TL;DR
- Most businesses still use AI manually. Open the CRM, find the data, paste it into ChatGPT, get an answer, then copy something back.
- MCP, or Model Context Protocol, creates a standard connection layer between AI and tools such as Notion, HubSpot, Stripe, Linear, Jira, databases, and other business systems.
- The AI can discover what connected tools allow it to do, retrieve live context, and, with the right permissions, take actions.
- MCP doesn’t replace APIs. It gives AI a standardized way to understand and use capabilities that often rely on APIs underneath.
- MCP infrastructure generally falls into four models: vendor-hosted, self-hosted local, self-hosted remote, and third-party hosted.
- The opportunity isn’t connecting everything. It’s finding workflows where people repeatedly move information between systems and removing that friction.
1. Your Employees Are Still Acting as the Integration Layer
AI has gotten dramatically better.
But look at how most people actually use it at work.
Open ChatGPT or Claude.
Open the CRM.
Find the customer.
Copy the notes.
Paste them into AI.
Ask the question.
Copy the answer.
Open another system.
Paste it there.
It works, but there’s something absurd about it.
The human has become the integration layer between AI and the company’s software.
MCP starts changing that.
2. MCP Gives AI a Standard Way Into Your Tools
MCP stands for Model Context Protocol.
Think of it as a common connection layer between AI applications and the systems your business already uses.
An MCP server can expose tools the AI can use, resources it can access, and reusable prompts or workflows.
Instead of manually feeding the AI everything it needs, you can potentially ask:
“What happened with the Acme account last week?”
The AI could retrieve the relevant context from connected business systems instead of waiting for someone to collect and paste it.
Or:
“Find the decision we made about the pricing change and create the implementation tasks.”
Now the AI isn’t just generating text.
It’s working with your systems.
That’s a much bigger productivity shift.
3. MCP Doesn't Replace APIs
This distinction matters.
Traditional APIs were primarily designed so developers could make one software system communicate with another.
MCP adds a standardized AI-facing layer.
The MCP server tells the AI what capabilities are available and how they can be used.
Underneath, an API, database connection, filesystem operation, or another interface may still do the actual work.
So this isn’t API versus MCP.
APIs provide the machinery. MCP helps AI understand which machinery it can use.
That means the AI can discover approved capabilities and select the appropriate tool for the task instead of requiring a developer to hard-code every interaction into every AI application.
[Read the full article: There’s an MCP for That →]
The full article covers MCP versus APIs, the four infrastructure models, practical MCPs businesses can start with, security considerations, and how multiple connected systems can turn AI into part of an operational workflow.
4. There Are Four Ways to Run MCP
This is where MCP gets more interesting from a business perspective.
Not every MCP connection has the same security, control, or maintenance implications.
Vendor-hosted: The software company runs the MCP server. This is usually the easiest option. Notion, HubSpot, Stripe, Linear, and other platforms have official implementations.
Self-hosted local: The MCP server runs on your computer. This works well for local files, development environments, and resources that don’t need a shared remote service.
Self-hosted remote: Your company runs the MCP server inside infrastructure you control. This gives you more control over authorization, logging, business rules, data residency, and team access.
Third-party hosted: A platform such as Composio, Smithery, or Glama operates the infrastructure or gateway. This can make deployment faster, but another company may sit between your AI and your business systems.
There isn’t one correct answer.
The right architecture depends on what the AI needs to access, who needs access, and what happens if something goes wrong.
5.Don't Connect Everything
This may be the most important part.
MCP makes it tempting to ask:
“How many systems can we connect?”
Wrong question.
Ask:
“Where are people repeatedly moving information between systems by hand?”
Maybe someone checks a CRM, searches email, looks through Slack, verifies a payment, and updates a project-management system every time a customer asks a certain question.
That’s a workflow worth investigating.
Connect the appropriate systems, give the AI only the permissions it needs, add human approval where consequences matter, and measure whether you’ve actually removed work.
Strategy first. MCP second.
Otherwise, we’ll just recreate the SaaS mess with a new generation of AI integrations.
Final Thought
The first phase of business AI was mostly:
Ask AI something. Get an answer.
The next phase looks different:
Give AI a goal. Let it securely gather context, use approved tools, and complete parts of the work.
MCP is one of the technologies making that transition possible.
And that’s why I think the phrase “There’s an MCP for that” is going to become increasingly common.
Look around your company for someone saying:
“I have to go into this system, grab this, paste it over there, check another system, then update this…”
That’s the signal.
There may already be an MCP for that.
Thanks for reading Signal Over Noise,
where we separate real business signal from AI noise.
where we separate real business signal from AI noise.
See you next Tuesday,
Avi Kumar
Founder: Kuware.com
Subscribe Link: https://kuware.com/newsletter/