Is MCP a product we have to buy?

No. MCP is an open standard. Anyone can build an MCP server, anyone can run one, and the AI tools that use it do not charge per server. The cost is the work of building the server and operating it — not a license fee.

Do we need a separate MCP server for every tool we use?

Often, yes — though many common tools (HubSpot, Slack, Drive, Notion) already have community-maintained MCP servers you can adapt. The work is gluing them to your specific account and rules, not writing every server from scratch.

Where does an MCP server actually run?

In your own cloud, on your accounts, with your credentials. The agent connects to it the same way it would connect to any other tool. We do not host the servers for you.

How is this different from connecting an AI tool to ChatGPT or Claude through their built-in integrations?

The built-in integrations cover the common cases on the platform vendor's terms. An MCP server you control is the way to do specific things — read your custom CRM, follow your refund rules, write to your internal tool — that the built-in integrations do not offer.

What does an MCP integration consultant actually do?

Three things. Decide which servers you actually need (and which you do not). Build the small parts that are specific to your business. Wire it into the rest of the framework — the rules, the cost caps, the audit log — so the AI is not making the call alone.

How long does an MCP integration take, and what does it cost?

It depends on how many tools, how custom the rules, and how much already exists in the open-source community. Our first checkpoint is always a fixed-price scope with a hard cap, so the answer is in writing before either of us starts.

microserv.io

Contact Book a call

MCP, in plain language

MCP servers, for the people paying the invoice.

If you have heard about MCP and the explanations all started with phrases like “protocol layer” and “tool calling primitives,” this page is for you. Same idea, said differently — for the founder, COO, or operations lead who is going to make the call on whether to build one.

What an MCP server is, in one sentence.

An MCP server is a small adapter that lets an AI use one of your business tools the same way one of your team would use it. That is the whole idea. The Model Context Protocol (MCP) is just the agreed-upon shape of those adapters, so any AI that speaks MCP can use any tool that has an MCP server in front of it.

Picture a hotel concierge desk. The concierge does not have keys to every room — they ring through to the right department. An MCP server is the phone line to one department: housekeeping, room service, the front desk. The agent is the concierge.

Without MCP servers, an AI is a smart talker with no hands. It can write you a draft response, but it cannot actually pull the customer record from your CRM, log a refund in your finance tool, or move a card on your project board. With MCP servers, it can — through narrow, well-defined doors that you control.

Why MCP at all

Three problems it quietly solves.

01
Reusable adapters
Before MCP, every AI tool reinvented its own way of talking to Gmail, to Slack, to your CRM. With MCP, that adapter is written once and reused — by Claude, by ChatGPT, by your in-house agent.
02
Permissions you control
An MCP server runs on your accounts, with your credentials, with the permissions you grant. The AI never sees your customers' records directly — it asks the server, which decides what to show.
03
A natural place for guardrails
Cost caps, approval thresholds, "this action requires a human signature" — all of those live cleanly in the MCP server. The AI is one customer of the server, not the only authority over your tools.

A small example

What an MCP server looks like in practice.

Imagine your team uses HubSpot for the customer record and Outlook for email. You want an AI to draft replies to common customer questions without ever sending one on its own authority.

We build (or adopt from open source) two MCP servers. One sits in front of HubSpot and offers two narrow tools: look-up-customer and note-on-customer. The other sits in front of Outlook and offers one tool: draft-reply-in-folder. That is the whole API surface the agent ever sees.

Now an inbound message arrives. The agent calls look-up-customer, the server checks the rules, returns just the relevant fields. The agent drafts a reply, calls draft-reply-in-folder, and the server places the draft in a folder a human reviews before anything is sent. The agent never had Outlook send-mail permission. It could not have, even if it tried.

That is what an MCP integration looks like in real life. Small, named tools. Narrow permissions. A human still in the loop where it matters.

When to hire someone

What an MCP integration consultant actually does.

MCP is open and the example above sounds simple, so the natural question is: do you need a consultant at all? Sometimes no. Often yes — but for narrower reasons than the marketing usually suggests.

Choose the right scope
Half the work is deciding which MCP servers you do not need. We have seen companies build six servers when two would have done the work.
Use what already exists
There are public MCP servers for many common tools. We bias toward adopting and adapting them rather than re-writing from scratch.
Wire in the guardrails
A server without cost caps, approval thresholds, and an audit log is a server you will turn off the first time something goes wrong.
Hand it over
You should be able to read, change, and operate the server without us. We write the runbook and walk your team through it.

Common questions

What people ask about MCP, in plain language.

Is MCP a product we have to buy?: No. MCP is an open standard. Anyone can build an MCP server, anyone can run one, and the AI tools that use it do not charge per server. The cost is the work of building the server and operating it — not a license fee.
Do we need a separate MCP server for every tool we use?: Often, yes — though many common tools (HubSpot, Slack, Drive, Notion) already have community-maintained MCP servers you can adapt. The work is gluing them to your specific account and rules, not writing every server from scratch.
Where does an MCP server actually run?: In your own cloud, on your accounts, with your credentials. The agent connects to it the same way it would connect to any other tool. We do not host the servers for you.
How is this different from connecting an AI tool to ChatGPT or Claude through their built-in integrations?: The built-in integrations cover the common cases on the platform vendor's terms. An MCP server you control is the way to do specific things — read your custom CRM, follow your refund rules, write to your internal tool — that the built-in integrations do not offer.
What does an MCP integration consultant actually do?: Three things. Decide which servers you actually need (and which you do not). Build the small parts that are specific to your business. Wire it into the rest of the framework — the rules, the cost caps, the audit log — so the AI is not making the call alone.
How long does an MCP integration take, and what does it cost?: It depends on how many tools, how custom the rules, and how much already exists in the open-source community. Our first checkpoint is always a fixed-price scope with a hard cap, so the answer is in writing before either of us starts.

30-minute call · no pitch deck

Have a tool you want an AI to use safely? Start with a 30-minute call.

We will tell you whether you need a custom MCP server, an existing one, or neither. Often the answer is not the one a consultancy would prefer.

Book a call Email the principals