MCP 服务器

发现优秀的 MCP 服务器

通过 MCP 服务器扩展您的代理能力，拥有 14,326 个能力。

全部14,326

Remote MCP Server on Cloudflare

PlayMCP Browser Automation Server

A comprehensive MCP server that provides powerful web automation tools using Playwright, enabling web scraping, testing, and browser interaction through natural language commands.

mcpserveraitools

刚刚发布了一个“模型上下文协议”（MCP）服务器，基本上就是JARVIS，但更有个性，而且没有英国口音！等机器人起义发生时，我肯定会在他们的“处理过这个人类的奇怪问题”名单上。

MCP Proxy POC

MCP 만들면서 원리 파헤쳐보기

Okay, here's a breakdown of the server and client implementation for a system conceptually similar to the MCP (Master Control Program) from the movie Tron, along with considerations for a modern implementation. Keep in mind that a real-world MCP would be incredibly complex, so this is a simplified, illustrative example. **Conceptual Overview** The MCP, in essence, is a central control system. In a modern context, we can think of it as a distributed system with the following key components: * **Server (MCP Core):** The central authority. It manages resources, schedules tasks, enforces security policies, and monitors the overall system health. * **Clients (Programs/Processes):** These are the individual applications or processes that interact with the MCP to request resources, execute tasks, and report their status. * **Communication Protocol:** A well-defined protocol for clients and the server to exchange information. **Implementation Considerations** * **Language Choice:** Python is a good choice for prototyping and scripting due to its readability and extensive libraries. For performance-critical components, consider languages like Go, Rust, or C++. * **Communication:** gRPC, ZeroMQ, or even a simple TCP socket-based protocol can be used for communication between the server and clients. gRPC is a modern, high-performance RPC framework that's well-suited for this kind of system. * **Security:** Authentication and authorization are crucial. Use strong authentication mechanisms (e.g., TLS certificates, API keys) and implement role-based access control (RBAC) to restrict access to sensitive resources. * **Resource Management:** The MCP needs to track available resources (CPU, memory, disk space, network bandwidth) and allocate them to clients based on their needs and priorities. * **Task Scheduling:** A scheduler determines the order in which tasks are executed. Prioritization, deadlines, and dependencies should be considered. * **Monitoring and Logging:** Comprehensive monitoring and logging are essential for detecting errors, performance bottlenecks, and security breaches. * **Fault Tolerance:** The MCP should be designed to be fault-tolerant. Consider using techniques like redundancy, replication, and failover to ensure that the system remains operational even if some components fail. **Simplified Python Example (Illustrative)** This is a very basic example to demonstrate the core concepts. It's not production-ready. ```python # server.py (MCP Core) import socket import threading import json import time HOST = '127.0.0.1' PORT = 65432 # Resource Management (very basic) available_cpu = 100 resource_lock = threading.Lock() def handle_client(conn, addr): print(f"Connected by {addr}") while True: try: data = conn.recv(1024).decode() if not data: break try: request = json.loads(data) print(f"Received request: {request}") if request['action'] == 'request_cpu': cpu_units = request['cpu_units'] with resource_lock: global available_cpu if available_cpu >= cpu_units: available_cpu -= cpu_units response = {'status': 'granted', 'cpu_units': cpu_units} print(f"Granted {cpu_units} CPU units. Remaining: {available_cpu}") else: response = {'status': 'denied', 'reason': 'Insufficient CPU'} conn.sendall(json.dumps(response).encode()) elif request['action'] == 'report_status': status = request['status'] print(f"Client reported status: {status}") conn.sendall(json.dumps({'status': 'received'}).encode()) elif request['action'] == 'release_cpu': cpu_units = request['cpu_units'] with resource_lock: available_cpu += cpu_units response = {'status': 'released', 'cpu_units': cpu_units} print(f"Released {cpu_units} CPU units. Remaining: {available_cpu}") conn.sendall(json.dumps(response).encode()) else: response = {'status': 'error', 'message': 'Invalid action'} conn.sendall(json.dumps(response).encode()) except json.JSONDecodeError: response = {'status': 'error', 'message': 'Invalid JSON'} conn.sendall(json.dumps(response).encode()) except ConnectionResetError: print(f"Connection reset by {addr}") break except Exception as e: print(f"Error handling client: {e}") break print(f"Closing connection with {addr}") conn.close() def main(): with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: s.bind((HOST, PORT)) s.listen() print(f"MCP Server listening on {HOST}:{PORT}") while True: conn, addr = s.accept() thread = threading.Thread(target=handle_client, args=(conn, addr)) thread.start() if __name__ == "__main__": main() ``` ```python # client.py (Program/Process) import socket import json import time HOST = '127.0.0.1' PORT = 65432 def request_cpu(conn, cpu_units): request = {'action': 'request_cpu', 'cpu_units': cpu_units} conn.sendall(json.dumps(request).encode()) response = json.loads(conn.recv(1024).decode()) return response def report_status(conn, status): request = {'action': 'report_status', 'status': status} conn.sendall(json.dumps(request).encode()) response = json.loads(conn.recv(1024).decode()) return response def release_cpu(conn, cpu_units): request = {'action': 'release_cpu', 'cpu_units': cpu_units} conn.sendall(json.dumps(request).encode()) response = json.loads(conn.recv(1024).decode()) return response def main(): with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: s.connect((HOST, PORT)) # Request CPU cpu_request_response = request_cpu(s, 20) print(f"CPU Request Response: {cpu_request_response}") if cpu_request_response['status'] == 'granted': # Report status status_response = report_status(s, 'Running task...') print(f"Status Response: {status_response}") time.sleep(5) # Simulate work # Release CPU cpu_release_response = release_cpu(s, 20) print(f"CPU Release Response: {cpu_release_response}") else: print("CPU request denied.") report_status(s, "Exiting") if __name__ == "__main__": main() ``` **How to Run:** 1. Save the code as `server.py` and `client.py`. 2. Run the server: `python server.py` 3. Run the client: `python client.py` (You can run multiple clients in separate terminals). **Explanation:** * **Server (server.py):** * Listens for incoming connections on a specified port. * Handles each client connection in a separate thread. * Receives JSON-encoded requests from clients. * Implements basic CPU resource management (request, release). * Responds to clients with JSON-encoded responses. * **Client (client.py):** * Connects to the server. * Sends JSON-encoded requests to the server (request CPU, report status, release CPU). * Receives JSON-encoded responses from the server. * Simulates a simple task that requests CPU, reports its status, and releases the CPU. **Improvements and Next Steps:** * **Error Handling:** Add more robust error handling to both the server and client. * **Authentication/Authorization:** Implement a proper authentication and authorization system. * **Resource Management:** Expand the resource management to include memory, disk space, and other resources. Implement a more sophisticated resource allocation algorithm. * **Task Scheduling:** Implement a task scheduler that can prioritize tasks, handle dependencies, and enforce deadlines. * **Monitoring:** Integrate monitoring tools to track system health and performance. * **Logging:** Use a logging library (e.g., `logging` in Python) to record events and errors. * **Concurrency:** Use asynchronous programming (e.g., `asyncio` in Python) for better concurrency and scalability. * **gRPC:** Migrate to gRPC for a more efficient and robust communication protocol. Define a gRPC service definition (a `.proto` file) that specifies the available RPC calls. * **Database:** Use a database to store information about users, resources, tasks, and system state. * **Distributed System:** Design the MCP as a distributed system with multiple server nodes for fault tolerance and scalability. Consider using a distributed consensus algorithm (e.g., Raft or Paxos) to ensure consistency across the nodes. **Translation to Chinese** Here's a translation of the conceptual overview and the improvements/next steps into Chinese: **概念概述 (Gài niàn ǒugài - Conceptual Overview)** MCP 本质上是一个中央控制系统。在现代背景下，我们可以将其视为一个分布式系统，具有以下关键组件： * **服务器 (MCP 核心):** 中央权威。它管理资源、调度任务、执行安全策略并监控整个系统健康状况。 * **客户端 (程序/进程):** 这些是与 MCP 交互以请求资源、执行任务和报告其状态的各个应用程序或进程。 * **通信协议:** 客户端和服务器交换信息的明确定义的协议。 **改进和下一步 (Gǎi jìn hé xià yī bù - Improvements and Next Steps)** * **错误处理 (Cuòwù chǔlǐ):** 向服务器和客户端添加更强大的错误处理。 * **身份验证/授权 (Shēnfèn yànzhèng/shòuquán):** 实施适当的身份验证和授权系统。 * **资源管理 (Zīyuán guǎnlǐ):** 扩展资源管理以包括内存、磁盘空间和其他资源。实施更复杂的资源分配算法。 * **任务调度 (Rènwù diàodù):** 实施一个任务调度程序，可以优先处理任务、处理依赖关系并强制执行截止日期。 * **监控 (Jiānkòng):** 集成监控工具以跟踪系统健康状况和性能。 * **日志记录 (Rìzhì jìlù):** 使用日志记录库（例如 Python 中的 `logging`）来记录事件和错误。 * **并发 (Bìngfā):** 使用异步编程（例如 Python 中的 `asyncio`）以获得更好的并发性和可伸缩性。 * **gRPC:** 迁移到 gRPC 以获得更高效和强大的通信协议。定义一个 gRPC 服务定义（一个 `.proto` 文件），指定可用的 RPC 调用。 * **数据库 (Shùjùkù):** 使用数据库来存储有关用户、资源、任务和系统状态的信息。 * **分布式系统 (Fēn bù shì xìtǒng):** 将 MCP 设计为具有多个服务器节点的分布式系统，以实现容错和可伸缩性。考虑使用分布式共识算法（例如 Raft 或 Paxos）来确保节点之间的一致性。 This provides a starting point for building a system inspired by the MCP. Remember to prioritize security, scalability, and fault tolerance as you develop your implementation. Good luck!

stdout-mcp-server

一个模型上下文协议（Model Context Protocol）服务器，它通过命名管道捕获和管理 stdout 日志，从而使应用程序的输出可用于在诸如 Cursor IDE 等 AI 工具中进行查询和调试。 (Alternative, slightly more literal translation): 一个模型上下文协议服务器，通过命名管道捕获并管理标准输出（stdout）日志，使得应用程序的输出能够在诸如 Cursor IDE 等人工智能工具中被查询和用于调试。

MCP Server for Danmarks Statistik

将丹麦统计局的 API 作为可编程资源公开，使其易于与语言模型和现代 AI 应用集成，从而能够使用自然语言查询统计数据。

Figma MCP Server

通过 REST API 或 MCP 协议提取节点层级结构，从而促进 Figma 文件结构的分析。

mcp-servers-latest

最新可用的 Minecraft 服务器。

GitHub MCP Server Integration

mcp-remote-macos-use

第一个开源的MCP服务器，使AI能够完全控制远程macOS系统。

Dify Workflows MCP Server

一个使用 TypeScript 实现的模型上下文协议 (MCP) 服务器，该服务器将 Dify 工作流作为工具暴露出来，供 AI 系统进行交互。

Terminal Control MCP

Enables AI agents to interact with terminal-based TUI applications by capturing visual terminal output as PNG screenshots and simulating keyboard input through a virtual X11 display.

NHL MCP Server

通过模型-上下文协议模式，提供对NHL（国家冰球联盟）数据的结构化访问，包括球队、球员、积分榜、赛程和统计数据。

Demo HTTP MCP Server

A demonstration MCP server that provides example tools for weather queries, time retrieval, and request handling, along with advice prompts. Supports both HTTP and stdio modes for testing MCP client integrations.

clickup-operator MCP server

镜子 (jìng zi)

Mixpanel MCP Server

Patronus MCP Server

QMT-MCP-Server

一个服务器应用程序，使大型语言模型能够通过 QMT 交易系统执行股票交易操作，提供账户查询、持仓管理和订单下单等功能。

LLV Helix Framework

Implements the Lines-Loops-Vibes creativity operating system with tools for building strategic flows, creating iterative loops, and managing energy states. Provides pre-built templates for innovation, strategic design, narrative strategy, and creative intelligence workflows.

Calendar AutoAuth MCP Server

镜子 (jìng zi)

raydium-launchlab-mcp

sourcesage

SourceSage 是一个 MCP（模型上下文协议）服务器，它可以高效地记忆代码库的关键方面——逻辑、风格和标准——同时允许动态更新和快速检索。它被设计为与语言无关，利用 LLM 对多种语言代码的理解。