# ToolHive — Run & Manage MCP Servers with thv

> Install, run, and manage MCP servers with the `thv` CLI (plus desktop/K8s options), with registry workflows and AI-client setup helpers.

## Install

Copy the content below into your project:

## Quick Use

1. Install (Homebrew):
   ```bash
   brew tap stacklok/tap
   brew install thv
   thv version
   ```
2. Explore the registry:
   ```bash
   thv registry list
   thv registry info fetch
   ```
3. Run a first server and connect a client:
   ```bash
   thv run fetch
   thv list
   thv client setup
   thv client status
   ```

## Intro

Install, run, and manage MCP servers with the `thv` CLI (plus desktop/K8s options), with registry workflows and AI-client setup helpers.

- **Best for:** developers who want a repeatable, containerized way to run MCP servers and wire them into AI clients safely
- **Works with:** Docker/Podman/Colima, MCP servers, AI clients such as Copilot/Cursor/Claude Code (optional)
- **Setup time:** 20–40 minutes

## Practical Notes

- Quant: start with one registry server (`fetch`) and record the assigned local port plus URL from `thv list`.
- Quant: define a “server budget” (CPU/RAM) per MCP server; scale only after 10 successful runs without crashes.

---

## A clean “MCP ops” baseline

Use ToolHive as the *operational* layer for MCP:

- One command to run/stop servers.
- A consistent local URL shape you can hand to clients.
- A place to centralize permissions and upgrade workflows.

## Suggested rollout

1. Start with a single server (`fetch`).
2. Add one real server you need (GitHub/Notion/etc.).
3. Lock down permissions and review the server list weekly.

## Troubleshooting hint

If `thv run` fails, verify your container runtime (Docker/Podman/Colima) is running before debugging anything else.

### FAQ

**Q: Do I need an AI client to use ToolHive?**
A: No. You can run/list servers without a client; client setup is optional.

**Q: Why use a container runtime?**
A: It isolates servers and makes runs repeatable across machines and teams.

**Q: What should I standardize first?**
A: Server inventory and permission boundaries—treat MCP servers like infra components.

## Source & Thanks

> Source: https://github.com/stacklok/toolhive
> License: Apache-2.0
> GitHub stars: 1,788 · forks: 216

---

<!-- ZH -->

## 快速使用

1. 安装（Homebrew）：
   ```bash
   brew tap stacklok/tap
   brew install thv
   thv version
   ```
2. 浏览 registry：
   ```bash
   thv registry list
   thv registry info fetch
   ```
3. 跑通一个 server 并接入客户端：
   ```bash
   thv run fetch
   thv list
   thv client setup
   thv client status
   ```

## 简介

用 `thv` CLI 安装、运行并管理 MCP servers（也支持桌面端与 K8s 形态），内置 registry 工作流与 AI 客户端接入向导，适合把 MCP 的安装、升级、权限与运行状态统一管起来。

- **适合谁：** 想用容器化、可复用的方式运行 MCP servers，并安全接入各类 AI 客户端的开发者
- **可搭配：** Docker/Podman/Colima、各类 MCP servers、可选 AI 客户端（Copilot/Cursor/Claude Code 等）
- **准备时间：** 20–40 分钟

## 实战建议

- 量化建议：先只跑 `fetch` 这一个 registry server，用 `thv list` 记录分配到的本地端口与 URL。
- 量化建议：为每个 server 定义 CPU/RAM 预算；至少连续成功跑 10 次不崩溃再扩容并行数量。

## 建议把 ToolHive 当作 “MCP 运维层”

ToolHive 适合承担 MCP 的运维工作：

- 统一 run/stop/list 的命令入口。
- 输出可直接交给客户端的本地 URL。
- 把权限、升级与可重复运行的流程集中管理。

## 推荐的落地顺序

1. 从一个最简单的 server（`fetch`）跑通全链路。
2. 再接入一个你真正需要的 server（GitHub/Notion 等）。
3. 逐步收紧权限，并每周复盘 server 清单与使用量。

## 排障提示

`thv run` 出问题时，先确认容器运行时（Docker/Podman/Colima）是否已启动，再排查其他因素。

### FAQ

**必须配合 AI 客户端才能用吗？**
答：不必须。你可以先只跑 server 并用 `thv list` 验证；客户端接入是可选项。

**为什么要用容器运行时？**
答：隔离更强、可复现更好，团队协作时也更容易统一环境。

**最该先标准化什么？**
答：server 清单与权限边界——把 MCP servers 当作基础设施组件来管理。

## 来源与感谢

> Source: https://github.com/stacklok/toolhive
> License: Apache-2.0
> GitHub stars: 1,788 · forks: 216


---
Source: https://tokrepo.com/en/workflows/toolhive-run-manage-mcp-servers-with-thv
Author: MCP Hub