MCP (Model Context Protocol) 介紹

What is MCP (Model Context Protocol) 介紹?

MCP (Model Context Protocol) 介紹 is a standardized tool-calling protocol released by Anthropic in November 2024. It provides a unified framework for large language models (LLMs) to dynamically discover, invoke, and combine external tools and APIs such as weather services, databases, and GitHub.

How to use MCP (Model Context Protocol) 介紹?

Configure an MCP server with either Stdio (local process communication) or SSE (HTTP long-polling) transport. The README shows configuring tools in the VSCode Cline client using JSON configuration; tool usage follows a three-stage cycle of context initialization, dynamic data exchange, and continuous optimization.

Key features of MCP (Model Context Protocol) 介紹

• Standardized tool discovery mechanism for LLMs
• JSON Schema–based parameter validation
• Dynamic tool discovery compared to static function definitions
• Support for composite multi-step task workflows
• Two transport modes: Stdio and SSE
• Compatible with both native function-call and adapter-based models

Use cases of MCP (Model Context Protocol) 介紹

• Querying weather and attractions for a travel destination
• Executing complex multi-step tasks across disparate APIs
• Integrating LLMs with databases, code repositories, or any external service

FAQ from MCP (Model Context Protocol) 介紹

What is the difference between MCP and Function Call?

MCP is a standardized and extended version of Function Call. It provides cross-platform uniformity, dynamic tool discovery, support for composite workflows, full JSON Schema validation, and multi-turn context handling.

Does the model need to natively support Function Call to use MCP?

No. If the model does not support native function calls, the client (e.g., Cline for VSCode) can parse the model’s text output, interpret tool-call intent, and execute the actual invocation.

What transport modes does MCP support and what are their differences?

MCP supports Stdio (local, synchronous stdin/stdout) for easy local testing, and SSE (server-sent events over HTTP) for distributed, network-based deployments.

How does MCP handle tool discovery?

Each MCP server publishes a tool’s name, description, parameter format (JSON Schema), and usage example. The model dynamically discovers and selects tools at runtime based on the task.

What runtime or dependencies are required?

The README shows Python-based server examples. No specific runtime requirements beyond the MCP server implementation are listed; the protocol is language-agnostic.