Model Context Protocol

You might say

Can the AI connect to our issue tracker and tell me who owns #1024?

A connection protocol AI applications use to discover and call external tools, data, and promptsWhen a user asks who owns issue #1024, the application discovers get_ticket from an MCP server and sends the lookup. The issue tracker checks read permission and returns Alex. MCP standardizes the connection; it does not replace the real system or its authorization.

Know first

AI Agent Tool Calling API

MCP ServerMCP Client

How does a real answer return after “Who owns issue #1024?”

User question“Who owns issue #1024?”

The user states what to look up; the AI application receives the question and decides which connections are available.

MCP standardizes tool discovery and calls; the external system still owns real data and access control.

Who does what in one get_ticket lookup?

Configuring and exposing tools: Add server command configs in your AI tool settings. The MCP server announces which tools (like get_ticket) it exposes and what arguments they expect.

The AI application requests the call: The model proposes get_ticket({ id: 1024 }), and the MCP client sends it to the issue service.

The issue tracker returns the real record: It checks the test account's read access and returns the owner. MCP does not store the issue or decide permission.

You can say this to an AI Agent

Use MCP to retrieve issue #1024 with a read-only test account. First confirm that the issue service exposes get_ticket, then show the actual arguments, account permission, and tool result. Do not expose close or delete actions.

Learn next

1Skill→2Sub-agent→

Recommended tool

Web fine-tuning tool · Selector Pro

Select page elements for precise AI edits

Click a button, text, or card, note the change, then copy its exact location to AI.

Open Selector Pro website

A user asks who owns issue #1024. After the application discovers get_ticket through MCP, what should happen next?

A. Because it uses MCP, the application can skip the issue tracker's account checkB. Let the model guess an owner from the issue number without calling the toolC. The app requests get_ticket({ id: 1024 }), and the issue tracker checks the current account before returning the real record