Correct way to return tool output directly as agent response (skip agent LLM after tool call)?

[ivishalanand]

I’m building a custom agent with multiple tools (a bit like a React agent). For most tools, I want the agent LLM to take the tool’s output, reason further, and then respond.

But for one specific tool, once it generates a response, I want that response to be sent directly back to the user - without sending it back through the agent LLM for rewriting/paraphrasing.

I am not sure how to use return_direct=True to allow for this?

Questions
1. What’s the recommended / idiomatic way in LangGraph to short-circuit tool outputs directly to the user?
2. Is there a built-in equivalent to the old return_direct=True from LangChain tools?

from collections.abc import Sequence
from typing import Annotated, TypedDict

from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import tool
from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph import END, StateGraph
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode
from langgraph.store.memory import InMemoryStore

# --- State ---
class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], add_messages]

# --- Example tools ---
@tool("Search-Docs")
def search_docs(query: str, config: RunnableConfig):
    """Searches and retrieves relevant documents."""
    return f"Results for: {query}"

from langgraph.prebuilt import InjectedState

@tool("Direct-Output-Tool", return_direct=True)
def direct_output_tool(state: Annotated[dict, InjectedState]):
    """This tool should return directly to the user (skip agent)."""
    return "Direct Context Response"

# --- ToolNode with all tools ---
tools = [search_docs, direct_output_tool]
tool_node = ToolNode(tools)

# --- Agent node (simplified) ---
def call_agent(state: AgentState, config: RunnableConfig):
    system_prompt = SystemMessage("Supervisor prompt here")
    response = model.invoke([system_prompt, *state["messages"]], config)
    return {"messages": [response]}

# --- Conditional router ---
def should_continue(state: AgentState):
    last_message = state["messages"][-1]
    if not isinstance(last_message, AIMessage):
        return "end"
    if not last_message.tool_calls:
        return "end"
    return "continue"

# --- Graph wiring ---
workflow = StateGraph(AgentState)
workflow.add_node("agent", call_agent)
workflow.add_node("tools", tool_node)
workflow.set_entry_point("agent")

workflow.add_conditional_edges("agent", should_continue, {
    "continue": "tools",
    "end": END,
})

workflow.add_edge("tools", "agent")

# --- Compile ---
app = workflow.compile(checkpointer=MemorySaver(), store=InMemoryStore())