Is the LangGraph dev server safe to expose publicly?

No. It is an in-memory development tool. For anything public, use a production deployment with authentication, persistent storage, and monitoring.

Do I need LangSmith to use LangGraph?

LangGraph itself works without it. The local dev server and Studio workflow described here use a LangSmith API key, and tracing is a recommended, optional way to observe your runs.

LangGraph Module 9: Local Server, Studio, and Deployment

Introduction

Scripts are fine for learning, but real applications need a server with an API, persistence, and good debugging tools. The LangGraph CLI provides a local development server and connects to Studio, a visual environment for running and inspecting graphs. This final module sets those up and discusses production options.

Prerequisites

Python 3.11 or newer
A free LangSmith account and API key
A graph you want to serve

Step-by-Step Guide

Step 1: Install the CLI

The CLI requires Python 3.11 or newer. The inmem extra includes the in-memory server used for local development.

pip install -U "langgraph-cli[inmem]"

Step 2: Create a project from a template

langgraph new scaffolds an application. Install the project in editable mode, then copy the example environment file to .env and add your LangSmith API key. Without a template option, the command shows an interactive menu.

langgraph new my-agent --template new-langgraph-project-python
cd my-agent
pip install -e .

# create .env from .env.example and set:
# LANGSMITH_API_KEY=your-langsmith-key

Step 3: Understand langgraph.json

A configuration file named langgraph.json tells the server which graphs to expose. The generated project includes one; a minimal version points a graph name at a variable holding your compiled graph. Check the application structure documentation for all options.

{
  "dependencies": ["."],
  "graphs": {
    "agent": "./src/agent/graph.py:graph"
  },
  "env": ".env"
}

Step 4: Start the dev server and open Studio

langgraph dev starts a local server and prints three addresses: the API at 127.0.0.1:2024, a link that opens Studio against your local server, and the interactive API documentation at /docs. In Studio you can run the graph, watch each step, inspect and edit state, and replay from earlier points. On Safari, add the --tunnel option. Under this server you compile without a checkpointer, because the server provides persistence.

langgraph dev

Step 5: Call the running graph and plan for production

Use the SDK (pip install langgraph-sdk) or plain HTTP to call the server, here with the synchronous client. The dev server is in-memory and meant for development only. For production, deploy the application with LangSmith Deployment or run the open-source server in your own infrastructure with a durable database, authentication, and monitoring.

from langgraph_sdk import get_sync_client

client = get_sync_client(url="http://localhost:2024")

for chunk in client.runs.stream(
    None,   # no thread: a stateless run
    "agent",
    input={"messages": [{"role": "human", "content": "What is 3 + 4?"}]},
    stream_mode="messages-tuple",
):
    print(chunk.event, chunk.data)

FAQs

Is the LangGraph dev server safe to expose publicly?: No. It is an in-memory development tool. For anything public, use a production deployment with authentication, persistent storage, and monitoring.
Do I need LangSmith to use LangGraph?: LangGraph itself works without it. The local dev server and Studio workflow described here use a LangSmith API key, and tracing is a recommended, optional way to observe your runs.

Summary

Use the LangGraph CLI to scaffold a project, describe your graphs in langgraph.json, run langgraph dev to get an API and Studio, and move to a durable, authenticated deployment for production. That completes the course from first graph to deployable agent.