| title | Include file | ||
|---|---|---|---|
| description | Include file | ||
| services | api-center | ||
| ms.service | azure-api-center | ||
| ms.topic | include | ||
| ms.date | 07/16/2025 | ||
| ms.custom |
|
AI agents are becoming widely adopted because of enhanced large language model (LLM) capabilities. However, even the most advanced models face limitations because of their isolation from external data. Each new data source potentially requires custom implementations to extract, prepare, and make data accessible for the models.
The model context protocol (MCP) helps solve this problem. MCP is an open standard for connecting AI models and agents with external data sources such as local data sources (databases or computer files) or remote services (systems available over the internet, such as remote databases or APIs).
MCP follows a client-server architecture where a host application can connect to multiple servers. Whenever your MCP host or client needs a tool, it connects to the MCP server. The MCP server then connects to, for example, a database or an API. MCP hosts and servers connect with each other through the MCP protocol.
The MCP architecture is built on JSON-RPC 2.0 for messaging. Communication between clients and servers occurs over defined transport layers, and supports primarily two modes of operation:
-
Remote MCP servers - MCP clients connect to MCP servers over the internet, establishing a connection using HTTP and server-sent events (SSE), and authorizing the MCP client access to resources on the user's account using OAuth.
-
Local MCP servers MCP clients connect to MCP servers on the same machine, using standard input/output as a local transport method.