langgraph-cli

Reference and operational guidance for the `langgraph` CLI, which manages the full lifecycle of LangGraph applications from scaffolding a new project to deploying it to LangGraph Platform (LangSmith Deployments). It is meant to be invoked when scaffolding, developing, building, or deploying LangGraph applications, or when understanding and editing the `langgraph.json` configuration that every command except `new` reads from the project root.

The CLI exposes a focused command set. `langgraph new` scaffolds a project from a template such as `agent-python` or `new-langgraph-project-js`. `langgraph dev` runs a local development server with hot reloading and no Docker, defaulting to localhost:2024, with flags for custom port, explicit config, disabling reload or the browser, binding to all interfaces, exposing a Cloudflare tunnel, and remote debugging. `langgraph build` builds a Docker image (the `-t` tag is required), `langgraph up` launches via Docker Compose including Postgres on default port 8123, and `langgraph deploy` builds and ships to LangGraph Platform, requiring Docker and, on Apple Silicon, Docker Buildx for cross-compiling to linux/amd64. Deploy subcommands cover list, delete, and logs, the latter supporting build versus runtime logs, follow streaming, level filtering, search, limits, and time ranges. `langgraph dockerfile` generates a Dockerfile and optionally Compose files without building.

Installation differs by ecosystem: Python uses `pip install 'langgraph-cli[inmem]'` (the inmem extra includes dev server support) or plain `langgraph-cli` for build, up, and deploy only, with equivalent `uv add` commands; JavaScript uses `npx @langchain/langgraph-cli` or a global npm install exposed as `langgraphjs`. The `langgraph.json` file is shared by dev, build, up, and deploy and requires `dependencies` (where `.` resolves local packages via pyproject.toml, setup.py, requirements.txt, or package.json) and `graphs` (a mapping of graph ID to a `file:variable` path that must be a CompiledGraph or a function returning one), plus optional keys such as `env`, `python_version`, `pip_config_file`, and `dockerfile_lines`. Deploying requires a LANGSMITH_API_KEY in the environment or `.env`.