An MCP server is the integration layer that exposes external tools and data sources to an AI application through the Model Context Protocol. It presents discoverable tools with names, descriptions, and input schemas, then translates structured agent requests into system-specific API calls and returns usable results. section 18.2
A server might expose document search, calendar access, database queries, or workflow actions. MCP standardizes exposure and invocation, but it does not decide whether a request is authorized; that decision belongs to the surrounding authorization architecture. section 18.2
{
"name": "documents.search",
"description": "Search customer documents available to the current task.",
"inputSchema": {
"type": "object",
"properties": {
"customerId": { "type": "string" },
"query": { "type": "string" }
},
"required": ["customerId", "query"]
}
}
A document-search tool listing shows how an MCP server describes an operation and constrains agent inputs with a JSON schema. section 18.2
