linux-harness-mcp
Exposes a Linux host with shell, Python, and filesystem tools as MCP tools over HTTPS, enabling AI agents to execute commands and manage files remotely.
README
linux-harness-mcp
A Model Context Protocol server that exposes this host as a Linux harness over public-trusted HTTPS. Designed for AI agents that have no Linux runtime of their own — they connect and get shell, Python, and filesystem tools that execute here.
Features
- 13 MCP tools: shell exec, Python exec, file read/write, directory ops, glob search, health check, opencode run/ultrawork
- Basic Auth or Bearer Token authentication
- Public HTTPS via ngrok (no certificate management needed)
- One-command install with interactive wizard (
lhm install) - Systemd service with auto-restart
- CLI manager (
lhm) for status, logs, config, diagnostics
Quick Start
Install via npm
npm install -g linux-harness-mcp
Interactive Setup
sudo lhm install
This walks you through bind address, port, authentication, and optional ngrok exposure.
Check Status
lhm status
Get Public Endpoint
lhm url
# https://your-domain.ngrok-free.dev/mcp
CLI Reference (lhm)
| Command | Description |
|---|---|
lhm install |
Interactive setup wizard |
lhm uninstall |
Stop + remove service + ngrok cleanup |
lhm status |
Show service status + public URL |
lhm restart |
Restart the service |
lhm logs |
View harness logs (-f to follow) |
lhm logs --ngrok |
View ngrok tunnel logs |
lhm url |
Print public MCP endpoint |
lhm config |
Show configuration (secrets masked) |
lhm config --edit |
Open config in $EDITOR |
lhm doctor |
Run health checks (incl. ngrok) |
lhm version |
Print version |
MCP Tools
| Tool | Description |
|---|---|
shell_exec |
Run shell commands via /bin/bash -c |
python_exec |
Execute Python scripts with system python3 |
file_read |
Read files (auto text/binary, base64) |
file_write |
Write text or base64 binary files |
list_dir |
List directory entries with metadata |
make_dir |
Create directories (recursive) |
delete_path |
Delete files or directories |
move_path |
Move/rename paths |
stat_path |
Get file metadata (type, size, mode, mtime) |
glob_search |
Glob pattern file search |
health_check |
Service health + host summary |
opencode_run |
Execute complex tasks via opencode CLI |
opencode_ultrawork |
Execute tasks with ultrawork mode |
Client Configuration
Basic Auth (recommended)
{
"mcpServers": {
"linux-harness": {
"url": "https://your-domain.ngrok-free.dev/mcp",
"headers": {
"Authorization": "Basic <base64(user:pass)>",
"ngrok-skip-browser-warning": "any"
}
}
}
}
Bearer Token
{
"mcpServers": {
"linux-harness": {
"url": "http://127.0.0.1:8765/mcp",
"headers": {
"Authorization": "Bearer <token>"
}
}
}
}
Architecture
Agent (MCP client)
| HTTPS (TLS 1.2/1.3, Let's Encrypt via ngrok)
v
ngrok :443 --> 127.0.0.1:8765 (uvicorn + FastMCP Streamable HTTP)
|
v
shell / python3 / filesystem (as root)
- Public ingress: ngrok tunnel to
127.0.0.1:8765 - MCP transport: Streamable HTTP, stateful sessions
- Auth: ASGI middleware (Basic Auth or Bearer Token)
/health: unauthenticated health endpoint
Environment Variables
| Variable | Default | Description |
|---|---|---|
HARNESS_HOST |
127.0.0.1 |
Bind address |
HARNESS_PORT |
8765 |
Bind port |
HARNESS_AUTH_USER |
— | Basic Auth username |
HARNESS_AUTH_PASS |
— | Basic Auth password |
HARNESS_TOKEN |
— | Bearer token (alternative to Basic Auth) |
HARNESS_ALLOWED_ORIGINS |
— | Comma-separated allowed origins |
Development
# Install in editable mode
pip install -e .
# Run tests
pytest tests/ -q
# Build wheel
python -m build --wheel
License
MIT
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。