矩阵爪
矩阵爪 是 AI Skill Hub 本期精选MCP工具之一。综合评分 7.5 分，整体质量较高。我们推荐使用将其纳入你的 AI 工具库，帮助提升工作效率。
📚 深度解析
通过安装 矩阵爪，你的 AI 助手将获得额外的工具调用能力，可以用自然语言直接操控该工具的功能，无需学习复杂的命令行语法。MCP 工具的核心价值在于"一次配置，永久增强"——配置完成后，每次与 AI 对话时都可以无缝调用这些工具。
在技术实现上，MCP 工具通过标准的 JSON-RPC 协议与 AI 客户端通信，工具的功能以"工具列表"的形式暴露给 AI 模型，AI 可以按需调用。矩阵爪 提供了结构化的工具调用接口，使 AI 模型能够精确地理解和使用每个功能点，显著降低 AI 在工具使用上的错误率。
与传统的 API 集成相比，MCP 工具的优势在于无需编写代码——用户只需在配置文件中添加几行 JSON，即可让 AI 获得全新能力。AI Skill Hub 将 矩阵爪 评为 AI 评分 7.5 分，属于同类工具中的优质选择。
📋 工具概览
矩阵爪 是一款遵循 MCP（Model Context Protocol）标准协议的 AI 工具扩展。通过 MCP 协议，它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务，实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用，都可以通过自然语言在 AI 对话中直接触发，极大提升生产效率。
📖 中文文档
矩阵爪 是一款遵循 MCP（Model Context Protocol）标准协议的 AI 工具扩展。通过 MCP 协议，它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务，实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用，都可以通过自然语言在 AI 对话中直接触发，极大提升生产效率。
- 通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
- 提供结构化工具调用接口，显著降低 AI 集成复杂度
- 支持 Claude Desktop 和 Claude Code 无缝接入，开箱即用
- 可与其他 MCP 工具组合叠加，构建完整 AI 工作站
- 轻量无侵入设计，不影响现有系统架构
- 在 Claude Desktop 对话中直接调用本地工具，实现 AI 与系统的深度联动
- 通过自然语言驱动复杂的多步骤自动化任务，代替繁琐手动操作
- 将多个 MCP 工具组合使用，构建个人专属 AI 工作站
# 方式一：通过 Claude Code CLI 一键安装
claude skill install https://github.com/Suren878/matrixclaw
# 方式二：手动配置 claude_desktop_config.json
{
"mcpServers": {
"---": {
"command": "npx",
"args": ["-y", "matrixclaw"]
}
}
}
# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json
- 确认已安装 Node.js（v18 或以上版本）
- 打开 Claude Desktop 或 Claude Code 的 MCP 配置文件
- 按「交给 Agent 安装 → Claude Desktop」标签中的 JSON 配置填入 mcpServers 字段
- 保存配置文件并重启 Claude 客户端
- 重启后，在对话中即可使用本工具
# 安装后在 Claude 对话中直接使用 # 示例： 用户: 请帮我用 矩阵爪 执行以下任务... Claude: [自动调用 矩阵爪 MCP 工具处理请求] # 查看可用工具列表 # 在 Claude 中输入："列出所有可用的 MCP 工具"
// claude_desktop_config.json 配置示例
{
"mcpServers": {
"___": {
"command": "npx",
"args": ["-y", "matrixclaw"],
"env": {
// "API_KEY": "your-api-key-here"
}
}
}
}
// 保存后重启 Claude Desktop 生效
MatrixClaw
MatrixClaw is an always-on local AI assistant and AI coding agent runtime for Terminal and Telegram.
MatrixClaw is an open-source, local-first OpenClaw alternative for users who want a self-hosted personal AI operator with durable sessions, SQLite memory, MCP tools, approvals, scheduled tasks, and Claude Code / Codex subagents.
Start a task in the terminal, continue from Telegram, approve tool calls remotely, and return later without losing the session.
matrixclaw is written in Go and runs as a small local daemon. It stores state in SQLite and gives your AI sessions a durable home outside any single app or chat window.
The core owns the session: context, files, tool history, approvals, provider settings, model choice, usage records, goals/plans, persistent memory, searchable history, and optional external-agent attachments. The Terminal TUI, Telegram bot, and future mobile clients are only interfaces connected to the same local runtime.
matrixclaw is built for personal work first: development, research, files, remote checks, reminders, provider switching, visible task plans, memory, subagents, and agent workflows where continuity and explicit control matter.
It is useful if you want a self-hosted or local-first AI assistant that can keep state across clients, run approved tools, switch between LLM providers, delegate focused tasks to MatrixClaw/Codex/Claude Code subagents, and expose the same runtime through Terminal, Telegram, or MCP.
<p align="center"> <img src="https://github.com/user-attachments/assets/8a3bdb40-1891-4a95-9bce-941ed48f4b72" alt="matrixclaw terminal demo"> </p>
What's New
Latest release highlights:
- Live subagent cards in the Terminal chat for delegate_task and spawn_subagent, with Matrix-style codenames, running/completed/failed states, expandable task details, and metadata previews. - Async subagent state is now merged back into existing tool cards so background work keeps animating after the spawn call returns. - The status bar separates the main model phase from active subagents and queued input, including Waiting for subagent: <name> when the parent run is blocked on child-agent work. - Busy input now queues by default while the assistant is running, with /queue, /steer, /interrupt, and /busy commands for explicit control. - Context management now has /context clear plus compacting. Clear creates a distinct Context cleared chat card and immediately fixes the header token estimate to use only effective post-clear context. - The context engine tracks clear markers, compact summaries, message counts, tool schemas, context window size, and provider usage in one ContextReport. - Command pickers and Terminal navigation were tightened for stable scrolling, clearer previews, and safer destructive confirmations. - Local voice runtime management was split into Piper, Supertonic, and Whisper.cpp drivers with install/status tests.
Install
Install the latest release:
curl -fsSL https://raw.githubusercontent.com/Suren878/matrixclaw/main/scripts/install.sh | bashThe installer downloads the matching GitHub Release archive, installs matrixclaw and matrixclawd into ~/.local/bin, prepares local config/state directories, and starts matrixclaw setup.
Local TTS/STT runtimes are optional because they install extra system packages and build native binaries. To prepare Piper, Supertonic, Whisper.cpp, and ffmpeg during install:
curl -fsSL https://raw.githubusercontent.com/Suren878/matrixclaw/main/scripts/install.sh | bash -s -- --voice-runtimeYou can also install them later:
curl -fsSL https://raw.githubusercontent.com/Suren878/matrixclaw/main/scripts/install_voice_runtime.sh | bashOn systems where packages are managed separately, install git, cmake, a C++ compiler, Python 3 with venv support, and ffmpeg, then run:
scripts/install_voice_runtime.sh --no-system-depsThe voice runtime installer prepares local binaries only. Voices and STT models are selected and downloaded later from the module UI, so an open-source install can stay small until you choose local audio features.
After setup is saved, run matrixclaw to open the terminal TUI. On a fresh machine, plain matrixclaw opens setup first and opens the TUI on later runs.
When the TUI starts, it checks the latest GitHub Release. If a newer version is available, it asks before updating. A successful TUI update then asks whether to restart the daemon so the background service uses the new binary.
Manual update:
matrixclaw update install
matrixclaw service restartUninstall keeps config and state by default:
curl -fsSL https://raw.githubusercontent.com/Suren878/matrixclaw/main/scripts/uninstall.sh | bashRemove config and state explicitly:
curl -fsSL https://raw.githubusercontent.com/Suren878/matrixclaw/main/scripts/uninstall.sh | bash -s -- --purgeCommon Use Cases
- Run a local-first AI assistant with durable sessions, tool history, approvals, memory, and searchable context. - Use a terminal AI TUI that can hand the same session to Telegram without losing state. - Compare or switch between OpenAI-compatible APIs, Codex subscription OAuth, Anthropic-compatible models, Gemini, and custom providers. - Delegate bounded work to native MatrixClaw subagents or external Codex and Claude Code subagents. - Connect MCP servers as assistant tools, or expose MatrixClaw tools to other MCP hosts. - Keep personal AI automation, scheduled tasks, voice, storage, and provider usage in one local SQLite-backed runtime.
OpenClaw, Claude Code, Codex, and OpenCode alternative
MatrixClaw is not just another terminal AI coding agent.
OpenClaw focuses on self-hosted personal AI assistance across chat apps. Claude Code, Codex, and OpenCode focus on coding-agent workflows in the terminal or developer environment.
MatrixClaw sits between them: an always-on local AI operator with a daemon, Terminal TUI, Telegram client, durable SQLite sessions, MCP tools, approvals, scheduled tasks, memory, and Claude Code / Codex subagents.
Use MatrixClaw if you want:
- an OpenClaw alternative with Terminal and Telegram access
- a Claude Code or Codex companion that can orchestrate subagents
- an OpenCode alternative focused on persistent sessions and automation
- a local-first AI agent with SQLite memory and approvals
- a self-hosted personal AI assistant that stays running between sessions
Is MatrixClaw an OpenClaw alternative?
Yes. MatrixClaw is an open-source, local-first OpenClaw alternative for users who want an always-on AI assistant reachable from Terminal and Telegram, with local SQLite state, approvals, MCP tools, memory, scheduled tasks, and subagents.
Is MatrixClaw a Claude Code alternative?
Partly. Claude Code is a terminal coding agent. MatrixClaw is a persistent local AI runtime that can use Claude Code as an external subagent while keeping the main session, memory, approvals, Telegram access, and automation in one daemon.
Is MatrixClaw a Codex alternative?
Partly. Codex CLI is a command-line coding agent. MatrixClaw can use Codex as an external subagent, but its main goal is durable personal AI automation across Terminal, Telegram, MCP tools, memory, approvals, and scheduled tasks.
Is MatrixClaw an OpenCode alternative?
MatrixClaw overlaps with OpenCode for terminal AI workflows, but focuses more on always-on personal automation, Telegram access, persistent sessions, approvals, and subagent orchestration.
FAQ
高质量的AI助手运行时
- 构建多智能体协作系统的 Agent 开发者
- Agent 任务先做 dry-run 验证工具调用链，再开启自主执行
- API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
- 云端托管：可放在 Vercel / Railway / Fly.io 等 PaaS 平台
⚡ 核心功能
- 通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
- 提供结构化工具调用接口，显著降低 AI 集成复杂度
- 支持 Claude Desktop 和 Claude Code 无缝接入，开箱即用
- 可与其他 MCP 工具组合叠加，构建完整 AI 工作站
- 轻量无侵入设计，不影响现有系统架构
- 构建多智能体协作系统的 Agent 开发者
- Agent 任务先做 dry-run 验证工具调用链，再开启自主执行
- API key 直接提交到 git 仓库（请用 .env 并加入 .gitignore）
👥 适合人群
🎯 使用场景
- 在 Claude Desktop 对话中直接调用本地工具，实现 AI 与系统的深度联动
- 通过自然语言驱动复杂的多步骤自动化任务，代替繁琐手动操作
- 将多个 MCP 工具组合使用，构建个人专属 AI 工作站
⚖️ 优点与不足
- +MIT 协议，可免费商用
- +标准化 MCP 协议，生态互联性强
- +与 Claude 官方生态无缝对接
- +即插即用，配置简单快捷
- −依赖 Claude 客户端，非 Claude 用户无法使用
- −MCP 协议仍在持续演进，接口可能变更
- −需要一定的配置步骤
AI Skill Hub 为第三方内容聚合平台，本页面信息基于公开数据整理，不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后，再部署至生产环境，并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一，可自由商用、修改、分发，仅需保留版权声明。
🔗 相关工具推荐
❓ 常见问题 FAQ
经综合评估，矩阵爪 在MCP工具赛道中表现稳健，质量良好。如果你已有明确的使用需求，可以直接上手体验；如果还在评估阶段，建议对比同类工具后再做决策。
| 原始名称 | matrixclaw |
| 原始描述 | 开源MCP工具：Local-first AI assistant runtime in Go: personal AI operator with daemon, TUI, T。⭐17 · Go |
| Topics | ai-agentai-assistantautomation |
| GitHub | https://github.com/Suren878/matrixclaw |
| License | MIT |
| 语言 | Go |
收录时间：2026-05-29 · 更新时间：2026-05-30 · License：MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。