taproot-mcp
Enables managing SSH-managed clusters with tools for executing commands, file operations, and tmux sessions across multiple nodes.
README
Taproot
Taproot is a local MCP server for SSH-managed nodes. Agents use taproot-mcp serve; the VS Code extension is the control panel for editing node config and checking status.
Remote nodes only need SSH access. They do not need Taproot installed.
Quick Start
1. Install the VS Code extension
Install taproot-mcp from the VS Code Marketplace.
For Remote-SSH, install it under SSH: <host>, not Local.
Open the Taproot panel and click Install/Update Backend.
Taproot runs on the same machine as the agent that connects to it. In a Remote-SSH VS Code window, extension commands run on the SSH host.
CLI-only install, without VS Code:
python -m pip install https://github.com/xiaoxiao27110/taproot/releases/download/v0.2.1/taproot_mcp-0.2.1-py3-none-any.whl
2. Add nodes
Open the Taproot panel in VS Code, add your SSH nodes, then run the connection check from the panel.
You can also check from a terminal:
taproot-mcp check --config /absolute/path/to/nodes.yaml
3. Start the local MCP server
In the Taproot panel or Command Palette, click Start Local MCP Server.
4. Connect your agent
http://localhost:8765/mcp
Claude Code:
claude mcp add --transport http taproot http://localhost:8765/mcp
Agent Examples
Manual server start, without the VS Code button:
taproot-mcp serve --config /absolute/path/to/nodes.yaml --transport http --host 127.0.0.1 --port 8765
Stdio Alternative
For MCP clients that launch servers by command, use stdio instead of a long-running HTTP server.
Codex ~/.codex/config.toml:
[mcp_servers.taproot]
command = "taproot-mcp"
args = ["serve", "--config", "/absolute/path/to/nodes.yaml"]
env = {}
Claude Code stdio:
claude mcp add taproot -- taproot-mcp serve --config /absolute/path/to/nodes.yaml
nodes.yaml
Example:
defaults:
user: admin
key: ~/.ssh/id_rsa
port: 22
nodes:
gpu-node-1:
host: 192.168.1.101
tags: [gpu, vllm]
dev-vm:
host: 192.168.1.200
user: dev
tags: [dev, build]
Config lookup order:
TAPROOT_CONFIG./nodes.yaml~/.config/taproot/nodes.yaml
Tools
Discovery:
cluster_nodes
Broadcast tools:
cluster_execcluster_read_filecluster_edit_filecluster_write_filecluster_list_dircluster_globcluster_system_infocluster_servicecluster_uploadcluster_download
Single-node tmux session tools:
cluster_session_opencluster_session_execcluster_session_readcluster_session_interruptcluster_session_closecluster_session_list
Safety
passwordandsudo_passwordinnodes.yamlare plaintext. Prefer SSH keys.- Do not commit
nodes.yaml,.taproot/, history files, approval files, SSH keys, or VSIX files. - Taproot enforces remote permissions on the MCP server side.
- Home-internal file tools run without approval, except protected directories such as
~/.ssh,~/.aws,~/.kube, and~/.taproot. - Paths outside home,
sudo=True, command execution, service mutations, and tmux command execution require CLI approval.
Approve or reject pending operations:
taproot-mcp approvals list --config /absolute/path/to/nodes.yaml --status pending
taproot-mcp approvals approve <approval-id> --config /absolute/path/to/nodes.yaml
taproot-mcp approvals reject <approval-id> --config /absolute/path/to/nodes.yaml
Development
python -m pip install -e ".[test]"
python -m pytest
VS Code extension:
cd taproot-plugin
npm install
npm test
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。