Host
The AI application a person uses: a chat assistant, a coding tool or an agent runtime.
Creates the clients, controls which connections are allowed, enforces security policy and consent, and handles the user’s authorisation decisions.
Based on the current stable MCP specification
What is MCP
MCP, the Model Context Protocol, is an open standard that lets AI applications connect to business systems in one consistent way. A system describes what it offers, such as data, actions and prompt templates, and any compatible AI application can discover and use it. MCP standardises the connection. It does not decide what the AI is allowed to do.
Management, IT leads and anyone new to MCP
You keep hearing about MCP and need to know what it changes before a decision.
A shared vocabulary, the protocol basics and the questions MCP leaves to you.
MCP is a connection standard, not a control system.
It makes connecting AI to systems easier and more uniform. Who may do what, which changes need a person to approve them, and how a mistake is noticed and reversed remain decisions your organisation has to design and enforce.
The problem MCP solves
Ten AI applications and ten business systems can mean a hundred separate integrations, each with its own credentials and failure behaviour. MCP replaces that with one protocol: a system is wrapped once as an MCP server, and any compatible AI application can use it. The specification compares this to the Language Server Protocol, which did the same for programming-language support in development tools.
Protocol basis: three roles
The AI application a person uses: a chat assistant, a coding tool or an agent runtime.
Creates the clients, controls which connections are allowed, enforces security policy and consent, and handles the user’s authorisation decisions.
A connector inside the host.
Each client talks to exactly one server and keeps the boundary between servers. A host usually runs several clients at once.
A program in front of one system, for example a CRM, a ticketing tool or a file store.
Exposes that system’s capabilities. It can run as a local process or as a remote service. It receives only the context it needs and cannot see the whole conversation or other servers.
Protocol basis: what a server offers
Functions the AI model can call, such as “create ticket” or “look up order status”. A tool can change data, so the specification treats tools as arbitrary code execution: hosts must obtain explicit user consent before invoking one.
Context and data that the user or the model can read, such as a document, a database record or a file.
Templated messages and workflows that a user chooses to run, such as “summarise this incident”.
The direction can also reverse: a client may offer features to servers, for example elicitation, where a server asks the user for missing information.
Protocol basis: where a server runs
Protocol basis: authorisation
Authorisation is optional in the core protocol. For servers reached over HTTP, the specification defines an OAuth-based model in which the MCP server acts as a resource server and checks an access token issued for it. The tools and resources a caller sees may depend on that authorisation. Business rules, such as which records a person may change or which amount needs a second approver, are not part of the protocol and have to be enforced by the server and the system behind it. Read the security and authorisation guide.
TechOne position
The specification itself states that MCP cannot enforce its security principles at the protocol level; implementers are expected to build consent and authorisation flows and access controls around it.
Where to go next
Next guide
Next guide
Next guide
Next guide
Primary sources
The source list distinguishes core protocol requirements, optional extensions and external implementation guidance. The surrounding enterprise control model is a TechOne architecture recommendation.
Direct answers
No. MCP is an open protocol with a public specification. Many AI applications and systems implement it; no single vendor owns its use.
No. An MCP server usually sits in front of a system’s existing API. The difference is that MCP describes its capabilities in a standard way, so any compatible AI application can discover and call them without a custom integration.
Not on its own. MCP carries requests and defines security principles, but the specification states that it cannot enforce them at the protocol level. Safety comes from the permissions, approvals, evidence and recovery designed around each server.
TechOne position: no. MCP pays off when several AI applications need the same systems, or when capabilities should be reusable and governed in one place. A single, narrow integration may not need it.
TechOne position: local servers suit personal and developer tools tied to one machine. Shared business systems are usually better served by a remote server with central identity, logging and a named owner.
Which operations it exposes, who may call each one, which actions need a human approval, what is recorded, and who owns and operates the server. The security and production guides cover each decision.