MCP Server
The server-side role in the Model Context Protocol (MCP) architecture. An MCP Server is a tool service program that implements the MCP protocol specification, exposing tool capabilities to AI models that support MCP, enabling plug-and-play invocation of those capabilities.
Core Facts
MCP采用客户端-服务器架构，基于标准化的JSON-RPC通信协议
MCP采用客户端-服务器架构，AI编程工具作为MCP客户端，外部服务作为MCP服务器
The design philosophy of MCP is similar to the unification concept of the USB-C interface, defining a universal client/server architecture that enables AI models to call any external tool in a standardized manner.
Timeline (last 90 days)
The MCP Server provided by Firebase exposes operations such as creating collections, configuring authentication, writing data and deploying functions as tools that AI can call, and the Agent can complete these operations autonomously after obtaining access permission through Google account authorization.
Legacy MCP Servers (such as aws.api-mcp-server or aws.knowledge-mcp-server) provide overlapping tools that may cause AI agent confusion or task failures, and should be removed from claude.json before installation.
iphone-use offers three integration methods: an HTTP API, a 21-tool MCP Server (designed specifically for Claude Code), and remote control (via a browser or iOS app).
The tutorial designs three collaborative services around a stock analysis scenario: tool declaration, calculating stock quotes for the past three days, and generating investment recommendation reports, forming a small Agent workflow of "data acquisition → data analysis → conclusion output"
Managing dependencies with Conda and launching the MCP Server in an activated virtual environment is also viable; the official recommendation of UV is more for consistency and convenience rather than a hard requirement
The MCP official documentation does not recommend starting the service directly, but instead recommends using UV to launch the MCP Server
In an MCP Server, a Tool is the smallest capability unit exposed for upper-level Agent invocation, and the tutorial uses a decorator approach for tool declaration
The MCP Server acts as an intermediary server responsible for translating LLM decisions into actual calls to external systems, as the AI model itself does not communicate directly with databases, logs, or APIs
Request timeouts can be configured via McpClientCustomizer — for example, if no response is received within 30 seconds, the MCP Server will automatically disconnect
After the expansion, agents can connect to closer endpoints, reducing request round-trip time and thereby lowering latency
18 more timeline events
All Facts (20)
MCP采用客户端-服务器架构，基于标准化的JSON-RPC通信协议
90%VerifiedMCP采用客户端-服务器架构，AI编程工具作为MCP客户端，外部服务作为MCP服务器
80%VerifiedThe design philosophy of MCP is similar to the unification concept of the USB-C interface, defining a universal client/server architecture that enables AI models to call any external tool in a standardized manner.
80%VerifiedFastMCP是MCP Python SDK的高层封装，借鉴了FastAPI的设计哲学，通过Python装饰器将普通函数转化为MCP工具
75%VerifiedIn an MCP Server, a Tool is the smallest capability unit exposed for upper-level Agent invocation, and the tutorial uses a decorator approach for tool declaration
65%VerifiedIn MCP's security design, tool invocations are stipulated to require explicit user authorization, and resource access is constrained by permission boundaries.
65%VerifiedA single MCP Host can simultaneously connect to multiple MCP Clients, with each Client interfacing with one Server to enable parallel invocation of multiple external capabilities.
65%VerifiedThe design of MCP Server follows the principle of least privilege, with each Server exposing only the tools within its area of responsibility
65%VerifiedImproperly configured MCP Servers can lead to security risks such as database password leakage, exposure of private files, and token theft.
65%UnverifiedMCP Server的设计借鉴了微服务架构（Microservices Architecture）思想，每个Server只负责一件事，独立开发、独立部署、独立维护
90%Unverified绝大多数MCP Server通过Node.js或Python在本地启动运行
85%Unverified通过MCP协议，AI在生成代码之前会先拉取所使用库的最新官方文档和示例，基于真实信息编写代码
85%UnverifiedMCP Server可以让智能体直接将生成的测试用例写入Jira、TestRail等工具，调用代码仓库读取最新代码变更，以及对接CI/CD流水线
75%UnverifiedMCP Server在Client首次连接时会执行握手过程，返回支持的工具列表、功能描述、输入参数的JSON Schema定义和返回值格式
60%UnverifiedThe MCP Server provided by Firebase exposes operations such as creating collections, configuring authentication, writing data and deploying functions as tools that AI can call, and the Agent can complete these operations autonomously after obtaining access permission through Google account authorization.
50%UnverifiedLegacy MCP Servers (such as aws.api-mcp-server or aws.knowledge-mcp-server) provide overlapping tools that may cause AI agent confusion or task failures, and should be removed from claude.json before installation.
50%Unverifiediphone-use offers three integration methods: an HTTP API, a 21-tool MCP Server (designed specifically for Claude Code), and remote control (via a browser or iOS app).
50%UnverifiedThe MCP official documentation does not recommend starting the service directly, but instead recommends using UV to launch the MCP Server
50%UnverifiedManaging dependencies with Conda and launching the MCP Server in an activated virtual environment is also viable; the official recommendation of UV is more for consistency and convenience rather than a hard requirement
50%UnverifiedThe tutorial designs three collaborative services around a stock analysis scenario: tool declaration, calculating stock quotes for the past three days, and generating investment recommendation reports, forming a small Agent workflow of "data acquisition → data analysis → conclusion output"
50%