Glossary · MCP protocol
MCP (Model Context Protocol)
MCP, or Model Context Protocol, is a protocol for exchanges between an AI application and services that provide data or functions. It defines shared conventions for discovering and using those capabilities, without itself being a model, an autonomous agent, or a guarantee that a connection is safe.
MCP (Model Context Protocol): the explanation
An application hosting an assistant uses a client component to communicate with an MCP server. The server exposes capabilities such as a search tool, a document resource, or a prompt template. Exchanges follow formats defined by the protocol, allowing the application to incorporate those capabilities into its workflow. A model may help choose which tool to request, but software components handle the protocol messages and execution. The model does not independently perform those communications simply by producing an answer.
MCP helps reduce the need for a different custom integration between every AI application and every service. A service can offer a common interface that several compatible applications can use. This does not remove the service’s own business logic or authorization rules. An MCP server may use an existing API to consult a calendar or search files. The protocol structures that connection while leaving many underlying service mechanisms in place, including the actual operations performed on your data.
Compatibility depends on supported features, protocol versions, and configuration on both sides. Two products advertising MCP support may therefore offer different practical capabilities. The protocol does not, by itself, decide what information you should share or which actions an agent should be allowed to perform. Permissions, authentication, confirmation flows, and the treatment of untrusted content remain important responsibilities of the application and the services you connect, rather than automatic consequences of using a common communication format.
A concrete example
You use an assistant to prepare a community meeting. Its application connects through MCP to an authorized document service and calendar. The assistant finds the previous meeting notes and checks relevant availability. If it proposes creating an event, the application applies its configured permissions and confirmation process. MCP provides the exchange framework; the documents available and actions permitted depend on the connected services and the authorization you have configured.
Why it matters
MCP can make adding a compatible service or switching applications easier. When choosing a tool, verify the capabilities you can actually use rather than relying on the presence of an MCP label. Check where information travels, who operates each server, and which permissions you grant. Being able to disable a connection, limit actions, and inspect outcomes matters as much as how easily the integration can be installed.
Common misconceptions
- An MCP connection does not automatically unlock all your accounts: available capabilities depend on configured servers, granted permissions, and restrictions enforced by both the application and the individual services involved in handling a particular request.
- MCP does not guarantee that a tool is trustworthy or harmless: a server can expose a faulty function, supply misleading content, or request excessive permissions even when its messages correctly follow the protocol’s communication requirements.
Related terms: mcp protocol
- MCP server
- An MCP server is a software component that exposes data or functions through the Model Context Protocol.
- MCP client
- An MCP client is the component of an application that communicates with a server using the Model Context Protocol.
- API
- An API, or application programming interface, defines how software can use another component’s functions or data.
- SDK
- SDK stands for software development kit.
- webhook
- A webhook is a mechanism that lets a service automatically send a notification to a web address when an event occurs.
- structured output
- Structured output is a response arranged according to a defined format, with expected fields and data types.
- A2A protocol
- The A2A protocol is an open set of rules that lets independent AI agents communicate and delegate tasks.
Where this site covers it
The term appears on 3,592 pages of this site. A few of them:
- Choosing an MCP server: trust, maintenance, permissions, and local or remote hosting Guide
- Claude Code reference guide: installation, subscriptions and API access, commands, CLAUDE.md, MCP, subagents, and best practices Guide
- AI agent skills: how they work, how to install them, and how to build and verify them Guide
- Cursor reference guide: the AI code editor, modes, rules, pricing, and limits Guide
- Securing an AI agent: permissions, secrets, sandboxing, human approval, and logging Guide
- Connect an AI agent to Telegram, WhatsApp, Slack, or Discord: methods and risks Guide
- Computer use agents: how they work, available tools, reliability, and security Guide
- AI automation in Make and Zapier: features, agents, limits, and pricing Guide
Go further
Frequently asked questions
Does MCP replace APIs?
Not necessarily. An MCP server may call an API to access a service and then expose capabilities to an AI application. Both interfaces can therefore operate at different layers within the same workflow that handles your request.
Do you need programming skills?
Some applications let you configure connections without writing code. Other installations require technical setup. The effort depends on the application, chosen server, and hosting arrangement, rather than on the protocol alone or the meaning of its name.
Where can you check server capabilities?
The [server feature specification](https://modelcontextprotocol.io/specification/draft/server/index) distinguishes tools, resources, and prompt templates. Also consult the documentation for your application and chosen server to establish which of those capabilities are actually available in the configuration you intend to use.
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score