mcp-systemctl
An MCP server that reports on and manages systemd services using systemctl and journalctl, enabling service listing, status, logs, and control operations.
README
mcp-systemctl
An MCP (Model Context Protocol) server that reports on and manages systemd services via systemctl and journalctl. Works with any MCP client — Claude Desktop, AI coding agents, or custom tooling.
Quick Start
npm install && npm run build
node dist/index.js
The server speaks JSON-RPC over stdio (the standard MCP transport). Connect it to your MCP client — see MCP_SYSTEMCTL_USAGE.md for client configuration and all available tools.
Tools
| Tool | Description |
|---|---|
list_services |
List systemd units with state/pattern/type filters |
get_service_status |
Detailed status of a specific service (PID, memory, CPU, uptime, etc.) |
list_failed_services |
All services in a failed/error state |
get_service_logs |
Recent journald log entries for a service |
service_control |
Start, stop, restart, reload, enable, or disable a service |
Full parameter details in MCP_SYSTEMCTL_USAGE.md.
Architecture
The project follows a clean three-layer separation:
src/index.ts — Server setup, tool registration
src/tools/*.ts — Zod input schemas + handler functions (thin glue)
src/systemctl.ts — Core logic: systemctl/journalctl exec, parsing, formatting
┌──────────────┐ stdio (JSON-RPC) ┌──────────────────────┐
│ MCP Client │ ◄──────────────────────► │ mcp-systemctl │
│ │ │ (Node.js process) │
│ │ │ │
│ │ │ tools/*.ts │
│ │ │ → systemctl.ts │
│ │ │ → systemd │
└──────────────┘ └──────────────────────┘
Layer Responsibilities
src/index.ts — Entry point. Creates the McpServer instance, calls server.registerTool() for each tool, then connects to the StdioServerTransport. Adding a new tool means importing its schema and handler and adding one more registerTool call here.
src/tools/ — One file per tool. Each exports:
- A schema object (Zod-defined input parameters with
describe()for documentation) - A handler function conforming to
ToolCallback<T>— receives validated args, calls intosystemctl.ts, formats the response as MCPcontentblocks
src/systemctl.ts — The engine. All systemctl and journalctl interactions live here:
- Shell execution via
execFile(no shell injection risk) - Structured result type
SystemctlResult<T>— every function returns{ ok: true, data }or{ ok: false, error, code, stderr } - Parsers for the systemctl table and key=value output formats
- Human-readable formatters for bytes, nanoseconds, etc.
Extending the Server
Adding a new tool takes four steps. This example adds a get_service_unit_file tool that returns a service's unit file contents.
1. Add the core logic in src/systemctl.ts
export async function getUnitFile(name: string): Promise<SystemctlResult<string>> {
const unitName = name.includes(".") ? name : `${name}.service`;
// Run: systemctl cat <unit>
const args = ["cat", unitName, "--no-pager", "-l"];
return await runSystemctl(args);
}
Use the existing helpers — runSystemctl() and runJournalctl() handle timeouts, permission errors, and missing commands consistently. If you need to parse new output formats, add a parser alongside the existing ones.
2. Create the tool file src/tools/getUnitFile.ts
import * as z from "zod";
import type { ToolCallback } from "@modelcontextprotocol/sdk/server/mcp.js";
import { getUnitFile as getUnitFileCore } from "../systemctl.js";
export const getUnitFileSchema = {
name: z
.string()
.min(1)
.describe("Service name (e.g. 'sshd', 'cron.service')"),
};
export const getUnitFileHandler: ToolCallback<typeof getUnitFileSchema> = async (args) => {
try {
const result = await getUnitFileCore(args.name);
if (!result.ok) {
return {
content: [{ type: "text", text: `Error: ${result.error}` }],
isError: true,
};
}
return {
content: [{ type: "text", text: result.data }],
};
} catch (err) {
return {
content: [{ type: "text", text: `Internal error: ${err instanceof Error ? err.message : String(err)}` }],
isError: true,
};
}
};
The pattern is always: import the core function → call it → check result.ok → format the response. Every handler returns { content: [...], isError?: boolean }.
3. Register the tool in src/index.ts
Add the imports:
import { getUnitFileSchema, getUnitFileHandler } from "./tools/getUnitFile.js";
Add the registration (alongside the existing ones):
server.registerTool(
"get_unit_file",
{
description: "Print the unit file contents of a systemd service.",
inputSchema: getUnitFileSchema,
},
getUnitFileHandler,
);
4. Build and test
npm run build
node dist/index.js
Your MCP client will now discover the new get_unit_file tool automatically — no client-side config changes needed.
Design conventions to follow
| Convention | Why |
|---|---|
Every core function returns SystemctlResult<T> |
Caller always handles errors explicitly; no thrown exceptions |
Use execFile, never exec or shell commands |
Prevents shell injection through service names or patterns |
Schema parameters use .describe() |
MCP clients use this description to document the parameter to the LLM |
| Handlers wrap core calls in try/catch | Last-resort safety net for unexpected errors (not for expected failures) |
Permission errors handled in systemctl.ts |
Consistent messaging across all tools |
Source files use .js extensions in imports |
Required by NodeNext module resolution with ESM |
What not to do
- Don't put shell execution logic in tool handlers — keep it in
systemctl.ts - Don't throw exceptions for expected failures (missing service, permission denied) — return error responses
- Don't skip error handling on
result.ok— everySystemctlResultmust be checked - Don't use
anytypes — the Zod schema provides full type inference - Don't add new dependencies without evaluating whether the existing patterns suffice
Project Structure
mcp-systemctl/
├── src/
│ ├── index.ts # Server entry point, tool registration
│ ├── systemctl.ts # Core systemctl/journalctl logic
│ └── tools/
│ ├── listServices.ts
│ ├── getServiceStatus.ts
│ ├── listFailedServices.ts
│ ├── getServiceLogs.ts
│ └── serviceControl.ts
├── dist/ # Build output (gitignored)
├── node_modules/ # Dependencies (gitignored)
├── MCP_SYSTEMCTL_USAGE.md
├── package.json
├── tsconfig.json
└── .gitignore
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。