发现优秀的 MCP 服务器

searxng-mcp

An MCP server that wraps a local SearXNG instance to provide private, customizable web search capabilities. It enables AI assistants to perform queries with support for specific parameters like results limits, language, and time ranges.

JSON Compare MCP Server

Enables deep, order-independent comparison of JSON files to detect differences including missing keys, value mismatches, and type differences. Provides detailed reports with path tracking for precise identification of variations between JSON structures.

MCP Google Custom Search Server

一个用于通过 Google 自定义搜索 API 进行搜索的 MCP 服务器。 (Yī gè yòng yú tōng guò Google zì dìngyì sōusuǒ API jìnxíng sōusuǒ de MCP fúwùqì.)

TimeChimp MCP Server

Enables interaction with the TimeChimp API v2 to manage projects, time entries, expenses, and invoices through natural language. It supports full CRUD operations across all major TimeChimp resources, including advanced OData query filtering and pagination.

Google Analytics MCP Server by CData

Google Analytics MCP Server by CData

EpicMe MCP

An application that demonstrates the future of user interactions through natural language with LLMs, enabling user registration, authentication, and data interaction exclusively via Model Context Protocol (MCP) tools.

mcp-server-odoo

An extensible MCP server that integrates Odoo with LLMs to enable querying and managing business data like partners, quotations, and sales orders. It supports custom tool registration and multiple transport protocols for both local and remote communication.

doc-tools-mcp

在 Node.js 中实现 Word 文档的读取和写入 MCP (Message Communication Protocol) 协议，需要分解成几个部分，并使用合适的库来处理 Word 文档和网络通信。 由于 MCP 通常指的是消息通信协议，与 Word 文档本身没有直接关系，因此我将假设你需要： 1. **读取和写入 Word 文档 (docx 格式)** 2. **使用 MCP 协议发送和接收 Word 文档的内容或相关信息** 以下是一个概念性的实现方案，包含代码示例和解释。 **1. 读取和写入 Word 文档 (docx 格式)** 可以使用 `docx` 库来处理 Word 文档。 ```bash npm install docx ``` ```javascript const { Document, Packer, Paragraph, TextRun } = require("docx"); const fs = require("fs"); // 创建一个新的 Word 文档 async function createWordDocument(data) { const doc = new Document({ sections: [{ children: [ new Paragraph({ children: [ new TextRun(data), ], }), ], }], }); // 将文档保存到文件 const buffer = await Packer.toBuffer(doc); fs.writeFileSync("my-document.docx", buffer); console.log("Word document created successfully!"); } // 读取 Word 文档 (需要额外的库，例如 mammoth.js) async function readWordDocument(filePath) { const mammoth = require("mammoth"); // 需要安装: npm install mammoth try { const result = await mammoth.extractRawText({ path: filePath }); const text = result.value; console.log("Word document content:", text); return text; } catch (error) { console.error("Error reading Word document:", error); return null; } } // 示例用法 async function main() { await createWordDocument("Hello, this is a test document created with Node.js!"); const content = await readWordDocument("my-document.docx"); if (content) { console.log("Successfully read the document."); } } main(); ``` **解释:** * **`docx` 库:** 用于创建和修改 Word 文档。 `Document`, `Paragraph`, `TextRun` 是 `docx` 库提供的类，用于构建文档结构。 * **`mammoth` 库:** 用于读取 Word 文档的内容。 `mammoth.extractRawText` 提取文档中的文本。 * **`fs` 模块:** Node.js 的文件系统模块，用于读写文件。 * **`createWordDocument` 函数:** 创建一个包含指定文本的 Word 文档，并将其保存到 `my-document.docx` 文件中。 * **`readWordDocument` 函数:** 读取指定路径的 Word 文档，并返回其文本内容。 **2. 使用 MCP 协议发送和接收 Word 文档的内容或相关信息** 这里需要定义 MCP 协议的具体格式。 假设 MCP 协议包含以下字段： * `type`: 消息类型 (例如 "document_content", "document_metadata") * `data`: 消息数据 (例如 Word 文档的内容，文档的元数据) 可以使用 Node.js 的 `net` 模块创建 TCP 服务器和客户端，并使用自定义的 MCP 协议进行通信。 ```javascript const net = require("net"); // MCP 协议编码函数 function encodeMCP(type, data) { const message = JSON.stringify({ type, data }); const length = Buffer.byteLength(message, 'utf8'); const lengthBuffer = Buffer.alloc(4); // 4 字节表示消息长度 lengthBuffer.writeInt32BE(length, 0); return Buffer.concat([lengthBuffer, Buffer.from(message, 'utf8')]); } // MCP 协议解码函数 function decodeMCP(buffer) { const length = buffer.readInt32BE(0); const message = buffer.slice(4, 4 + length).toString('utf8'); return JSON.parse(message); } // 服务器端 function startServer(port) { const server = net.createServer((socket) => { console.log("Client connected."); let receivedData = Buffer.alloc(0); socket.on("data", (data) => { receivedData = Buffer.concat([receivedData, data]); while (receivedData.length >= 4) { const length = receivedData.readInt32BE(0); if (receivedData.length >= 4 + length) { const messageBuffer = receivedData.slice(0, 4 + length); const message = decodeMCP(messageBuffer); console.log("Received message:", message); // 处理消息 if (message.type === "document_content") { console.log("Received document content:", message.data); } receivedData = receivedData.slice(4 + length); // 移除已处理的消息 } else { break; // 等待更多数据 } } }); socket.on("end", () => { console.log("Client disconnected."); }); socket.on("error", (err) => { console.error("Socket error:", err); }); }); server.listen(port, () => { console.log(`Server listening on port ${port}`); }); } // 客户端 function connectToServer(port, message) { const client = net.createConnection({ port: port }, () => { console.log("Connected to server."); const encodedMessage = encodeMCP(message.type, message.data); client.write(encodedMessage); }); client.on("data", (data) => { console.log("Received data from server:", data.toString()); client.end(); }); client.on("end", () => { console.log("Disconnected from server."); }); client.on("error", (err) => { console.error("Client error:", err); }); } // 示例用法 async function main() { const port = 8080; // 启动服务器 startServer(port); // 等待服务器启动 await new Promise(resolve => setTimeout(resolve, 1000)); // 读取 Word 文档内容 const documentContent = await readWordDocument("my-document.docx"); if (documentContent) { // 创建 MCP 消息 const message = { type: "document_content", data: documentContent, }; // 连接到服务器并发送消息 connectToServer(port, message); } } main(); ``` **解释:** * **`net` 模块:** Node.js 的网络模块，用于创建 TCP 服务器和客户端。 * **`encodeMCP` 函数:** 将消息编码为 MCP 协议格式。 它将消息类型和数据转换为 JSON 字符串，然后计算字符串的长度，并将长度作为 4 字节的大端整数添加到消息的前面。 * **`decodeMCP` 函数:** 将 MCP 协议格式的消息解码为 JavaScript 对象。 它首先读取消息长度，然后读取消息内容，并将其解析为 JSON 对象。 * **`startServer` 函数:** 启动一个 TCP 服务器，监听指定端口。 当客户端连接时，它会接收数据，解码 MCP 消息，并处理消息。 * **`connectToServer` 函数:** 连接到指定端口的 TCP 服务器，并发送 MCP 消息。 * **示例用法:** 首先启动服务器，然后读取 Word 文档的内容，创建一个包含文档内容的 MCP 消息，并将其发送到服务器。 **关键点:** * **MCP 协议定义:** 你需要根据实际需求定义 MCP 协议的格式。 上面的示例使用 JSON 格式，并添加了消息长度字段。 * **错误处理:** 在实际应用中，需要添加更完善的错误处理机制，例如处理网络连接错误，数据解析错误等。 * **数据分块:** 如果 Word 文档非常大，可能需要将文档内容分成多个块进行传输。 * **安全性:** 如果需要传输敏感数据，需要考虑使用加密技术，例如 TLS/SSL。 * **库的选择:** `docx` 和 `mammoth` 只是处理 Word 文档的其中两种库。 还有其他的库，例如 `officegen`，可以用于创建更复杂的 Word 文档。 选择合适的库取决于你的具体需求。 **总结:** 这个方案提供了一个基本的框架，用于在 Node.js 中实现 Word 文档的读取和写入，并使用 MCP 协议进行通信。 你需要根据实际需求调整代码，并添加必要的错误处理和安全性措施。 记住安装所需的 npm 包：`npm install docx mammoth`。 如果需要更复杂的 Word 文档处理功能，可以考虑使用其他的 Word 文档处理库。

Brevo MCP Server

A comprehensive MCP server providing Claude with full access to Brevo's marketing automation platform through the official SDK, featuring tools for email operations, contact management, campaigns, SMS, conversations, webhooks, e-commerce, and account management.

Terra Config MCP Server

Enables LLMs to configure the TerraAPI dashboard by managing health and fitness integrations, destinations, and provider credentials. It allows users to programmatically interact with the Terra ecosystem to handle developer settings and data source configurations.

Search Intent MCP

一个基于 MCP 的服务，用于分析用户搜索关键词以确定其意图，并提供分类、推理、参考资料和搜索建议，以支持 SEO 分析。

Useful-mcps

以下是一些实用的小型 MCP 服务器，包括： * docx\_replace：替换 Word 文档中的标签 * yt-dlp：基于章节提取章节和字幕 * mermaid：使用 mermaidchart.com API 生成和渲染图像

MCP-Bridge

一个中间件，提供一个与 OpenAI 兼容的端点，可以调用 MCP 工具。 (Yī gè zhōngjiànjiàn, tígōng yī gè yǔ OpenAI xiāng róng de duāndiǎn, kěyǐ diàoyòng MCP gōngjù.)

Microsoft 365 Core MCP Server

Provides comprehensive management of Microsoft 365 services including Exchange, SharePoint, Teams, Azure AD, Intune device management, security & compliance frameworks, and universal access to 1000+ Microsoft Graph API endpoints with advanced features like batch operations, delta queries, and real-time webhooks.

brain-trust

Enables AI agents to ask questions and review planning documents by connecting to OpenAI's GPT-4. Provides context-aware question answering and multi-level plan analysis with structured feedback including strengths, weaknesses, and suggestions.

Chrome MCP Server

Enables AI assistants to control and automate your Chrome browser directly, leveraging existing login states and configurations for tasks like content analysis, semantic search across tabs, screenshots, network monitoring, and interactive operations.

Fusion MCP

A Model Context Protocol server that connects LLMs with Autodesk Fusion, enabling CAD operations through natural language dialogue.

MCP Fullstack

A comprehensive operations platform providing browser automation, web search/crawling, database operations, deployment tools, secrets management, and analytics capabilities for full-stack software engineering workflows. Features AI-driven automation, real-time monitoring dashboard, and cross-platform service management.

MCP Jieba Server

Provides high-performance Chinese text segmentation, POS tagging, and keyword extraction using Rust-based Jieba implementation. Supports both local STDIO and remote HTTP deployment modes.

fastmcp-opengauss

Enables interaction with openGauss databases through multiple transport methods (Stdio, SSE, Streamable-Http). Supports database operations and queries with configurable connection parameters.

mcp-servers

To make an AI agent more general, you need to focus on its ability to handle a wider range of tasks, environments, and situations without requiring specific retraining or modifications. Here's a breakdown of key strategies and considerations: **1. Broaden the Training Data:** * **Diverse Datasets:** Train the agent on a vast and diverse dataset that covers a wide spectrum of scenarios, tasks, and environments. This includes variations in data quality, format, and context. * **Synthetic Data Augmentation:** Generate synthetic data to supplement real-world data, especially for rare or underrepresented scenarios. This can help the agent generalize to unseen situations. * **Curriculum Learning:** Start with simpler tasks and gradually increase the complexity of the training data. This helps the agent learn fundamental skills before tackling more challenging problems. * **Self-Supervised Learning:** Leverage unlabeled data to learn general representations of the world. This can be particularly useful when labeled data is scarce. **2. Improve the Agent's Architecture:** * **Modular Design:** Break down the agent into modular components that can be easily reused and adapted to different tasks. This promotes flexibility and reduces the need for extensive retraining. * **Attention Mechanisms:** Incorporate attention mechanisms to allow the agent to focus on the most relevant information in a given situation. This improves its ability to handle complex and noisy inputs. * **Memory Networks:** Use memory networks to store and retrieve information from past experiences. This allows the agent to learn from its mistakes and adapt to changing environments. * **Meta-Learning (Learning to Learn):** Train the agent to learn new tasks quickly and efficiently. This enables it to adapt to novel situations with minimal training data. * **Hierarchical Reinforcement Learning:** Structure the agent's decision-making process into a hierarchy of sub-goals. This allows it to break down complex tasks into smaller, more manageable steps. **3. Enhance the Agent's Reasoning and Planning Abilities:** * **Symbolic Reasoning:** Integrate symbolic reasoning techniques to enable the agent to reason about abstract concepts and relationships. This can improve its ability to solve complex problems and make informed decisions. * **Planning Algorithms:** Implement planning algorithms to allow the agent to anticipate the consequences of its actions and choose the best course of action to achieve its goals. * **Common Sense Reasoning:** Equip the agent with common sense knowledge to help it understand the world and make reasonable assumptions. This can improve its ability to handle ambiguous or incomplete information. * **Causal Reasoning:** Enable the agent to understand cause-and-effect relationships. This is crucial for understanding how its actions affect the environment and for making predictions about future events. **4. Robustness and Adaptability:** * **Adversarial Training:** Train the agent to be robust to adversarial attacks and noisy inputs. This can improve its ability to handle real-world data and prevent it from being easily fooled. * **Domain Adaptation:** Develop techniques to adapt the agent to new domains or environments with minimal retraining. This is essential for deploying the agent in a variety of settings. * **Transfer Learning:** Leverage knowledge learned from one task or domain to improve performance on another. This can significantly reduce the amount of training data required for new tasks. * **Continual Learning (Lifelong Learning):** Enable the agent to continuously learn and adapt over time without forgetting previously learned knowledge. This is crucial for long-term deployment in dynamic environments. **5. Evaluation and Monitoring:** * **Generalization Metrics:** Use appropriate metrics to evaluate the agent's generalization performance on unseen data and tasks. * **Regular Monitoring:** Continuously monitor the agent's performance in real-world settings to identify potential issues and areas for improvement. * **Explainable AI (XAI):** Develop techniques to explain the agent's decisions and reasoning processes. This can help identify biases and improve trust in the agent. **Key Considerations:** * **Computational Resources:** Training and deploying general AI agents can be computationally expensive. Consider using cloud computing resources and optimizing the agent's architecture for efficiency. * **Ethical Implications:** Be mindful of the ethical implications of deploying general AI agents. Ensure that the agent is fair, unbiased, and does not cause harm. * **Safety:** Implement safety mechanisms to prevent the agent from taking unintended or harmful actions. * **Trade-offs:** There is often a trade-off between generality and performance. A more general agent may not perform as well on specific tasks as a specialized agent. **In summary, making an AI agent more general requires a multi-faceted approach that involves broadening the training data, improving the agent's architecture, enhancing its reasoning and planning abilities, and ensuring its robustness and adaptability. Careful evaluation and monitoring are also essential for ensuring that the agent performs as expected in real-world settings.** --- **Chinese Translation:** 要使人工智能代理更通用，您需要专注于其处理更广泛的任务、环境和情况的能力，而无需进行特定的重新训练或修改。以下是关键策略和考虑因素的细分： **1. 扩大训练数据：** * **多样化的数据集：** 在涵盖各种场景、任务和环境的庞大而多样化的数据集上训练代理。 这包括数据质量、格式和上下文的变化。 * **合成数据增强：** 生成合成数据以补充真实世界的数据，特别是对于罕见或代表性不足的场景。 这可以帮助代理推广到未见过的情况。 * **课程学习：** 从简单的任务开始，逐渐增加训练数据的复杂性。 这有助于代理在处理更具挑战性的问题之前学习基本技能。 * **自监督学习：** 利用未标记的数据来学习世界的通用表示。 当标记数据稀缺时，这尤其有用。 **2. 改进代理的架构：** * **模块化设计：** 将代理分解为可以轻松重用并适应不同任务的模块化组件。 这提高了灵活性，并减少了大量重新训练的需要。 * **注意力机制：** 结合注意力机制，使代理能够专注于给定情况下最相关的信息。 这提高了它处理复杂和嘈杂输入的能力。 * **记忆网络：** 使用记忆网络来存储和检索过去经验的信息。 这使代理能够从错误中学习并适应不断变化的环境。 * **元学习（学习学习）：** 训练代理快速有效地学习新任务。 这使其能够以最少的训练数据适应新的情况。 * **分层强化学习：** 将代理的决策过程构建为子目标的层次结构。 这使其能够将复杂的任务分解为更小、更易于管理的步骤。 **3. 增强代理的推理和规划能力：** * **符号推理：** 集成符号推理技术，使代理能够推理抽象概念和关系。 这可以提高其解决复杂问题和做出明智决定的能力。 * **规划算法：** 实施规划算法，使代理能够预测其行为的后果，并选择实现其目标的最佳行动方案。 * **常识推理：** 为代理配备常识知识，以帮助其理解世界并做出合理的假设。 这可以提高其处理模糊或不完整信息的能力。 * **因果推理：** 使代理能够理解因果关系。 这对于理解其行为如何影响环境以及预测未来事件至关重要。 **4. 鲁棒性和适应性：** * **对抗训练：** 训练代理对对抗性攻击和嘈杂的输入具有鲁棒性。 这可以提高其处理真实世界数据的能力，并防止其轻易被愚弄。 * **领域自适应：** 开发技术，以最少的重新训练将代理适应到新的领域或环境。 这对于在各种环境中部署代理至关重要。 * **迁移学习：** 利用从一项任务或领域中学到的知识来提高另一项任务的性能。 这可以显着减少新任务所需的训练数据量。 * **持续学习（终身学习）：** 使代理能够随着时间的推移不断学习和适应，而不会忘记先前学到的知识。 这对于在动态环境中进行长期部署至关重要。 **5. 评估和监控：** * **泛化指标：** 使用适当的指标来评估代理在未见过的数据和任务上的泛化性能。 * **定期监控：** 持续监控代理在真实世界环境中的性能，以识别潜在问题和需要改进的领域。 * **可解释人工智能 (XAI)：** 开发技术来解释代理的决策和推理过程。 这可以帮助识别偏见并提高对代理的信任。 **关键考虑因素：** * **计算资源：** 训练和部署通用人工智能代理可能在计算上很昂贵。 考虑使用云计算资源并优化代理的架构以提高效率。 * **伦理影响：** 注意部署通用人工智能代理的伦理影响。 确保代理是公平、公正的，并且不会造成伤害。 * **安全：** 实施安全机制以防止代理采取意外或有害的行动。 * **权衡：** 通用性和性能之间通常存在权衡。 更通用的代理在特定任务上的表现可能不如专门的代理。 **总而言之，使人工智能代理更通用需要一种多方面的方法，包括扩大训练数据、改进代理的架构、增强其推理和规划能力，并确保其鲁棒性和适应性。 仔细的评估和监控对于确保代理在真实世界环境中按预期执行也至关重要。**

Keboola Explorer MCP Server

This server facilitates interaction with Keboola's Storage API, enabling users to browse and manage project buckets, tables, and components efficiently through Claude Desktop.

Rootstock MCP Server

A backend service that enables seamless interaction with the Rootstock blockchain using the Model Context Protocol, providing standardized APIs for querying, transacting, and managing assets on Rootstock.

Nearest Tailwind Colors

Finds the closest Tailwind CSS palette colors to any given CSS color value. Supports multiple color spaces and customizable result filtering to help match designs to Tailwind's color system.

MCP Hello World Server

A demonstration Model Context Protocol server that provides a configurable 'Hello World' tool with multi-language support and input validation. It supports both local stdio and remote HTTP deployment via Docker, facilitating testing and integration with platforms like Cursor and Manus.

Pega DX MCP Server

Transforms complex Pega Platform interactions into intuitive, conversational experiences by exposing Pega DX APIs through the standardized Model Context Protocol, enabling AI applications to interact with Pega through natural language.

TechMCP - PSG College of Technology MCP Server

Enables AI assistants to access PSG College of Technology e-campus portal data including CA marks, attendance records, timetable schedules, and course information through natural language queries.

MCP OpenDART

Enables AI assistants to access South Korea's financial disclosure system (OpenDART), allowing users to retrieve corporate financial reports, disclosure documents, shareholder information, and automatically extract and search financial statement notes through natural language queries.