MCP 服务器
开放的 Model Context Protocol 服务器注册表。按类别、传输方式或使用场景,为 AI 智能体寻找合适的工具、资源和提示词。
服务器
177
工具
471
类别
11
贡献者
155
Official Box MCP server for enterprise content management. Lets AI agents search files, read documents, extract text and metadata, ask questions with Box AI, and manage folders via the Box API. Useful for document analysis, knowledge retrieval, and automating content workflows over files stored in Box.
MCP server for Dropbox cloud storage. Lets AI agents list folders, upload and download files, search content, move and delete items, and create shared links via the Dropbox API with OAuth. Useful for document workflows, backups, and giving agents access to files stored in Dropbox.
Google's MCP Toolbox provides secure, ready-made and custom tools for database queries, schema operations, and semantic search in agent workflows.
MCP server for Gmail that lets AI agents read, search, draft, and send email, manage labels, and organize the inbox via the Gmail API with OAuth. Handy for triaging mail, drafting replies, and automating routine inbox workflows from an AI client.
MCP server for Salesforce that lets AI agents query and modify CRM data using SOQL, manage standard and custom objects (Accounts, Contacts, Opportunities, Cases), describe object metadata, and execute Apex anonymous blocks. Supports both production and sandbox orgs via OAuth or username-password flows.
AgentMail MCP gives AI agents secure email operations through hosted MCP access and local stdio bridges for agent-native inbox workflows.
E2B's MCP server gives AI agents the ability to run arbitrary code in secure, isolated cloud sandboxes. Each sandbox is a fast-booting micro-VM where models can execute Python and shell commands, install packages, read and write files, and capture stdout/stderr — ideal for code interpretation, data analysis, and agentic workflows that need real execution.
Skills 与 MCP 服务器
有什么区别?Skills定义“做什么”
Skill 将说明、示例提示词和推荐模型组织在一起,让智能体稳定地完成任务。Skills 增加的是知识,而不是新的连接。
MCP 服务器定义“如何连接”
MCP 服务器通过连接数据库、API 和文件等真实系统,为智能体增加新能力。MCP 提供连接和操作,而不是任务说明。
简单来说:当模型需要把某项任务做好时选择 Skill;当模型需要访问工具或系统时选择 MCP 服务器。两者可以组合使用,Skill 可以依赖 MCP 服务器提供的工具。
开发了 MCP 服务器?
将它提交到由社区维护的开源注册表。