一个由社区贡献的MCP服务器目录，涵盖与学术图书馆、科研和高等教育相关的内容。由学术图书馆实践社区中的AI从业者维护。

欢迎对目录和指南内容的贡献。有关如何贡献的信息，请参见 contributing.md。

重用与归因：欢迎你根据具体受众/用例分叉并调整，或链接到本资源。

目录

• 什么是MCP？
• 入门指南
• 学术MCP服务器目录
• 如何评估服务器

什么是MCP？

模型上下文协议（MCP）是由Anthropic开发的一项开源标准，标准化了AI应用与外部数据源、工具（工具调用）及系统之间的连接。

为什么MCP重要

在MCP出现之前，连接AI应用与外部系统需要定制集成。每个应用都需要自己的连接器，且必须单独构建和维护。MCP用通用标准取代了这一点：一旦系统或数据源发布了MCP服务器，任何兼容MCP的应用程序都可以直接连接到它。

Academic MCP Guide and Directory

A community-contributed directory of MCP servers relevant to academic libraries, research, and higher education. Maintained by the AI Practitioners in Academic Libraries Community of Practice.

Contributions to both the directory and guide content are welcome. For information on how to contribute, see contributing.md.

Reuse and Attribution: You are welcome to fork and adapt to your specific audience/use case or link to this resource.

Table of Contents

• What is MCP?
• Getting Started
• Academic MCP Server Directory
• How to Evaluate Servers

What is MCP?

Model Context Protocol (MCP) is an open-source standard developed by Anthropic that standardized the connection between AI applications and external data sources, tools (tool calling), and systems.

Why MCP Matters

Before MCP, connecting AI applications with external systems required custom integrations. For each application, every tool, database, and service would need its own connector, which would have to be built and maintained individually. MCP replaces that with a universal standard: once a system or data source publishes an MCP server, any MCP-compatible application can connect to it directly. This dramatically lowers the engineering effort required on both sides. Developers, database providers, and publishers only need to implement the standard once, and their tools, publications, and datasets will immediately be accessible to any AI application. For application users, MCP facilitates access to a growing library of tools, data sources, and other services. For example, Claude’s desktop app support “connectors” that are built on MCP that let users link tools like Google Drive, Gmail, or other external services directly into a conversation.

Key Terms

MCP Server: a program that exposes data, tools, or capabilities to AI applications via the MCP standard. It acts as a bridge between the AI and an external system, such as a database or third-party service. It takes requests from MCP clients and returns relevant context or results.

MCP Client: the component within an AI application that initiates communication with MCP servers. It sends requests for data or tool use and processes the responses returned by the server.

Host: the AI application or environment in which the MCP client runs. Examples include Claude Desktop, an IDE plugin, or a custom AI assistant. The host manages the client’s connections to MCP servers.

MCP vs RAG

Another common approach to connecting AI with external information is Retrieval-Augmented Generation, or RAG. Both RAG and MCP helps maintain knowledge currency and allow AI applications or models to draw on outside sources rather than relying solely on what the model learned during training. They differ in how that outside information is managed.