# langchain-mcp-adapters — MCP Tools in LangGraph

> Use MCP servers inside LangChain/LangGraph by loading MCP tools into agents via MultiServerMCPClient, without writing one-off wrappers per server.

## Install

Save as a script file and run:

# langchain-mcp-adapters — MCP Tools in LangGraph

> Use MCP servers inside LangChain/LangGraph by loading MCP tools into agents via MultiServerMCPClient, without writing one-off wrappers per server.

## Quick Use

1. Install:
   ```bash
   pip install langchain-mcp-adapters
   ```
2. Run:
   ```bash
   python -c "from langchain_mcp_adapters.client import MultiServerMCPClient; print('ok')"
   ```
3. Verify:
   - Connect to one local MCP server and confirm `get_tools()` returns a non-empty list.


---

## Intro

Use MCP servers inside LangChain/LangGraph by loading MCP tools into agents via MultiServerMCPClient, without writing one-off wrappers per server.

- **Best for:** LangGraph builders who want to reuse MCP tools across agents with a single client layer
- **Works with:** Python, LangChain/LangGraph, MCP servers over stdio/HTTP/SSE (per repo docs)
- **Setup time:** 12 minutes


### Quantitative Notes

- GitHub stars (verified): see Source & Thanks
- GitHub forks (verified): see Source & Thanks
- Setup time ~12 minutes (Python env + one demo server)


---

## Practical Notes

Treat MCP as your **tool surface**, and LangGraph as your **orchestration surface**. A practical pattern is: (1) run one MCP server locally (stdio) for filesystem or a custom tool, (2) connect via MultiServerMCPClient, (3) feed the resulting tool list into a tool-aware agent. Once that works, add a second server and verify tool names, permissions, and error handling. TokRepo related reading: https://tokrepo.com/en/topics

**Safety note:** Keep MCP connections scoped (servers, tools, env vars). Don’t forward unrestricted shell/file tools into production agents.

### FAQ

**Q: What does it actually do?**
A: It converts MCP tools into LangChain-compatible tools, so LangGraph agents can call them like any other tool node.

**Q: Can it connect to multiple MCP servers?**
A: Yes. You can configure multiple server connections and load tools across them in one client.

**Q: How do I avoid tool name collisions?**
A: Prefix tool names per server (or enable the library option) so `search` from different servers stays unique.

---

## Source & Thanks

> GitHub: https://github.com/langchain-ai/langchain-mcp-adapters
> Owner avatar: https://avatars.githubusercontent.com/u/126733545?v=4
> License (SPDX): MIT
> GitHub stars (verified via `api.github.com/repos/langchain-ai/langchain-mcp-adapters`): 3,526
> GitHub forks (verified via `api.github.com/repos/langchain-ai/langchain-mcp-adapters`): 424


---

<!-- ZH -->

# langchain-mcp-adapters——在 LangGraph 里用 MCP 工具

> 把 MCP server 的 tools/prompts/resources 转为 LangChain/LangGraph 可用工具：用 MultiServerMCPClient 连接多 server，并像原生 tool 一样调用。

## 快速使用

1. 安装：
   ```bash
   pip install langchain-mcp-adapters
   ```
2. 运行：
   ```bash
   python -c "from langchain_mcp_adapters.client import MultiServerMCPClient; print('ok')"
   ```
3. 验证：
   - Connect to one local MCP server and confirm `get_tools()` returns a non-empty list.


---

## 简介

把 MCP server 的 tools/prompts/resources 转为 LangChain/LangGraph 可用工具：用 MultiServerMCPClient 连接多 server，并像原生 tool 一样调用。

- **适合谁（Best for）:** 想在 LangGraph agent 中复用 MCP 工具、并希望用统一 client 管理多 server 的开发者
- **兼容工具（Works with）:** Python、LangChain/LangGraph、通过 stdio/HTTP/SSE 暴露的 MCP servers（仓库文档）
- **安装时间（Setup time）:** 12 分钟


### 量化信息

- GitHub stars（已核验）：见「来源与感谢」
- GitHub forks（已核验）：见「来源与感谢」
- 安装/跑通约 12 分钟（准备 Python 环境 + 一个 demo server）


---

## 实战要点

把 MCP 当作「工具面」（tool surface），把 LangGraph 当作「编排面」（orchestration surface）。推荐的落地路径是：先本地跑一个 MCP server（stdio），再用 MultiServerMCPClient 连接并拉取工具列表，把工具交给支持 tool 的 agent；跑通后再加第二个 server，重点检查同名工具、权限边界和错误处理。TokRepo 相关阅读：https://tokrepo.com/en/topics

**安全提示：** 把 MCP 连接做成「最小权限」：限制 server、工具集合与环境变量；不要把无限制的 shell/文件工具带进生产 agent。

### FAQ

**Q: 它到底做了什么？**
A: 它把 MCP tools 转成 LangChain 可用的 tool 形态，让 LangGraph agent 能像调用普通工具一样调用 MCP 工具。

**Q: 能连多个 MCP server 吗？**
A: 可以。你可以配置多个 server 的连接参数，然后一次性拉取并加载所有工具。

**Q: 不同 server 同名工具会冲突吗？**
A: 可能会。实践中建议给工具名加 server 前缀，或启用库的前缀选项，避免 `search` 之类重名。

---

## 来源与感谢

> GitHub：https://github.com/langchain-ai/langchain-mcp-adapters
> Owner avatar：https://avatars.githubusercontent.com/u/126733545?v=4
> 许可证（SPDX）：MIT
> GitHub stars（已通过 `api.github.com/repos/langchain-ai/langchain-mcp-adapters` 核验）：3,526
> GitHub forks（已通过 `api.github.com/repos/langchain-ai/langchain-mcp-adapters` 核验）：424


---
Source: https://tokrepo.com/en/workflows/langchain-mcp-adapters-mcp-tools-in-langgraph
Author: Agent Toolkit