MCP Ops Toolkit
A self-hosted MCP operations gateway for Linux control hosts, enabling safe filesystem and Git operations, authenticated HTTP transport, and optional SSH/SFTP and infrastructure adapters.
README
MCP Ops Toolkit
中文说明 | English
A self-hosted Model Context Protocol (MCP) operations gateway for Linux control hosts. It provides a compact tool surface for safe filesystem and Git operations, authenticated HTTP transport, ChatGPT attachment upload, native file reading, SSH/SFTP adapters, and optional infrastructure integrations.
Security warning: this project can expose highly privileged operations. The unrestricted root-shell path is disabled by default. Audit the code, keep the service on loopback behind an authenticated reverse proxy or tunnel, use a dedicated Unix account, and allowlist every privileged bridge.
Highlights
- Compact FastMCP facade designed to keep tool discovery manageable.
- Constant-time Bearer or
X-API-Keyauthentication for Streamable HTTP. - Atomic binary attachment upload with size limits, SHA-256 verification, overwrite protection, and backup support.
- Native MCP text, image, and blob responses for local files.
- Structured filesystem, Git, Python, Node, and npm operations through a root-owned bridge.
- Persistent SSH sessions, argv-based remote execution, terminal I/O, and SFTP-style transfer through an external allowlisted bridge.
- Optional Azure, Cloudflare, control-plane, and encrypted secret-broker adapters.
- Optional full root shell with bounded runtime, bounded output files, output paging, and explicit opt-in.
Repository layout
compact_app.py: the compact public MCP tool surface.http_server.py: authenticated Streamable HTTP entrypoint.mcp_main.py: stdio entrypoint.safe_ops.py/safe_ops_bridge.py: structured local operations.file_content.py: native MCP file responses.server.py: control-plane and optional root-shell implementation.azure_tools.py,cloudflare_tools.py,ssh_tools.py,secret_tools.py: optional bridge adapters.deploy/systemd/: generic systemd and sudoers examples.
Quick start
git clone https://github.com/gaojiefengxswhuhit/mcp-ops-toolkit.git
cd mcp-ops-toolkit
python3 -m venv .venv
. .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
export MCP_OPS_ACCESS_TOKEN="$(python3 -c 'import secrets; print(secrets.token_urlsafe(48))')"
python http_server.py
The default listener is 127.0.0.1:8792; /healthz is unauthenticated and all MCP routes require the access token. For stdio transport, run python mcp_main.py.
Privileged bridges
The MCP process should run as an unprivileged account. Privileged operations call narrowly named, root-owned executables under /usr/local/sbin/. The repository includes the safe-operations bridge and the optional root-shell bridge. Azure, Cloudflare, SSH, and secret-broker adapters define JSON or argv contracts, but their deployment-specific credential and host-resolution implementations are intentionally not included.
Expected bridge paths can be changed in source or wrapped with stable symlinks:
/usr/local/sbin/mcp-ops-toolkit-safe-ops
/usr/local/sbin/mcp-ops-toolkit-root-shell
/usr/local/sbin/mcp-ops-toolkit-azure-cli
/usr/local/sbin/mcp-ops-toolkit-cloudflare
/usr/local/sbin/mcp-ops-toolkit-ssh-bridge
/usr/local/sbin/mcp-ops-toolkit-secret-broker
Security defaults
- The HTTP server binds to loopback by default.
- DNS-rebinding protection is enabled.
- Allowed hosts and origins are configured through environment variables.
- Full root shell is disabled unless
MCP_OPS_ENABLE_FULL_SHELL=1. - File reads use allowlisted roots and deny common key and secret locations.
- Uploaded files are copied atomically and verified after the copy.
- Real tokens, host aliases, private domains, account identifiers, and credentials must stay outside Git.
See SECURITY.md before production use.
Tests
python -m compileall -q .
python test_server_contract.py server.py
python test_compact_contract.py .
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。