LangGraph

In LangGraph, MFS is a retrieval node: one step in the graph fans a query out across every indexed source and puts the results into state. It reuses the same MFSRetriever from the LangChain page — define that first.

Install

pip install langgraph langchain-core
# plus the MFS Python SDK (mfs_sdk) — see the SDKs page

A retrieve → generate graph

from typing import TypedDict

from langgraph.graph import START, END, StateGraph

retriever = mfs_retriever(top_k=4)  # from the LangChain page


class State(TypedDict):
    question: str
    context: str
    answer: str


def retrieve(state: State) -> State:
    hits = retriever.invoke(state["question"])
    return {"context": "\n\n".join(h.page_content for h in hits)}


def generate(state: State) -> State:
    # standard LangChain generation — swap in your LLM and prompt
    answer = llm.invoke(f"Answer using only this context:\n{state['context']}\n\nQ: {state['question']}")
    return {"answer": answer.content}


graph = StateGraph(State)
graph.add_node("retrieve", retrieve)
graph.add_node("generate", generate)
graph.add_edge(START, "retrieve")
graph.add_edge("retrieve", "generate")
graph.add_edge("generate", END)
app = graph.compile()

app.invoke({"question": "what happens when a client exceeds the rate limit?"})

The retrieve node is the only MFS-specific part — everything else is ordinary LangGraph. Because MFS searches every connected source in one call, that single node covers code, docs, chat, and databases at once; there is no per-source retriever to wire up or keep in sync.

As a tool instead of a fixed node

For an agentic graph, expose MFS as a tool the model calls when it decides it needs context, rather than a mandatory step:

from langchain_core.tools import create_retriever_tool

mfs_tool = create_retriever_tool(
    retriever, "search_mfs", "Search the team's code, docs, chat, and databases via MFS."
)
# bind mfs_tool to your model / ToolNode as usual

Scope a node or tool with mfs_retriever(scope="…") when a step should look at one source, or leave it open to search everything.