What Is Langflow? Is It Free, Docker and Desktop Setup, MCP, API, Security and License

In Langflow you place components on a canvas and connect them to build an AI agent or a flow that chats with your documents (RAG); you can run the flow you build through an API or expose it to AI clients as an MCP tool. It is open source (MIT); it is not fully no-code, since every component's Python code can be edited. The software is free: it runs as Langflow Desktop for Windows/Mac, as a pip package with uv, or as a Docker image. The interface is not in Turkish. The official security page describes it as a platform that executes code, and the repository has 15 security advisories for 2026: use the latest version (1.12.6) and don't expose it to the internet.

What it does: Builds agent and RAG flows by connecting components on a canvas
Free: Yes: MIT, self-hosted
Install: Desktop · pip · Docker
Turkish: No Turkish interface
Security: Runs code, stay up to date
Cloud: No pricing page

What is Langflow for? Can I use it without code?

In Langflow you build an AI flow by connecting a chat input, a model, a few tools and an output component on a canvas. You test the flow right away in the built-in Playground, then expose it as an API or an MCP tool. MCP is the standard AI assistants use to connect to external tools.

Build agent and RAG flows with drag and dropWhat you get: A flow tested in the Playground and running as an APINeeds a separate setup Self-host with DockerWhat you get: A server running on port 7860Needs a separate setup Expose flows as MCP toolsWhat you get: MCP toolsNeeds a connection

No-code? You start with the visual canvas and there are ready-made templates (for example Simple Agent); but Langflow also describes itself as a development environment with a code editor: you can open and change the Python code of components. You don't have to write code, but setting up your own server and adding the model provider's key takes technical knowledge. For those who don't want to install anything, the official site promises a free cloud account; plan and limit details are not in the official documentation.

Who it's for: developers, data teams and technically curious business owners who want to prototype an AI agent or a bot that searches their own documents.

How do you install Langflow? Desktop, pip and Docker

There are three ways; the first is the one recommended in the official docs. In all of them, the interface opens in the browser or in the app.

Click Download Langflow at langflow.org/desktop.
Fill in the form: first name, last name, email, country and company. Below the form there is a consent line saying IBM may use this information to send you information.
Mac (macOS 13+): mount the app and install it. Windows: double-click the downloaded .msi file.

If you get a C++ Build Tools Required error on Windows, install Microsoft C++ Build Tools or Visual Studio following the on-screen instructions. According to the docs, Desktop does not have the Shareable Playground or Voice mode.

Requires Python 3.10-3.14 and the uv package manager; at least a dual-core CPU and 2 GB RAM, recommended multi-core and more than 4 GB RAM.

Install› uv pip install langflow

Start› uv run langflow run

The first launch can take a few minutes. Address: http://127.0.0.1:7860. Auto-login is off in official releases: before starting, set a password in the LANGFLOW_SUPERUSER_PASSWORD variable; the username is langflow.

Official image (change the password)

› docker run -p 7860:7860 -e LANGFLOW_AUTO_LOGIN=false -e LANGFLOW_SUPERUSER_PASSWORD=STRONG_PASSWORD langflowai/langflow:latest

Then go to http://localhost:7860 and sign in with the langflow user and the password you set. Official images require authentication by default.

The docs' Docker page has a sentence in the same section saying auto-login is on, while the command does the opposite. The safe option is to set a password as in the command.

Updating: with pip, uv pip install langflow -U. The docs recommend backing up your existing installation before upgrading.

How do you use Langflow? First flow, API and MCP

Add your model key. Profile icon › Settings then Model Providers. Choose a provider, enter the API key and turn on the models you want to use. A local Ollama can also be connected.
Start from a template. New Flow › Simple Agent. In the Agent component, choose the model with Setup Provider.
Test it in the Playground. Ask a math question (the Calculator tool runs) or ask for current news (the URL tool runs). The Playground shows which tool the agent picked.

To run the flow from your own app, first generate a Langflow API key: profile icon › Settings then Langflow API Keys then Add New. Then you call the flow over HTTP:

Run the flow via API (official Quickstart)

› curl --request POST --url "http://LANGFLOW_SERVER_ADDRESS/api/v1/run/FLOW_ID" --header "Content-Type: application/json" --header "x-api-key: $LANGFLOW_API_KEY" --data '{"output_type": "chat", "input_type": "chat", "input_value": "Hello"}'

A Langflow API key carries all the permissions of the user who generated it; don't share it with anyone and don't embed it in code.

Exposing it as MCP: your flow must have a Chat Output component. When you create a project, the flows inside it automatically become MCP tools; if authentication is on, a separate API key is generated for the project. The default transport is streamable HTTP (path /streamable). The docs' example client is Cursor; you connect to Cursor and other MCP-compatible clients with a JSON config. The docs have no Claude-specific step.

The reverse works too: you can connect external MCP servers to a Langflow flow and make their components the agent's tools.

First test in the Playground"Add 4 and 4, then summarize one of today's important tech news stories."If you write in Turkish, whether the model answers in Turkish depends on the model you connect; the official documentation does not guarantee Turkish.

Is Langflow safe? Known vulnerabilities

The official security page describes Langflow as a code execution platform: its built-in code editor can run Python on the server with full privileges, there is no isolation between users, and flows that run AI-generated code are risky. In a multi-user setup, you provide isolation at the infrastructure level.

Issue	Affected versions	Fix
Unauthenticated remote code execution, public flow endpoint (critical, March 2026)	1.8.2 and earlier	1.9.0
Command execution via MCP stdio (critical, September 2026)	1.1.2 – 1.10.2	1.10.3
Python REPL component runs code without a sandbox (critical, September 2026)	Before 1.10.1	1.10.1

Use the latest version: the latest release is 1.12.6 (October 13, 2026), newer than all the fixes above.
Don't expose the port directly to the internet; keep auto-login off, use a strong LANGFLOW_SECRET_KEY, and publish behind an authenticated reverse proxy.
Run flows open to untrusted users or to model-generated code in an isolated container.

The repository has 15 published security advisories for 2026 (March 16 – September 28). This does not mean "no open vulnerabilities right now"; it only means the recorded advisories are fixed in this version.

Is Langflow free? Pricing and cloud

Option	Cost	Note
Self-hosted (Desktop, pip, Docker)	$0 (MIT)	Server, maintenance and updates are on you; model API fees separate
Langflow cloud account	Not stated in the official documentation	The homepage says "free"; no plan or limits page

There is no fee for the software itself. Costs come from two places: API usage of model providers (such as OpenAI and Anthropic; their pricing applies) and a server if you will run the flow continuously. Server costs are not in the official documentation; no estimate is given here.

The official homepage mentions a free cloud account, but langflow.org/pricing does not open and the docs list no cloud plan, limits or prices. Before choosing the cloud, check the plan and terms of use on the site yourself. There is no Turkish lira pricing.

Langflow license, data and Turkish

License

MIT. You are free to use, modify and distribute it commercially; keeping the copyright and license notice is required.

Data

If it runs locally, your flows and data stay with you. For text you send to a connected model provider, that provider's rules apply. Check Langflow's own privacy policy on its official site.

Telemetry

According to the docs, anonymous usage statistics are collected; turn it off with the DO_NOT_TRACK=True setting.

Turkish

The interface is not in Turkish (language files: German, English, Spanish, French, Japanese, Korean, Portuguese, Chinese). Replies in Turkish depend on the model you connect.

Frequently asked questions

What is Langflow and what is it used for?

An open-source tool for building AI agents and document-querying flows on a visual canvas. You can expose a flow as an API or an MCP tool.

Is Langflow free?

The software is free (MIT); you use it on your own computer or server. Model API and server costs are separate. The plan for the free cloud account is not clear in the official documentation.

How do you install Langflow with Docker?

Start Docker and run the langflowai/langflow image with a password (LANGFLOW_SUPERUSER_PASSWORD), then go to localhost:7860. The command is in the Installation section.

Does Langflow work with MCP?

Yes. You can expose your flows as MCP tools to clients like Cursor or connect external MCP servers to your flow. The flow needs a Chat Output component.

Is Langflow safe?

The official page describes it as a platform that executes code, and the repository has 15 security advisories for 2026. Use the latest version and don't expose it directly to the internet.

Is Langflow available in Turkish?

The interface is not in Turkish. Replies in Turkish depend on the model you connect.

Related guides

Sources

Not clear in the official sources: Plan, price and limits of the cloud version (the pricing page does not open) · Langflow's own privacy policy and model training terms · Whether the Company field in the Desktop download form is required · The step for connecting MCP to Claude (the docs only have a Cursor example) · GPU requirements for local models · Turkish lira pricing. Each is covered in the relevant section along with what you can do.