and the second one is where most firms will get stuck.
Not a wiki. Not a Notion full of SOPs. SOPs capture procedure. The asset is judgment, and judgment never fits neatly in a wiki page.
What you need is a system that holds the reasoning behind your recommendations, the decisions you've made for clients and why, the patterns you've seen across dozens of engagements. The stuff that lives in your head and comes out on calls. If it isn't captured somewhere a machine can read, there's nothing for an MCP to expose.
This is the hard part.
Your knowledge includes things you learned from Client A that you'd never want Client B to see. It includes internal pricing logic, candid assessments, and frameworks you'd happily give a paying client but not a prospect. An MCP for agencies has to know who's asking and what they're allowed to get.
Get this wrong and you've either leaked something or you've locked it down so tight the MCP is useless. Get it right and you can hand a client an always-on version of your firm with a straight face.
I'll keep these anonymous, but they're real.
The founder of a resilience consultancy changed his business model around it. The people he trains can now ask his methodology questions between sessions instead of saving them up for the next workshop, and he's selling that access rather than only selling the workshops.
A GTM agency built one so its clients get the firm's outbound frameworks inside their own AI tools, which means the agency's point of view shows up in the work whether or not someone from the agency is in the room.
A third firm is using it to put its onboarding methodology in front of new clients before the kickoff call, so the first meeting starts further ahead than it used to.
None of them are technical founders. None of them wrote code. They documented what they know, decided who should see what, and plugged it in. That's the whole process, and it's the process we built Gia around.
It's been enlightening to watch this unfold in real time with these early adopters. Six months ago I'd have called it a nice-to-have. I don't anymore.
Here's the part that pushes this from "interesting" to "table stakes."
Right now, a human at your client's company sits in Claude and asks a question. Soon, it won't be a human. It'll be an agent your client has set up to run their outbound, manage their ads, or plan their content. That agent will go looking for context before it acts.
If your expertise has an MCP, the agent finds it and uses it. Your firm's judgment shapes the work.
If it doesn't, the agent uses whatever's available. Generic best practice. A competitor's published framework. Nothing at all.
So this isn't just about your clients needing your context. Soon your clients' agents will need it too. And an agent can't book a call.
You don't need to build anything this quarter. You need to start capturing.
Pick the three questions clients ask you most often. Write down how you actually answer them, including the reasoning, including the exceptions. Do that for a month and you'll have the seed of something an MCP can expose. Do it for a year and you'll have something your competitors can't copy, because it's built on fifteen years of client work they never saw.
The firms that get there first stop selling hours.
The rest will be the ones explaining to clients why the agency's AI doesn't know anything about the agency.
