Overview
Langflow is a drag-and-drop canvas for building LLM applications that compiles down to real Python. You wire together components - model calls, retrievers, vector stores, memory, tools, conditional routing - and Langflow exposes the finished graph as an API endpoint or an MCP server. It began in 2023 as a LangChain visual layer, was acquired by DataStax in 2024, and is now framework-agnostic: the LangChain dependency is optional rather than structural. The project carries an MIT license and over 150,000 GitHub stars, and ships a packaged desktop app for macOS and Windows that spares you the usual Python environment wrangling. Two things separate it from other visual builders. First, bi-directional MCP support - a Langflow flow can both consume MCP servers and be published as one, which is unusual since most builders are clients only. Second, an inline custom-component escape hatch, so you can drop raw Python into a node instead of hitting a wall when the visual paradigm runs out. The honest limitation is scale: flows get hard to read past roughly 30 nodes, version control is JSON export rather than git-native diffs, and community component quality varies. Treat it as a prototyping and internal-tools platform, not a production application framework.
Key Features
- Visual drag-and-drop canvas with 200+ prebuilt components
- Bi-directional MCP: consume MCP servers and publish flows as MCP servers
- Every flow exposed as a REST API endpoint out of the box
- Inline custom Python components when the visual layer runs out
- Native RAG and Graph RAG retrieval pipelines
- MIT-licensed and self-hostable via pip or Docker
- Free desktop app for macOS and Windows with dependencies bundled
Pricing
| Plan | Price | For |
|---|---|---|
| Open Source | $0 (MIT) | Self-hosters, full feature access |
| Langflow Desktop | $0 | Local prototyping on Mac and Windows |
| Self-hosted VM | ~$20-100/mo infrastructure | Solo devs and small production flows |
| Enterprise | Custom (DataStax / IBM) | HA clusters, managed DB, compliance |
Comparison
Compared to Flowise, Langflow’s Python-native components and MCP publishing give it more reach, while Flowise stays lighter and faster to deploy for simple chatbots. Compared to Dify, Langflow is the developer’s tool - Dify ships a fuller ops layer with datasets, annotation and usage analytics, whereas Langflow assumes you bring your own observability. n8n is the better pick when the workflow is mostly SaaS plumbing with one LLM step; Langflow wins when the LLM graph itself is the product. Budget accordingly: the software is free, but you still pay for hosting and model tokens.