🗓️ 19092026 2037

MCP

Model Context Protocol (MCP) is an open standard that lets AI applications connect to external systems through one common interface. An AI application can use many compatible servers without building a custom integration for each one.

The big picture​

AI application (host) -> MCP client -> MCP server -> external system
  • The host is the AI application.
  • A client is the host's connection to one server.
  • A server exposes useful capabilities, such as a database query or a project file.

MCP makes the connection interoperable. The host still decides which servers to connect, which capabilities reach the model, and what actions are allowed.

What a server can offer​

CapabilityPurposeWho normally chooses it
ToolsPerform an operationModel, within host controls
ResourcesProvide context or dataHost application
PromptsSupply a reusable interaction templateUser

Start with llm_tool_use to understand tool calls. Continue to mcp_architecture for how participants communicate, or custom_mcp_servers if you need to expose an internal system.

Important boundary​

MCP does not make a model autonomous, guarantee that a tool is safe, or grant access to data. Treat every tool call as a request that the application and server must validate.

References​