Deploy

langgraph dev runs your graph as a local server. A langgraph.json file tells it where your graph is and what it depends on.

Last updated: 29 Sep, 2026 · LangGraph Platform

Once a graph works, you serve it so a UI or another service can call it. The dev server needs no Docker and reloads as you edit.

The debugging video moves its tool-calling graph into agent.py, where make_tool_graph() builds and compiles it and the variable tool_agent holds the result. Next to it goes a langgraph.json that points at tool_agent as "./agent.py:tool_agent", with env set to "../.env", the keys file one folder up. langgraph dev, from langgraph-cli[inmem], run in that folder, starts a local server and opens the graph in LangGraph Studio, where you can send a message, watch each node run, and pause the run at a node. The same server lists its API routes on a docs page. Below, the same file is built one field at a time for the support graph.

The langgraph dev command

pip install "langgraph-cli[inmem]"
langgraph dev            # starts a local server with hot reload

The dev server reads a langgraph.json file to find your graph. Build that file up one field at a time.

The graphs field

{
  "graphs": { "support": "./your_package/agent.py:graph" }
}

graphs is required. It maps a name to a pointer of the form "./file.py:variable". Here the server imports the graph variable from agent.py and serves it under the name support.

The dependencies field

{
  "graphs": { "support": "./your_package/agent.py:graph" },
  "dependencies": ["langchain", "./your_package"]
}

dependencies lists what to install: packages by name, such as "langchain", and your own code as a local path, such as "./your_package".

The env field

{
  "graphs": { "support": "./your_package/agent.py:graph" },
  "dependencies": ["langchain", "./your_package"],
  "env": "./.env"
}

env points at a .env file so the server can read your keys and settings. This field is optional.

The complete config file

All three fields together, saved next to your code.

{
  "dependencies": ["langchain", "./your_package"],
  "graphs": { "support": "./your_package/agent.py:graph" },
  "env": "./.env"
}

With this file in place, run langgraph dev and the server loads the support graph.

What each field does

• graphs maps a name to a pointer of the form "./file.py:variable", where variable is the compiled graph the server imports.
• dependencies lists the packages and local folders to install.
• env points at a .env file for keys and settings.

Required vs optional fields

When to serve a graph

• Running your agent behind an endpoint a UI can call.
• The step between a script on your machine and a served service.

Related

• Previous: Integrations
• Next: LangSmith
• Reference: Application structure