Wiring an LLM to MCP Tools

The missing piece

Everything through Lesson 14 called tools yourself, with hardcoded names and arguments. The whole point of MCP, though, is letting an AI decide which tool to call and with what arguments, exactly like LangChain's tool-calling lesson did with local @tool functions. This lesson connects those two worlds: MCP tools on one side, a LangChain model on the other.

langchain-mcp-adapters: the bridge

langchain-mcp-adapters converts MCP tools into ordinary LangChain BaseTool objects, so anything you already know how to do with bind_tools just works.

from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "calculator": {
        "transport": "stdio",
        "command": "python",
        "args": [str(server_script)],
    }
})
tools = await client.get_tools()

MultiServerMCPClient takes a dict of named server connections ("calculator" is just a label you choose) and get_tools() launches each server, lists its tools, and wraps every one as a LangChain tool. Notice there's no manual stdio_client/ClientSession here, the adapter manages the connection lifecycle for you.

bind_tools, same as always

from langchain_google_genai import ChatGoogleGenerativeAI

model = ChatGoogleGenerativeAI(model="gemini-3.5-flash-lite").bind_tools(tools)
response = model.invoke("What is 12 plus 30? Use the tool.")
response.tool_calls
# [{"name": "add", "args": {"a": 12, "b": 30}, "id": "...", "type": "tool_call"}]

This is identical to LangChain's tool-calling lesson. The model has no idea these tools came from a separate server over a protocol, it just sees LangChain tools with a name, description, and schema, precisely the "the AI never sees the implementation" principle from Lesson 3, now crossing a process boundary and a LangChain adapter on its way to Gemini.

What this lesson doesn't do yet

Gemini decided to call add, but nothing here actually executes that call and feeds the result back. That loop, ask, call, respond, is exactly LangChain's manual tool-calling loop, and it's Lesson 16's subject.