# MCP Toolbox Python SDKs — Use Toolbox Tools in Apps

> Python SDKs that load MCP Toolbox tools as callable functions/objects, with packages for core apps, Google ADK, LangChain/LangGraph, and LlamaIndex.

## Install

Save as a script file and run:

## Quick Use

1. Ensure you have a running MCP Toolbox server (see docsite).
2. Install the package you need:
   ```bash
   pip install toolbox-core
   # or: toolbox-adk / toolbox-langchain / toolbox-llamaindex
   ```
3. Follow the Python quickstart tutorial:
   - https://mcp-toolbox.dev/documentation/getting-started/local_quickstart/

## Intro

Python SDKs that load MCP Toolbox tools as callable functions/objects, with packages for core apps, Google ADK, LangChain/LangGraph, and LlamaIndex.

- **Best for:** Python teams who want Toolbox-defined tools usable inside LangChain/LlamaIndex or custom code
- **Works with:** Python; MCP Toolbox service; optional integration packages for ADK/LangChain/LlamaIndex
- **Setup time:** 10–25 minutes

## Practical Notes

- Quant: the repo lists **4** packages (`toolbox-core`, `toolbox-adk`, `toolbox-langchain`, `toolbox-llamaindex`).
- Quant: install is a single `pip install ...` and validation is a docsite tutorial run (copy-paste friendly).

## A practical “Toolbox-first” architecture

If you already have agents calling tools via MCP, the SDKs are useful when you need **the same tool definitions** inside your Python app (batch jobs, evaluation harnesses, or offline pipelines).

Recommended approach:

1. **Keep tool definitions in Toolbox** (single source of truth).
2. **Expose tools to agents via MCP** for interactive workflows.
3. **Reuse tools via SDK** for Python-side batch execution (tests, evals, scheduled runs).

This avoids the classic drift where “agent tools” and “backend tools” diverge over time.

### FAQ

**Q: Do I need LangChain to use it?**
A: No. `toolbox-core` is framework-agnostic; integrations are optional.

**Q: Is this the Toolbox server?**
A: No. The README says you should have a running Toolbox server, and these packages integrate with it.

**Q: What should I test first?**
A: Start with one simple tool, run the docsite quickstart, then wire it into your app’s CI job.

## Source & Thanks

> Source: https://github.com/googleapis/mcp-toolbox-sdk-python
> License: Apache-2.0
> GitHub stars: 182 · forks: 49

---

<!-- ZH -->

## 快速使用

1. 先确保 MCP Toolbox 服务端已运行（见官方文档）。
2. 安装对应 SDK：
   ```bash
   pip install toolbox-core
   # 或：toolbox-adk / toolbox-langchain / toolbox-llamaindex
   ```
3. 按 Python Quickstart 教程跑通：
   - https://mcp-toolbox.dev/documentation/getting-started/local_quickstart/

## 简介

MCP Toolbox 的 Python SDK 能把 Toolbox 定义的工具加载为 Python 函数/对象使用；提供 core/ADK/LangChain(LangGraph)/LlamaIndex 等不同集成包，便于接到你的 GenAI 应用里。

- **适合谁：** 想在 Python 代码（含 LangChain/LlamaIndex）里复用 Toolbox 工具的团队
- **可搭配：** Python；MCP Toolbox 服务；可选 ADK/LangChain/LlamaIndex 集成包
- **准备时间：** 10–25 分钟

## 实战建议

- 量化信息：仓库列出了 **4** 个包（core / ADK / LangChain / LlamaIndex）。
- 量化信息：安装是单条 `pip install ...`；验证路径是 docsite 的 Quickstart（可直接复制执行）。

## 推荐用法：Toolbox 作为“工具定义的单一事实源”

如果你已经在用 MCP 让 Agent 调工具，这套 SDK 的价值在于：**同一份工具定义**也能在 Python 应用里复用（批处理、评测、离线流水线）。

推荐结构：

1. **工具定义放在 Toolbox**（唯一事实源）。
2. **交互式流程让 Agent 走 MCP**（可控、可审计）。
3. **批量/离线执行用 SDK**（测试、评测、定时任务）。

这样能避免“Agent 侧工具”和“后端工具”长期漂移。

### FAQ

**必须用 LangChain 才能用吗？**
答：不需要。`toolbox-core` 是框架无关的，可选集成包才针对框架。

**它是服务端吗？**
答：不是。README 说明需要先运行 Toolbox 服务端，这里是 SDK。

**第一步怎么验证？**
答：先选 1 个最简单的工具跑通 docsite Quickstart，再接入 CI 做回归验证。

## 来源与感谢

> Source: https://github.com/googleapis/mcp-toolbox-sdk-python
> License: Apache-2.0
> GitHub stars: 182 · forks: 49


---
Source: https://tokrepo.com/en/workflows/mcp-toolbox-python-sdks-use-toolbox-tools-in-apps
Author: MCP Hub