World Monitor
Real-time global intelligence on conflict, markets, energy, trade, shipping and cyber. 100 tools.
MCP server for the Longhorn distributed storage manager API, with a --read-only mode.
列表按使用分(满分 100)排序,由以下部分组成
| 热度 | 0/80 | GitHub 星标或每周下载量:较大的计 60,较小的计 20 |
|---|---|---|
| 新鲜度 | 12/12 | 30 天内有发布或代码更新得满分 |
| 仍在响应 | 0/8 | 上次检查时云端端点有响应 |
| 使用分 | 12/100 |
质量: 基础检查通过 3/3 项(仅展示,不用于排序) · 上升势头: 暂无数据 (只用于“上升中”标签)
选择你使用的 AI 应用,然后复制设置。
用一句话让 AI 帮你安装
请帮我安装 Longhorn MCP。按照 https://mcp.popnx.com/i/rafaelgaspar-longhorn-mcp.md 操作,先把命令给我看,等我同意后再运行。粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文
通过 3/3 项检查 · 在你的电脑上运行 · ✓ 命令已检查 2026年11月13日
第一步:你用哪个 AI?
选择你用的应用,步骤会显示在下方。
可连接 Claude,但需要先准备
需要先在电脑上安装 Node.js(只需一次) · 仅限电脑上的 Claude 应用(Windows 或 Mac),网页和手机不可用
把这段话复制到 Claude 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude 安装 MCP server Longhorn MCP,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "longhorn": { "command": "npx", "args": [ "-y", "longhorn-mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 Longhorn MCP 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -vMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: node -v
安装一次,所有需要 Node.js 的工具都能用
SettingsDeveloperEdit Config
{
"mcpServers": {
"longhorn": {
"command": "npx",
"args": [
"-y",
"longhorn-mcp"
]
}
}
}用 Longhorn MCP 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude 说正在使用 Longhorn MCP,或回答中出现 Longhorn MCP 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
ChatGPT 暂时无法连接 Longhorn MCP
Longhorn MCP 需要安装在你的电脑上,而 ChatGPT 只能连接在网上运行的服务器
A · 使用能连接 Longhorn MCP 的应用
B · 其他 类中能连接 ChatGPT 的其他服务器
从同类中使用较多的里挑选,仅限用地址即可连接、无需密钥的
可连接 Claude Code,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
把这段话复制到 Claude Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude Code 安装 MCP server Longhorn MCP,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):claude mcp add --scope user longhorn -- npx -y longhorn-mcp 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 Longhorn MCP 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -vMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: node -v
安装一次,所有需要 Node.js 的工具都能用
Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell
claude mcp add --scope user longhorn -- npx -y longhorn-mcp应能看到 Longhorn MCP 在列表中并显示已连接
用 Longhorn MCP 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude Code 说正在使用 Longhorn MCP,或回答中出现 Longhorn MCP 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
可连接 Cursor,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
把这段话复制到 Cursor 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Cursor 安装 MCP server Longhorn MCP,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "longhorn": { "command": "npx", "args": [ "-y", "longhorn-mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 Longhorn MCP 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -vMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: node -v
安装一次,所有需要 Node.js 的工具都能用
用 Longhorn MCP 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Cursor 说正在使用 Longhorn MCP,或回答中出现 Longhorn MCP 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同
可连接 VS Code,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
把这段话复制到 VS Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 VS Code 安装 MCP server Longhorn MCP,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "servers": { "longhorn": { "type": "stdio", "command": "npx", "args": [ "-y", "longhorn-mcp" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 Longhorn MCP 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -vMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: node -v
安装一次,所有需要 Node.js 的工具都能用
用 Longhorn MCP 告诉我你能帮我做什么,然后做一件给我看✓ 如果 VS Code 说正在使用 Longhorn MCP,或回答中出现 Longhorn MCP 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同
Claude Code
claude mcp add --scope user longhorn -- npx -y longhorn-mcpClaude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"longhorn": {
"command": "npx",
"args": [
"-y",
"longhorn-mcp"
]
}
}
}Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"longhorn": {
"command": "npx",
"args": [
"-y",
"longhorn-mcp"
]
}
}
}VS Code (mcp.json)
{
"servers": {
"longhorn": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"longhorn-mcp"
]
}
}
}这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。
在你的电脑上运行,适用于
ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月19日
注册名称已通过 GitHub 账户 rafaelgaspar 验证,该账户拥有此仓库
代码最后更新于今天
使用 MIT
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
OSV.dev 中没有 longhorn-mcp@0.1.8 的记录
只检查这个版本的包本身,不包括所有依赖 · 检查于 2026年11月16日
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/servers/rafaelgaspar-longhorn-mcp)提问、分享技巧或报告无法使用。按“有帮助”数量排序。
还没有评价
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 3 项适用的基础检查中通过了 3 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于今天,最新版本(0.1.8)于42天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:World Monitor, openstatus, Oh My Posh Validator, Strata, Jitsu
同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。
如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Real-time global intelligence on conflict, markets, energy, trade, shipping and cyber. 100 tools.
Manage monitors, status pages, incidents and maintenances in your openstatus workspace.
Validate oh-my-posh configurations and segment snippets against the official schema.
MCP server for progressive tool usage at any scale (see https://klavis.ai)
Manage Jitsu data pipelines: destinations, streams, connections, functions, live events.
| 本项 Longhorn MCP | World Monitor | openstatus | Oh My Posh Validator | Strata | Jitsu | |
|---|---|---|---|---|---|---|
| 运行位置 | 需要安装在电脑上 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 |
| 需要密钥 | 否 | 否 | 否 | 否 | 否 | 否 |
| 可在 Claude 中按 + 添加 | 否 | 否 | 否 | 否 | 否 | 否 |
| 基础检查 | 3/3 | 4/4 | 4/4 | 4/4 | 1/4 | 4/4 |
| 许可证 | MIT | AGPL-3.0 | AGPL-3.0 | MIT | Apache-2.0 | MIT |
| 价格 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 |
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
An MCP server for the Longhorn distributed storage manager API. Full read/write coverage of Longhorn's REST API (volumes, snapshots, backups, backing images, nodes, settings, engine images, recurring jobs, and more), a --read-only mode, both stdio and Streamable HTTP transports, and an official Helm chart.
AI-assisted development. The majority of this codebase — implementation, tests, and this documentation — was written with AI assistance (Claude). Review accordingly, especially before running write-mode tools against a production Longhorn cluster.
npx -y longhorn-mcp --longhorn-url=http://<longhorn-manager-host>:9500
Longhorn's manager API is unauthenticated — reachability is the only access control. If you're not running this against an in-cluster Longhorn (see Connecting popular MCP clients → Remote/HTTP), you'll need a VPN, SSH tunnel, or kubectl port-forward svc/longhorn-backend 9500:9500 to reach it safely. Never expose Longhorn's manager API directly to an untrusted network.
| Flag | Default | Description |
|---|---|---|
--http | off (stdio) | Serve Streamable HTTP instead of stdio. |
--port | 3000 | HTTP listen port (with --http). |
--host | 0.0.0.0 | HTTP listen address (with --http). |
--read-only | off | Register only read tools; the client also refuses any non-GET request. |
--longhorn-url | http://longhorn-backend.longhorn-system.svc.cluster.local:9500 | Base URL of the Longhorn manager API. Standard Kubernetes cluster-local DNS by default — a cluster with a custom cluster domain overrides this via --longhorn-url in its own Deployment rather than changing this package's default. |
--allowed-hosts | localhost,127.0.0.1 | Comma-separated Host/Origin allowlist for --http mode. Must include every hostname/IP:port the server will actually be reached on — see MCP spec DNS rebinding guidance. |
An official chart is published alongside every release:
helm install longhorn-mcp oci://ghcr.io/rafaelgaspar/longhorn-mcp/charts/longhorn-mcp --version 0.1.0 \
--set args[0]=--http \ --set args[1]=--longhorn-url=http://longhorn-backend.longhorn-system.svc.cluster.local:9500
Key values:
| Value | Default | Notes |
|---|---|---|
image.repository / image.tag | ghcr.io/rafaelgaspar/longhorn-mcp / chart appVersion | |
args | ["--http"] | Append --read-only, --longhorn-url, --allowed-hosts, etc. — see CLI reference. |
service.enabled / .type / .port | true / ClusterIP / 3000 | |
route.enabled | false | Optional, vendor-neutral core Gateway API HTTPRoute (gateway.networking.k8s.io/v1) — no Envoy Gateway-specific resources. Bring your own Gateway and set route.parentRefs/route.hostnames. |
serviceAccount.create | true |
Full schema: chart/values.yaml / chart/values.schema.json.
The chart is also attached as an OCI referrer directly on the image manifest, for tooling that discovers charts via the OCI Referrers API instead of pulling a separately tagged artifact:
oras discover ghcr.io/rafaelgaspar/longhorn-mcp:0.1.0
Each client's exact config file path/key can drift between versions — check the client's own docs if these don't work. All of them accept the same underlying command/args shape for a local stdio server.
claude mcp add longhorn -- npx -y longhorn-mcp --longhorn-url=http://localhost:9500
Edit claude_desktop_config.json:
{
"mcpServers": { "longhorn": { "command": "npx", "args": ["-y", "longhorn-mcp", "--longhorn-url=http://localhost:9500"] } } }
Edit .cursor/mcp.json (project-scoped) or ~/.cursor/mcp.json (global):
{
"mcpServers": { "longhorn": { "command": "npx", "args": ["-y", "longhorn-mcp", "--longhorn-url=http://localhost:9500"] } } }
.vscode/mcp.json uses a servers key (not mcpServers):
{
"servers": { "longhorn": { "command": "npx", "args": ["-y", "longhorn-mcp", "--longhorn-url=http://localhost:9500"] } } }
To point a client at an already-running Streamable HTTP instance (e.g. this server deployed in-cluster, see Deploying with Helm) instead of spawning a local npx process, use that client's remote-MCP configuration with the server's /mcp URL. Clients without native remote-HTTP MCP support can bridge via mcp-remote.
Longhorn's manager API — and by extension this server — has no authentication of its own; reachability is the access control (see Quick start). For a multi-tenant or externally-reachable deployment, fronting --http mode with an MCP-aware gateway (e.g. Envoy AI Gateway or similar) for auth, rate limiting, and routing is a good idea. It's not required, though — a private network boundary (VPN, SSH tunnel, cluster-internal only) is a perfectly reasonable standalone posture too, and is exactly what the Helm chart defaults to (route.enabled: false, no external exposure).
--read-only is an enforced, structural guarantee (see below); the following is a softer, model-facing one — it shapes what a well-behaved LLM *should* do, not what the server *can* refuse.
Every tool whose action is destructive or hard to reverse (deletes, force-detaches, reverts, restores cluster-wide state, evicts a node, deploys an arbitrary engine image, redirects a backup target, etc.) has its description suffixed with an explicit instruction not to call it unless the user has explicitly authorized that specific action in the current conversation, and never proactively, speculatively, or as a side effect of an unrelated task. The server's MCP instructions field (surfaced by clients that support it, in addition to each tool's own description) repeats the same guidance at the server level as a backstop for tools called without reading their full description closely.
This is advisory, not enforced — a client or model can ignore it. It reduces the chance of an LLM calling volume_delete or systemrestore_create as an unprompted "helpful" side effect; it is not a substitute for --read-only where that's the guarantee you actually need.
Most tool descriptions end with a Longhorn docs: link to the relevant page in Longhorn's official documentation, pinned to the Longhorn version this server's default --longhorn-url target runs (currently 1.13.0) rather than "latest," so the linked content matches the API version actually in use. A handful of tools with no clearly corresponding doc page (instancemanager_*, volumeattachment_*, longhorn_events, volume_job_list, the longhorn_raw_request/longhorn_list_resource_types/longhorn_describe_resource_type introspection tools) don't have one.
Every Longhorn resource returned by this server (a volume, node, backing image, ...) embeds an actions map of the RPC actions currently valid for it, e.g. a volume's actions.attach. Longhorn's own API returns these as URLs pointing at the manager's in-cluster address — useless to an MCP client, which can only call tools, not arbitrary HTTP endpoints. This server rewrites each one into the MCP tool that performs it instead, alongside that tool's parameter shape (name -> compact type, ? suffix for optional) so a client can call it without a separate schema lookup, e.g.:
"actions": {
"attach": { "tool": "volume_attach", "shape": { "name": "string", "hostId": "string", "disableFrontend": "boolean?", "attachedBy": "string?", "attacherType": "string?", "attachmentID": "string?" } }, "updateReplicaCount": { "tool": "volume_update_setting", "shape": { "name": "string", "setting": "enum(19 values)", "value": "string|number|boolean" }, "args": { "setting": "replicaCount" } } }
args (as above on updateReplicaCount) appears when the action already pins down one of the tool's parameters — Longhorn's ~19 single-field volume update* actions all route through the one volume_update_setting tool, distinguished only by which setting value to pass. shape always lists every parameter the tool takes, including ones already pinned via args, so a pinned value can always be resolved against shape without a separate lookup.
Every action every resource type's schema currently declares maps to a named tool this way — including volume's undocumented jobList, via volume_job_list (Longhorn publishes no input/output schema for it, so that tool accepts an open-ended extra object; see longhorn_describe_resource_type('volume') to check the deployed version's exact shape). Any action a future Longhorn version adds that this server hasn't caught up to yet falls back to longhorn_raw_request instead, with args.method/args.path pre-filled from the original URL. Since Longhorn only lists actions currently valid for a resource's state (e.g. salvage only appears on a faulted volume), this reflects what's actually callable right now, not just what the tool schema allows.
--read-only enforces itself two ways:
tools/list, so a read-only instance's tool set is a strict subset of the read-write one.LonghornClient refuses any non-GET request regardless of which tool triggered it. This is what protects longhorn_raw_request, the one escape-hatch tool that stays registered in both modes (its write-ness isn't known until it's actually called).Run both a read-write and a read-only instance under different names in the same client config if you want an LLM session to default to read-only but have an explicit, differently-named write-capable tool set available:
{
"mcpServers": { "longhorn": { "command": "npx", "args": ["-y", "longhorn-mcp", "--longhorn-url=http://localhost:9500"] }, "longhorn-readonly": { "command": "npx", "args": ["-y", "longhorn-mcp", "--read-only", "--longhorn-url=http://localhost:9500"] } } }
npm ci
npm run typecheck npm test npm run build
scripts/introspect-schema.ts dumps a live Longhorn manager's self-described /v1/schemas — useful for cross-checking this package's hand-written tool coverage against whatever Longhorn version you're actually running:
npm run introspect-schema -- http://localhost:9500
npm run introspect-schema -- http://localhost:9500 volume # one resource type in full
Contributions are welcome — see CONTRIBUTING.md for dev setup, coding conventions, and how releases work.
MIT — see LICENSE.