知识图谱记忆
AI 与智能体编辑精选Memory
by Anthropic
Memory 是一个基于本地知识图谱的持久化记忆系统,让 AI 记住长期上下文。
帮 AI 和智能体补上“记不住”的短板,用本地知识图谱沉淀长期上下文,连续对话更聪明,数据也更可控。
什么是 知识图谱记忆?
Memory 是一个基于本地知识图谱的持久化记忆系统,让 AI 记住长期上下文。
如何使用 知识图谱记忆
在终端运行。加 --scope project 可写入项目根目录的 .mcp.json,与团队共享。
claude mcp add --transport stdio memory -- npx -y @modelcontextprotocol/server-memoryREADME
Model Context Protocol servers
This repository is a collection of reference implementations for the Model Context Protocol (MCP), as well as references to community-built servers and additional resources.
[!IMPORTANT] If you are looking for a list of MCP servers, you can browse published servers on the MCP Registry. The repository served by this README is dedicated to housing just the small number of reference servers maintained by the MCP steering group.
[!WARNING] The servers in this repository are intended as reference implementations to demonstrate MCP features and SDK usage. They are meant to serve as educational examples for developers building their own MCP servers, not as production-ready solutions. Developers should evaluate their own security requirements and implement appropriate safeguards based on their specific threat model and use case.
The servers in this repository showcase the versatility and extensibility of MCP, demonstrating how it can be used to give Large Language Models (LLMs) secure, controlled access to tools and data sources. Typically, each MCP server is implemented with an MCP SDK:
- C# MCP SDK
- Go MCP SDK
- Java MCP SDK
- Kotlin MCP SDK
- PHP MCP SDK
- Python MCP SDK
- Ruby MCP SDK
- Rust MCP SDK
- Swift MCP SDK
- TypeScript MCP SDK
🌟 Reference Servers
These servers aim to demonstrate MCP features and the official SDKs.
- Everything - Reference / test server with prompts, resources, and tools.
- Fetch - Web content fetching and conversion for efficient LLM usage.
- Filesystem - Secure file operations with configurable access controls.
- Git - Tools to read, search, and manipulate Git repositories.
- Memory - Knowledge graph-based persistent memory system.
- Sequential Thinking - Dynamic and reflective problem-solving through thought sequences.
- Time - Time and timezone conversion capabilities.
Archived
The following reference servers are now archived and can be found at servers-archived.
- AWS KB Retrieval - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime.
- Brave Search - Web and local search using Brave's Search API. Has been replaced by the official server (
@brave/brave-search-mcp-server). - EverArt - AI image generation using various models.
- GitHub - Repository management, file operations, and GitHub API integration.
- GitLab - GitLab API, enabling project management.
- Google Drive - File access and search capabilities for Google Drive.
- Google Maps - Location services, directions, and place details.
- PostgreSQL - Read-only database access with schema inspection.
- Puppeteer - Browser automation and web scraping.
- Redis - Interact with Redis key-value stores.
- Sentry - Retrieving and analyzing issues from Sentry.io.
- Slack - Channel management and messaging capabilities. Now maintained by Zencoder
- SQLite - Database interaction and business intelligence capabilities.
🚀 Getting Started
Using MCP Servers in this Repository
TypeScript-based servers in this repository can be started with npx.
For example, this will start the Memory server:
npx -y @modelcontextprotocol/server-memory
Python-based servers in this repository can be started with uvx or pip. uvx is recommended for ease of use and setup.
For example, this will start the Git server:
# With uvx
uvx mcp-server-git
# With pip
pip install mcp-server-git
python -m mcp_server_git
Follow these instructions to install uv / uvx and these to install pip.
Using an MCP Client
However, running a server on its own isn't very useful, and should instead be configured into an MCP client. For example, here's the Claude Desktop configuration to use the above server:
{
"mcpServers": {
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
}
}
}
On Windows, wrap npx with cmd /c:
{
"mcpServers": {
"memory": {
"command": "cmd",
"args": ["/c", "npx", "-y", "@modelcontextprotocol/server-memory"]
}
}
}
Additional examples of using the Claude Desktop as an MCP client might look like:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
},
"git": {
"command": "uvx",
"args": ["mcp-server-git", "--repository", "path/to/git/repo"]
},
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
}
On Windows, apply the same wrapper to the npx-based filesystem entry above by changing its "command" to "cmd" and prepending "/c", "npx" to its existing args. Leave the uvx entries unchanged.
Transports: stdio and Streamable HTTP
An MCP server talks to its client over a transport. The protocol defines two standard ones, and they run very differently:
| Transport | How the server runs | How a client reaches it |
|---|---|---|
| stdio | The client launches the server as a subprocess and talks to it over the process's stdin and stdout. The server opens no port. | A command and args in the client's configuration, as in the examples above |
| Streamable HTTP | The server runs on its own as an HTTP service, and can serve many clients | A URL, such as http://localhost:3001/mcp |
Every server in this repository serves stdio by default, and every one can also serve Streamable HTTP on http://127.0.0.1:3001/mcp (set PORT and HOST to move it): Everything with its streamableHttp argument (its README also covers its deprecated HTTP+SSE transport), and the others with --transport streamable-http, as each server's README shows. Started without one, a server uses stdio, so when you run uvx mcp-server-git or npx -y @modelcontextprotocol/server-memory in a terminal, it opens no port, prints at most a one-line startup notice, and then appears to hang. That is expected: it is waiting for a client to send it messages on stdin. You do not normally start a stdio server yourself; your MCP client starts it from its configuration and stops it when it is done.
To try a server without setting up a client, use the MCP Inspector. Given a stdio server's command, it launches the server for you and lets you browse and call its tools; given a URL, it connects to a running Streamable HTTP server:
# A stdio server: the Inspector launches it
npx @modelcontextprotocol/inspector uvx mcp-server-git
# A Streamable HTTP server: start it first, then connect to its URL
npx @modelcontextprotocol/server-everything streamableHttp # or: uvx mcp-server-time --transport streamable-http
npx @modelcontextprotocol/inspector --transport http --server-url http://localhost:3001/mcp
For more, see the transport layer in the MCP architecture overview, and Transports in the specification.
🛠️ Creating Your Own Server
Interested in creating your own MCP server? Visit the official documentation at modelcontextprotocol.io for comprehensive guides, best practices, and technical details on implementing MCP servers.
📚 Learn More
See ADDITIONAL.md for a curated list of frameworks and resources that simplify building MCP servers and clients.
🤝 Contributing
See CONTRIBUTING.md for information about contributing to this repository.
📦 Releasing
See RELEASING.md for how packages are published (OIDC trusted publishing from CI — no registry tokens) and how to retry a failed publish.
🔒 Security
See SECURITY.md for reporting security vulnerabilities.
📜 License
This project is licensed under the Apache License, Version 2.0 for new contributions, with existing code under MIT - see the LICENSE file for details.
💬 Community
⭐ Support
If you find MCP servers useful, please consider starring the repository and contributing new servers or improvements!
Managed by Anthropic, but built together with the community. The Model Context Protocol is open source and we encourage everyone to contribute their own servers and improvements!
常见问题
知识图谱记忆 是什么?
使用本地知识图谱实现持久化记忆,提供长期上下文。
如何安装 知识图谱记忆?
运行命令:npx -y @modelcontextprotocol/server-memory
安装排障(国内网络)
npx 拉包很慢、超时或报 ETIMEDOUT 怎么办?
国内直连 npm 官方源经常超时。把 registry 切到 npmmirror(淘宝镜像)即可。全局切换对所有客户端生效;只想对这个 MCP 生效,就在配置的 env 里加 NPM_CONFIG_REGISTRY。bun 和 pnpm 用各自的写法。
npm config set registry https://registry.npmmirror.comnpx --registry=https://registry.npmmirror.com -y @modelcontextprotocol/server-memory"env": {
"NPM_CONFIG_REGISTRY": "https://registry.npmmirror.com"
}pnpm config set registry https://registry.npmmirror.com[install]
registry = "https://registry.npmmirror.com"终端里开了代理,客户端里的 MCP 还是连不上?
从 Dock、开始菜单启动的 Claude Desktop、Cursor 等客户端不会继承终端里 export 的代理变量,MCP 进程拿不到代理。把代理地址写进该 MCP 配置的 env(端口换成你自己的)。注意 Node 自带的 fetch 默认不读 HTTPS_PROXY,Node 24 及以上可再加 NODE_USE_ENV_PROXY=1;仍不行就开代理软件的 TUN / 增强模式。
"env": {
"HTTPS_PROXY": "http://127.0.0.1:7890",
"HTTP_PROXY": "http://127.0.0.1:7890",
"NODE_USE_ENV_PROXY": "1"
}npm config set proxy http://127.0.0.1:7890
npm config set https-proxy http://127.0.0.1:7890报 SyntaxError、fetch is not defined 或 Unsupported engine?
多半是 Node 版本太低。大多数 MCP Server 需要 Node 18 以上,建议用 LTS(20 或 22)。用 nvm 装好后,客户端可能仍在用系统自带的旧 Node:用 which npx 查出完整路径,把配置里的 command 换成这个绝对路径。
node -vnvm install --lts
nvm alias default 'lts/*'which npxWindows 上提示 spawn npx ENOENT 或客户端无法启动?
Windows 下 npx 实际是 npx.cmd,部分客户端不能直接启动它。把 command 改成 cmd,在 args 前面加上 /c 和 npx。
"command": "cmd",
"args": ["/c","npx","-y","@modelcontextprotocol/server-memory"]相关 Skills
Claude接口
by anthropics
面向接入 Claude API、Anthropic SDK 或 Agent SDK 的开发场景,自动识别项目语言并给出对应示例与默认配置,快速搭建 LLM 应用。
✎ 想把Claude能力接进应用或智能体,用claude-api上手快、兼容Anthropic与Agent SDK,集成路径清晰又省心
前端审美
by leonxlnx
给 Claude Code 的前端输出补上审美判断,支持调节设计变化、动效强度和视觉密度,适合做更有层次的页面与组件,避开千篇一律的普通 UI。
✎ 给 AI 补上前端审美短板,可细调设计变化、动效强度和视觉密度,让生成界面告别模板味。
科研技能
by k-dense-ai
面向科研场景的 Claude Code Skill 合集,覆盖科学计算、数据分析与实验流程,让 AI 智能体直接执行专业科研任务,适合研究人员和需要处理科学数据的开发者使用。
相关 MCP Server
顺序思维
编辑精选by Anthropic
Sequential Thinking 是让 AI 通过动态思维链解决复杂问题的参考服务器。
✎ 这个服务器展示了如何让 Claude 像人类一样逐步推理,适合开发者学习 MCP 的思维链实现。但注意它只是个参考示例,别指望直接用在生产环境里。
by deusdata
持久化的代码库知识图谱,可跨会话保留上下文,在 session 重启或上下文压缩后仍能继续使用。
✎ 专治 AI 编程助手“会话失忆”,把代码库沉淀为持久知识图谱,重启或压缩上下文后也能无缝续上开发状态。
PraisonAI
编辑精选by mervinpraison
PraisonAI 是一个支持自反思和多 LLM 的低代码 AI 智能体框架。
✎ 如果你需要快速搭建一个能 24/7 运行的 AI 智能体团队来处理复杂任务(比如自动研究或代码生成),PraisonAI 的低代码设计和多平台集成(如 Telegram)让它上手极快。但作为非官方项目,它的生态成熟度可能不如 LangChain 等主流框架,适合愿意尝鲜的开发者。