Stockertim Dictionary · Artificial intelligence · Technology
Model Context Protocol
Also called MCP
In short
The Model Context Protocol is an open standard that defines how AI applications connect to outside tools and data through servers offering tools, resources and prompt templates.
- Intermediate
- Technology
- Checked
In detail
Before a shared standard, every pairing of an AI application with a business system needed its own custom integration. The Model Context Protocol, published as an open specification in November 2024, gives them a common language instead. A system is wrapped once as an MCP server, and any application that supports the protocol can then use it.
The protocol has three roles. The host is the AI application a person uses, such as a chat assistant or a coding tool. Inside it, a client keeps a connection to each server. Servers offer three kinds of capability: tools, which are actions the model can ask to run, such as searching tickets or creating a draft; resources, which are data the application can read, such as files or records; and prompts, which are reusable templates. Messages between client and server use JSON-RPC 2.0, carried over standard input and output for servers running on the same machine or over HTTP for remote ones.
When a session starts, the client asks each server what it offers, and the descriptions of the tools are passed to the model so it knows what it may request. The host decides whether a request runs, which is where approval prompts and permission rules sit.
The protocol standardises the connection, not the safety. A server can expose more power than a task needs, tool descriptions and returned content can carry hostile instructions, and a poorly built server can leak credentials, so the same controls apply as to any tool an agent can call.
Worked example
Fernhill Accountancy connects its AI assistant to two systems through the protocol.
- An MCP server for the document store offers read-only resources: client folders and letters.
- An MCP server for the practice database offers two tools: look up a client and create a draft task.
- When a manager asks which clients have returns due this month, the assistant calls the look-up tool, reads the results and lists them.
- Creating draft tasks for those clients waits for the manager to confirm each call.
One assistant reaches two systems through a common standard, with reading free and writing approved.
In practice
Teams use the protocol to give AI assistants and agents access to internal systems, prefer servers from the system's own maintainer or build them in-house, start with read-only tools, and review each server's tool list before enabling it.
Watch out
- Supporting the protocol says nothing about whether a particular server is safe or well built.
- Tool descriptions and results reach the model, so they can carry prompt injection like any other content.
Learn it properly
Official sources
- Guidelines for secure AI system development · National Cyber Security Centre
- Top 10 for large language model applications · OWASP Foundation
Facts last checked on 9 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.