Langgraph auth inject from runnable config

LangGraph Auth Inject Node

When to use

LangGraph servers expose authenticated user info in config["configurable"]["langgraph_auth_user"]. You want that info inside state (so every node can access it without re-parsing config), and you want graceful fallback when running without auth (CLI, tests).

How it works

A single LangGraph node is set as the entry point. It reads from config["configurable"]["langgraph_auth_user"] plus thread_id/run_id, falls back to existing state keys, and returns a partial-update dict that LangGraph merges into state.

Example

def _extract_auth(state, config) -> dict[str, str]:
    configurable = config.get("configurable", {})
    auth = configurable.get("langgraph_auth_user", {})
    thread_id = configurable.get("thread_id", "") or state.get("thread_id", "")
    run_id    = configurable.get("run_id", "")    or state.get("run_id", "")
    return {
        "org_id":        auth.get("org_id")           or state.get("org_id", ""),
        "user_id":       auth.get("identity")         or state.get("user_id", ""),
        "user_email":    auth.get("email", ""),
        "user_name":     auth.get("full_name", ""),
        "organization_slug": auth.get("organization_slug", ""),
        "thread_id":     thread_id,
        "run_id":        run_id,
    }

def inject_auth_node(state, config):
    return _extract_auth(state, config)

graph.add_node("inject_auth", inject_auth_node)
graph.set_entry_point("inject_auth")

Gotchas

• The node receives config as a positional argument because it has two parameters; LangGraph binds it automatically.
• Always fall back to state values so CLI/tests that pre-populate state still work.
• Keep the partial dict flat — nesting auth in a sub-key would force every downstream node to re-flatten.