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 14,364 个能力。

开发者工具3,065

playwright-mcp-fetch

这个工具提供了一个模型上下文协议（MCP）服务器，用于从网站获取内容，并使用 Playwright 将其转换为不同的格式。

mcp-server-hub

创建一个单例 MCP 服务器中心，可供 Cline/Roo 代码使用。

RobotFrameworkLibrary-to-MCP

Okay, here's a breakdown of how to turn a Robot Framework library into an MCP (Message Center Protocol) server, along with explanations and considerations: **Understanding the Goal** The core idea is to expose the functionality of your Robot Framework library as a service that can be accessed remotely via MCP. This allows other systems (potentially written in different languages or running on different machines) to trigger actions within your Robot Framework library. **Key Concepts** * **Robot Framework Library:** A collection of keywords (functions) that can be used in Robot Framework test cases. * **MCP (Message Center Protocol):** A lightweight protocol for inter-process communication. It's often used for sending commands and receiving responses between different applications or services. * **MCP Server:** A process that listens for MCP requests, processes them, and sends back responses. * **MCP Client:** A process that sends MCP requests to an MCP server. * **Serialization/Deserialization:** Converting data structures (like Python objects) into a format suitable for transmission over a network (e.g., JSON) and then converting them back on the receiving end. **General Approach** 1. **Choose an MCP Library/Framework:** You'll need a Python library that handles the MCP protocol. Some options include: * **`mcp` (Python Package):** A dedicated MCP library for Python. This is likely the most direct and appropriate choice. You can install it with `pip install mcp`. * **ZeroMQ (with MCP Implementation):** ZeroMQ is a powerful messaging library that can be used to implement MCP. This is a more general-purpose solution, but it might be overkill if you only need MCP. * **Other Messaging Libraries:** You *could* potentially use other messaging libraries (like RabbitMQ or Redis Pub/Sub), but you'd need to implement the MCP protocol on top of them, which is more complex. 2. **Create an MCP Server:** Write a Python script that: * Imports your Robot Framework library. * Uses the chosen MCP library to create a server that listens on a specific port. * Registers handlers for different MCP commands. Each handler will correspond to a keyword in your Robot Framework library. * When a command is received, the handler will: * Extract the arguments from the MCP message. * Call the corresponding Robot Framework keyword with those arguments. * Capture the return value (if any) from the keyword. * Serialize the return value (e.g., to JSON). * Send the serialized result back to the MCP client as an MCP response. 3. **Create an MCP Client (if needed for testing):** Write a Python script (or use a tool like `netcat`) that: * Uses the chosen MCP library to connect to the MCP server. * Sends MCP requests with the appropriate command name and arguments. * Receives and deserializes the MCP response. * Prints or processes the result. **Example using the `mcp` Python Package** ```python # server.py (MCP Server) import mcp import json from robot.libraries.BuiltIn import BuiltIn # Or import your custom library # Instantiate your Robot Framework library (or use BuiltIn for demonstration) # my_library = MyRobotLibrary() builtin = BuiltIn() def execute_keyword(keyword_name, *args): """Executes a Robot Framework keyword and returns the result.""" try: result = builtin.run_keyword(keyword_name, *args) return result except Exception as e: return {"error": str(e)} # Handle errors gracefully class MyMCPHandler(mcp.Handler): def handle_message(self, message): """Handles incoming MCP messages.""" try: command = message.command arguments = message.arguments if command == "log_message": # Example: Expose the 'Log' keyword result = execute_keyword("Log", arguments.get("message", "")) # Pass arguments as needed return mcp.Response(result=result) elif command == "get_variable_value": #Example: Expose the 'Get Variable Value' keyword variable_name = arguments.get("name") default_value = arguments.get("default", None) result = execute_keyword("Get Variable Value", variable_name, default_value) return mcp.Response(result=result) else: return mcp.Response(error="Unknown command: {}".format(command)) except Exception as e: return mcp.Response(error=str(e)) if __name__ == "__main__": server = mcp.Server(handler=MyMCPHandler()) server.start() # Defaults to port 7000 print("MCP server started on port 7000...") try: server.join() # Keep the server running except KeyboardInterrupt: print("Shutting down server...") server.stop() ``` ```python # client.py (MCP Client - for testing) import mcp import json def send_mcp_request(command, arguments): """Sends an MCP request and returns the response.""" try: client = mcp.Client() response = client.send_message(mcp.Message(command=command, arguments=arguments)) client.close() return response.result, response.error except Exception as e: return None, str(e) if __name__ == "__main__": # Example 1: Call the 'Log' keyword result, error = send_mcp_request("log_message", {"message": "Hello from MCP!"}) if error: print("Error:", error) else: print("Log Result:", result) # Example 2: Call the 'Get Variable Value' keyword result, error = send_mcp_request("get_variable_value", {"name": "${TEMPDIR}"}) if error: print("Error:", error) else: print("Variable Value:", result) ``` **Explanation of the Example** * **`server.py`:** * Imports the `mcp` library and `robot.libraries.BuiltIn`. Replace `robot.libraries.BuiltIn` with your actual Robot Framework library. * `execute_keyword` function: This is the crucial part. It takes a keyword name and arguments, and then uses `BuiltIn().run_keyword()` (or the equivalent for your library) to execute the keyword. Error handling is included. * `MyMCPHandler`: This class inherits from `mcp.Handler` and overrides the `handle_message` method. This method is called whenever the server receives an MCP message. * It extracts the command name and arguments from the message. * It uses a series of `if/elif/else` statements to determine which Robot Framework keyword to call based on the command name. * It calls `execute_keyword` to execute the keyword. * It creates an `mcp.Response` object with the result (or an error message) and returns it. * The `if __name__ == "__main__":` block creates an `mcp.Server` instance, starts it, and keeps it running until a `KeyboardInterrupt` (Ctrl+C) is received. * **`client.py`:** * Imports the `mcp` library. * `send_mcp_request` function: This function takes a command name and arguments, creates an `mcp.Message` object, sends it to the server, and returns the response. * The `if __name__ == "__main__":` block shows how to use the `send_mcp_request` function to call the `log_message` and `get_variable_value` commands. **How to Run the Example** 1. **Install `mcp`:** `pip install mcp` 2. **Save the code:** Save the server code as `server.py` and the client code as `client.py`. 3. **Run the server:** `python server.py` 4. **Run the client:** `python client.py` (in a separate terminal) You should see the "Hello from MCP!" message logged by the Robot Framework `Log` keyword, and the value of the `${TEMPDIR}` variable printed by the client. **Important Considerations and Enhancements** * **Error Handling:** The example includes basic error handling, but you should add more robust error handling to catch exceptions and return meaningful error messages to the client. * **Security:** MCP itself doesn't provide any security features. If you need to secure your MCP server, you'll need to implement your own security mechanisms (e.g., authentication, encryption). Consider using TLS/SSL for encryption. * **Argument Handling:** The example assumes that the arguments are passed as a dictionary. You might need to adjust the argument handling to match the specific requirements of your Robot Framework keywords. Consider using a more structured data format like JSON Schema to define the expected arguments for each command. * **Data Serialization:** The example uses JSON for serialization. You can use other serialization formats (e.g., Pickle, MessagePack) if needed. JSON is generally a good choice for interoperability. * **Asynchronous Operations:** If your Robot Framework keywords perform long-running operations, consider using asynchronous programming (e.g., `asyncio`) to prevent the MCP server from blocking. * **Configuration:** Use a configuration file (e.g., YAML, JSON) to store the server's port number, logging settings, and other configuration parameters. * **Logging:** Add logging to the server to track requests, responses, and errors. Use a logging library like `logging`. * **Command Discovery:** Implement a mechanism for the client to discover the available commands and their arguments. This could be done by adding a special "describe" command to the server. * **Robot Framework Listener:** You could potentially use a Robot Framework listener to automatically register keywords as MCP commands. This would reduce the amount of manual configuration required. * **Testing:** Write unit tests and integration tests to ensure that your MCP server is working correctly. **In summary, turning a Robot Framework library into an MCP server involves creating a Python script that listens for MCP requests, calls the appropriate Robot Framework keywords, and sends back the results as MCP responses. The `mcp` Python package provides a convenient way to implement the MCP protocol.** Remember to consider error handling, security, argument handling, and other important factors to create a robust and reliable service.

code-rules-mcp

用于将代码规则可靠地读入 Agentic AI 上下文的 MCP 服务器。 (MCP 服务器用于可靠地将代码规则读入自主智能体 (Agentic AI) 的上下文。)

PostgreSQL Model Context Protocol (PG-MCP) Server

Bilibili API MCP Server

用于Bilibili API的MCP服务器，支持各种操作。

mcp-demo

一个使用 mcp-server 库的小型演示项目。

Kibela MCP Server

镜子 (jìng zi)

Ixnetwork Mcp

IxNetwork 会话管理的 MCP 服务器

Bitrefill MCP Server

镜子 (jìng zi)

MCP Hello World Example

一个用于创建 MCP 服务器和客户端项目的模板项目

Openai Mcp Server

适用于 OpenAI 协议的 MCP 服务器

Local MCP-Server-with-HTTPS-and-GitHub-OAuth

这个项目是一个安全的 MCP 服务器，使用 Node.js 和 Express 构建。它具有以下特性：使用自签名证书的 HTTPS 加密、GitHub OAuth 身份验证，以及额外的安全措施，例如速率限制和 HTTP 头部保护。

Magnet Desktop

用于探索 MCP (模型上下文协议) 服务器的 MacOS 和 Windows 桌面应用程序

Cross-System Agent Communication MCP Server

镜子 (jìng zi)

YouTube MCP Server

用于 YouTube 集成的媒体控制协议服务器

Fastapi Mcp Server

FastAPI MCP Server 是一个使用 FastAPI 构建的 MCP (模型上下文协议) 服务器。该项目旨在统一对各种工具和数据源的访问。

Mcp Server Basic Sample

MCP to LangChain/LangGraph Adapter

Okay, I understand. You want me to translate the following English sentence into Chinese: "Adapter that turns MCP server tools into langchain usable tools" Here are a few options, depending on the nuance you want to convey: **Option 1 (Most Literal):** * **将 MCP 服务器工具转换为 Langchain 可用工具的适配器** (Jiāng MCP fúwùqì gōngjù zhuǎnhuàn wéi Langchain kěyòng gōngjù de shìpèiqì) * This is a direct translation and is generally understandable. **Option 2 (Slightly More Natural):** * **一个将 MCP 服务器工具适配成 Langchain 可用工具的工具** (Yī gè jiāng MCP fúwùqì gōngjù shìpèi chéng Langchain kěyòng gōngjù de gōngjù) * This emphasizes that the adapter *is* a tool itself. It also uses "适配成" (shìpèi chéng) which implies adapting *into* something. **Option 3 (Focus on Integration):** * **用于将 MCP 服务器工具集成到 Langchain 的适配器** (Yòng yú jiāng MCP fúwùqì gōngjù jíchéng dào Langchain de shìpèiqì) * This focuses on the *integration* aspect, using the word "集成" (jíchéng), which means "integrate." **Breakdown of the terms:** * **Adapter:** 适配器 (shìpèiqì) * **MCP server tools:** MCP 服务器工具 (MCP fúwùqì gōngjù) * **turns into / adapts to:** 转换成 (zhuǎnhuàn chéng) / 适配成 (shìpèi chéng) * **Langchain usable tools:** Langchain 可用工具 (Langchain kěyòng gōngjù) * **Integrate:** 集成 (jíchéng) **Which option is best depends on the specific context.** If you want a simple, direct translation, Option 1 is fine. If you want to emphasize the adapter's role as a tool, Option 2 is better. If you want to highlight the integration aspect, Option 3 is the best choice.