Definition
An MCP server is a service that exposes tools, resources or prompts to compatible AI clients through the Model Context Protocol.
What an MCP server provides
Tools let a model request actions or calculations. Resources provide data that a client can place into model context. Prompts provide reusable message templates. A server can expose one or more of these primitives and describes its capabilities when a client connects.
The client is the MCP-speaking part of the AI application. It discovers the server's capabilities, presents or selects them for the model and carries requests and results between the model and the server.
What MCP does not decide
MCP standardizes the connection. It does not decide the server's authentication method, custody model, permissions or business rules. Those belong to the product behind the server. A trading MCP server can expose market data and order tools, but its documentation must still explain who holds funds and what limits apply.
What it does not mean
An MCP server is not an AI model and does not make decisions on its own. It gives a compatible client a structured way to reach outside systems.
Where this term fits
Questions about mcp server
- Is an MCP server the same as an API?
- An MCP server can wrap APIs, but it adds a standard way for AI clients to discover tools, resources and prompts and exchange structured requests and results.
- Does every AI assistant support every MCP server?
- No. The client must support the server's transport, authentication and protocol features. Product documentation should list tested connection paths.
- Can an MCP server take actions?
- Yes, when it exposes tools that perform actions. A server may also be read only, and individual tools can have different effects and permissions.
See these concepts in context
Review the Fensory MCP connection, tools and account controls.