What is an MCP server?
Definition
An MCP server is a program that exposes data and actions to AI applications using the Model Context Protocol. It describes what it offers as tools, resources and prompts, and any MCP client, such as an AI assistant or coding tool, can connect to it and use those capabilities without a custom integration.
Updated
An MCP server is the other end of the Model Context Protocol. The AI assistant holds the client; the system that has the data runs the server. Once a server exists, any MCP-capable assistant can use it.
How it works
When a client connects, the server announces its capabilities. Tools are functions the model can call, with a name, a description and a typed input schema. Resources are pieces of data the application can read. Prompts are templates the user can pick. The model reads the tool descriptions, decides which to call, and the server runs the call and returns the result.
Servers come in two shapes. A local server runs on the user’s machine and talks to the client over standard input and output, which suits developer tools reading local files. A remote server is reached over HTTP and handles authentication itself, which suits company systems where many people connect with different rights.
What makes a good one
The protocol says little about whose data a server returns. That is left to the server, and it is where most of the risk sits. A server built for company context should:
- authenticate each user and answer only with what that user is permitted to see
- return records with their source, so an answer can be checked
- keep reads auditable, so an owner can see what was accessed
- describe its tools precisely, because the model chooses tools from those descriptions
These are the properties described under permission-scoped context.
MCP server vs connector
A connector, in most products, is a one-way integration that copies data from one system into another vendor’s index. An MCP server leaves the data where it is and answers questions about it on request. The data stays under the owner’s access rules, and the same server serves every compatible client.
How In Parallel approaches it
In Parallel runs a hosted MCP server over its shared context. It offers tools to read decisions, risks, open questions and workspace pages, and to correct them when they are wrong. Each request is answered with the connected person’s access, as set out on the security page, and AI agents need no seat of their own.
FAQ
Common questions
- Do I need to host an MCP server myself?
- Not necessarily. Some MCP servers run locally on a user's machine; others are hosted by the vendor and reached over HTTP with the user's credentials. A hosted server suits shared company data, because access control can be enforced in one place.
- Is it safe to connect an MCP server to an AI assistant?
- It is as safe as the server's access controls. A good server authenticates each user, returns only what that user may see, and logs what was read. Check those three things before connecting one to company data.
- Can an MCP server change data, or only read it?
- Both are possible. Tools can perform actions as well as queries. Well-designed servers separate read and write tools clearly so the host application can ask the user before anything is changed.
Start with your next meeting.
Bring your team in. Use every surface, including every other team's. From €29 per user per month, billed monthly, no commitment.
30-minute demo. We'll show you how meeting signals become a shared record of decisions, owners and risks.