🔀 Langflow：147k Stars 的可视化 AI Agent 编排平台，拖拽式搭建工作流，还能当 MCP Server 用
每次想搭一个 AI 工作流，你是不是也得先装 LangChain、配 API Key、写几十行 callback 函数？老实说，我踩过这个坑太多次了。后来发现 Langflow——一个让你用拖拽代替写代码的 AI Agent 编排平台，147k Stars，Python 写的，底层用 React Flow 做可视化引擎，支持所有主流 LLM 和向量数据库。
项目数据： 147,926 Stars | Python | MIT 协议 | 日下载量超 10 万 | 支持 Windows/macOS Desktop
它能干啥
🔀 可视化拖拽构建 — 不用写代码。拖个 LLM 节点，拖个 Prompt 节点，连起来，就跑通了。所有组件用 Python 写的，你随时可以打开源码改逻辑。
⚡ 内置 MCP Server — 这招最骚。你搭好的工作流可以直接部署成 MCP Server，别的 AI Agent 通过 MCP 协议直接调你的流当工具用。相当于你搭一个链，它就变成了一个 API。
🛠 多智能体编排 — 支持多 Agent 对话管理 + 检索增强。不是单链条，是真·多智能体协同。
📦 一键部署 API — 工作流搭好后，点一下就能部署成 REST API，或者导出 JSON 给 Python 应用用。
3 步上手
装 Langflow 前你需要 Python 3.10–3.13 和 uv（推荐）:
# 1. 安装
uv pip install langflow -U
# 2. 启动
uv run langflow run
浏览器打开 http://127.0.0.1:7860，就能看到可视化界面了。
想用 Docker？一行搞定：
docker run -p 7860:7860 langflowai/langflow:latest
不想折腾环境？官方有 Desktop 版，Windows 和 macOS 直接下载安装包，依赖全打包好了。
和 ComfyUI 有点像，但方向不同
用过 ComfyUI 的人看到 Langflow 应该很眼熟——都是节点编排。但 ComfyUI 管的是 Stable Diffusion 的图片生成管线，Langflow 管的是 LLM + Agent 的逻辑编排。一个是画图的，一个是干活的。
实际开发中注意
Langflow 的 MCP Server 功能是 2.0 之后加的，如果你跑旧版本，记得升级到最新。还有，节点多了之后界面会有点卡，建议把复杂工作流拆成子流。
要点总结：
🔀 Langflow: 147k Stars Visual AI Agent Orchestration Platform — Drag, Drop, and Deploy Workflows as MCP Servers
Every time you want to build an AI workflow, it's the same story — install LangChain, configure API keys, write dozens of lines of callback functions. I've been there too many times. Then I found Langflow: a visual AI agent builder that replaces code with drag-and-drop. 147k Stars, built with Python and React Flow, supports every major LLM and vector database out there.
Project stats: 147,926 Stars | Python | MIT License | 100k+ daily downloads | Windows/macOS Desktop
What it does
🔀 Visual drag-and-drop builder — Zero code required. Drop an LLM node, drop a Prompt node, connect them, and it runs. Every component is written in Python, so you can open the source and tweak it anytime.
⚡ Built-in MCP Server — Here's the killer feature. Your finished workflow can be deployed as an MCP Server instantly. Other AI agents call your flow as a tool via the MCP protocol. Build a chain, get an API.
🛠 Multi-agent orchestration — Conversation management + retrieval augmented generation for real multi-agent collaboration, not just a single chain.
📦 One-click API deployment — Export your flow as a REST API or JSON for Python apps.
Getting started in 3 steps
You'll need Python 3.10–3.13 and uv (recommended):
# 1. Install
uv pip install langflow -U
# 2. Run
uv run langflow run
Open http://127.0.0.1:7860 in your browser and you're looking at the visual editor.
Prefer Docker?
docker run -p 7860:7860 langflowai/langflow:latest
Don't want to deal with environments? Grab the Desktop app for Windows or macOS — all dependencies included.
Think of it as ComfyUI, but for LLMs
If you've used ComfyUI, Langflow will feel familiar — same node-based visual programming. ComfyUI orchestrates Stable Diffusion image pipelines; Langflow orchestrates LLM + Agent logic. One draws pictures, the other gets work done.
A few things I learned
The MCP Server feature landed in Langflow 2.0+. If you're running an older version, upgrade first. Also, the UI can get sluggish with too many nodes — split complex flows into sub-flows.
Quick recap: