In recent months MCP, or Model Context Protocol, has entered the vocabulary of people working with AI agents, automation and tools connected to company data. The risk, as often happens in digital, is treating it as yet another acronym to chase.
LU3G reads it differently: MCP isn’t interesting because it “lets AI talk to other software”. It’s interesting because it forces companies to look at a deeper problem: where data lives, who can use it, with what permissions, inside which processes and with what human oversight.
Let’s start simple: what is MCP, without turning it into a technical lecture?
According to the official specification, MCP is an open protocol designed to connect applications based on language models to external data and tools. Its architecture distinguishes between host, client and server, and it uses JSON-RPC to exchange messages.
Put more simply: MCP tries to give AI agents a standard way to access what they need to do their work. Not just text in a chat, but tools, archives, databases, documents, internal systems and operational functions.
The point isn’t to say that MCP “solves integration”. The point is that it sharpens a question: if an AI agent is going to really help a company, where does it get its context and under what rules can it act?
Why does this matter to a company, and not just to developers?
Because the value of AI in a company doesn’t come from the language model alone. It comes from the connection between the model, proprietary data, internal tools and real processes.
AI4Business highlights a point consistent with this reading: core AI technology tends to become a commodity, while value shifts to proprietary data, business processes, governance and the ability to generate concrete impact.
In other words: if two companies use the same AI model, the model alone doesn’t make the difference. What does is the quality of the data, how knowledge is organised, permissions, integrations and the ability to turn an answer into a useful action.
So is MCP a shortcut to better AI agents?
No, and it’s important to say so. MCP can be a technical enabler, not a strategic shortcut.
An AI agent becomes useful when it works on a clear task: reading sales data, querying a CRM, preparing a report, opening a request, comparing documents, suggesting priorities, triggering a workflow. MCP can help structure the connection between the agent and its tools, but it doesn’t decide on its own which processes make sense.
LU3G’s view is this: before asking “can we use MCP?”, a company should ask “which tasks deserve an AI agent, and what data does it need to work well?”.
How are MCP, company data and governance related?
Directly. If an AI agent accesses tools and data, the issue isn’t just technical: it becomes organisational, legal, operational and reputational.
Agenda Digitale, writing about digital repatriation, points to a broader trend: businesses are seeking more control over cloud, supply chain, security and data, driven partly by AI and digital sovereignty. This is about control as much as infrastructure.
MCP fits into the same picture: the more AI connects to company systems, the more rules are needed on access, permissions, traceability, accountability and limits. Without them, the risk is building powerful automation on fragile foundations.
Where does the problem show up in companies?
It shows up when information is scattered: Excel files, outdated documents, an incomplete CRM, procedures that live only in people’s heads, data copied between different tools.
EconomyUp, in the case it reports on data products and the industrialisation of data, describes exactly this shift from one-off requests and Excel sheets to more structured management of a company’s information assets. It also helps in reading MCP: an AI agent connected to messy data quickly produces confusion, not value.
That is why MCP shouldn’t be read as “let’s plug AI into everything”. It should be read as “let’s properly prepare what AI can read, use and trigger”.
What is the most common mistake when talking about MCP and AI agents?
The most common mistake is starting from the tool instead of the process.
A weak question is: “How do we implement MCP?”. A more useful one is: “Which repetitive or high-value task do we want to improve with an AI agent?”. Only then does it make sense to ask whether MCP is the right standard to connect that agent to the systems it needs.
A second mistake is confusing access with authorisation. Just because an agent can connect to a tool doesn’t mean it should be able to do everything. Limits, roles and checks are needed.
How should an SME approach the question of whether it really needs MCP?
LU3G would recommend a four-step path.
First: choose a concrete use case, not a generic one. For example: sales support, analysis of customer requests, marketing reports, internal help desk, knowledge base updates.
Second: map the data and tools involved: CRM, email, documents, databases, tickets, analytics, e-commerce, management software.
Third: define permissions and human oversight: what the agent can read, what it can propose, what it can execute and what must stay subject to a person’s approval.
Fourth: assess whether MCP can simplify the architecture compared with isolated integrations that are hard to maintain.
What role can LU3G play in this kind of project?
The role isn’t to sell an acronym. It’s to help the company connect strategy, data, automation and AI within a sustainable path.
LU3G’s pages on AI for businesses and AI agent development are the natural context for this topic: MCP can become one piece of the puzzle when an agent needs to work with company tools and sources, but the project has to start from goals, processes and responsibilities.
In other words: first you design the work the agent has to do. Then you decide the architecture. Not the other way round.
So the final question is: will MCP change the way companies use AI?
LU3G’s hypothesis: yes, but not because every company will be talking about MCP. It will change things if it helps bring order to the connection between AI agents, data and tools.
The real shift isn’t from chat to agent. It’s from the isolated answer to the governed action. MCP is interesting because it sits exactly there: at the point where AI stops being just conversation and starts to enter processes.
But the value isn’t in the protocol alone. It lies in the company’s ability to know what data it has, which processes it wants to improve and which decisions must never be delegated blindly.
LU3G insights
Sources
- Model Context Protocol – official specification: Primary technical source: MCP is an open protocol for connecting applications based on language models to external data and tools; host/client/server architecture and use of JSON-RPC.
- AI4Business – Perché l’AI genera valore solo se governa i dati aziendali: Independent source: the value of AI is shifting towards proprietary data, business processes, governance and concrete impact, while core technology tends to become a commodity.
- Agenda Digitale – Digital repatriation, perché le aziende riportano i dati sotto al proprio controllo: Independent source: businesses are seeking more control over cloud, supply chain, security and data; AI and digital sovereignty are pushing towards more governable models.
- EconomyUp – Basta con l’anarchia dei fogli Excel: il metodo di Banca Patrimoni Sella per industrializzare i dati: Independent source: data management is evolving towards data products and the industrialisation of information assets, to move beyond one-off requests and the chaos of Excel files.
Giordano Vinci
Giordano Vinci is CEO of LU3G, a digital agency that works alongside companies and SMEs on websites, SEO strategy, advertising campaigns and online growth. His approach combines consulting, method and strategic vision, with the aim of turning digital into a practical business tool.
Giordano Vinci
Giordano Vinci is CEO of LU3G, a digital agency that works alongside companies and SMEs on websites, SEO strategy, advertising campaigns and online growth. His approach combines consulting, method and strategic vision, with the aim of turning digital into a practical business tool.