# LangChain MCP Adapters Ref: - https://github.com/langchain-ai/langchain-mcp-adapters This notebook demonstrates how to use MCP (Model Context Protocol) tools with LangChain agents using `langchain-mcp-adapters`. ## Setup Configure `.env` before running. See `.env.sample`. ```python import rich from dotenv import load_dotenv load_dotenv() ``` True ## MCP Servers MCP server files are located in `mcp_servers/` directory. ```python from pathlib import Path from rich.syntax import Syntax mcp_servers_dir = Path("mcp_servers") math_server = mcp_servers_dir / "math_server.py" string_server = mcp_servers_dir / "string_server.py" rich.print(Syntax(math_server.read_text(), "python", padding=(1, 2))) ``` from mcp.server.fastmcp import FastMCP mcp = FastMCP("Math") @mcp.tool(name="math__add") def add(a: int, b: int) -> int: """Add two numbers.""" return a + b @mcp.tool(name="math__multiply") def multiply(a: int, b: int) -> int: """Multiply two numbers.""" return a * b if __name__ == "__main__": mcp.run(transport="stdio") ## Load MCP Tools Use `load_mcp_tools` to load tools from an MCP server session. **Note:** There are two ways to load tools: - `load_mcp_tools(session)`: Uses an existing session. Tools execute within the same session. More efficient for multiple tool calls. - `client.get_tools()`: Creates a new session for each tool execution. Simpler but has connection overhead per call. Useful for stateless deployments like LangGraph API Server. ```python from langchain_mcp_adapters.tools import load_mcp_tools from mcp import ClientSession, StdioServerParameters from mcp.client.stdio import stdio_client server_params = StdioServerParameters( command="python", args=[str(math_server.resolve())], ) async def list_tools() -> None: """List available tools from the MCP server.""" async with stdio_client(server_params) as (read, write): async with ClientSession(read, write) as session: await session.initialize() tools = await load_mcp_tools(session) rich.print("tools =", tools) await list_tools() ``` tools = [ StructuredTool( name='math__add', description='Add two numbers.', args_schema={ 'properties': {'a': {'title': 'A', 'type': 'integer'}, 'b': {'title': 'B', 'type': 'integer'}}, 'required': ['a', 'b'], 'title': 'addArguments', 'type': 'object' }, response_format='content_and_artifact', coroutine=.call_tool at 0x112dcaac0> ), StructuredTool( name='math__multiply', description='Multiply two numbers.', args_schema={ 'properties': {'a': {'title': 'A', 'type': 'integer'}, 'b': {'title': 'B', 'type': 'integer'}}, 'required': ['a', 'b'], 'title': 'multiplyArguments', 'type': 'object' }, response_format='content_and_artifact', coroutine=.call_tool at 0x112dcb1a0> ) ] ## Create Agent with MCP Tools Load tools from the MCP server and create a LangChain agent. ```python from typing import Any from langchain.agents import create_agent from langchain_anthropic import ChatAnthropic from langgraph.graph.state import CompiledStateGraph model = ChatAnthropic(model="claude-sonnet-4-5-20250929") async def run_agent() -> None: """Run agent with MCP tools.""" async with stdio_client(server_params) as (read, write): async with ClientSession(read, write) as session: await session.initialize() tools = await load_mcp_tools(session) agent: CompiledStateGraph[Any] = create_agent( model=model, tools=tools, system_prompt="You are a helpful math assistant.", ) response = await agent.ainvoke({"messages": [{"role": "user", "content": "What is 5 + 3?"}]}) rich.print("response =", response) await run_agent() ``` response = { 'messages': [ HumanMessage( content='What is 5 + 3?', additional_kwargs={}, response_metadata={}, id='934fdfbb-9f0a-4fd4-932a-a1a7af5709a4' ), AIMessage( content=[ { 'id': 'toolu_013up3WT5Ubnsw7oerjJQj7y', 'input': {'a': 5, 'b': 3}, 'name': 'math__add', 'type': 'tool_use' } ], additional_kwargs={}, response_metadata={ 'id': 'msg_01Se6RdeWyn28BCWgiMtKEo2', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'tool_use', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 649, 'output_tokens': 70, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--684d7f12-8ed8-442f-935a-b9cca3dbcac2-0', tool_calls=[ { 'name': 'math__add', 'args': {'a': 5, 'b': 3}, 'id': 'toolu_013up3WT5Ubnsw7oerjJQj7y', 'type': 'tool_call' } ], usage_metadata={ 'input_tokens': 649, 'output_tokens': 70, 'total_tokens': 719, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ), ToolMessage( content='8', name='math__add', id='a9d3722b-d0b4-4609-8b14-7e4fa8020c72', tool_call_id='toolu_013up3WT5Ubnsw7oerjJQj7y' ), AIMessage( content='The answer is **8**.', additional_kwargs={}, response_metadata={ 'id': 'msg_01Ws6WXpmPCgrdPfS5uU9Wd7', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'end_turn', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 733, 'output_tokens': 9, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--d6223114-2a27-46e5-9571-6241a0565bf3-0', usage_metadata={ 'input_tokens': 733, 'output_tokens': 9, 'total_tokens': 742, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ) ] } ## Stream with MCP Tools You can also stream responses when using MCP tools. ```python async def stream_agent() -> None: """Stream agent output with MCP tools.""" async with stdio_client(server_params) as (read, write): async with ClientSession(read, write) as session: await session.initialize() tools = await load_mcp_tools(session) agent: CompiledStateGraph[Any] = create_agent( model=model, tools=tools, system_prompt="You are a helpful math assistant.", ) async for chunk in agent.astream({"messages": [{"role": "user", "content": "What is 7 * 8?"}]}): rich.print("chunk =", chunk) await stream_agent() ``` chunk = { 'model': { 'messages': [ AIMessage( content=[ { 'id': 'toolu_01NHme3fRFwVjQ6sfLKdqVrh', 'input': {'a': 7, 'b': 8}, 'name': 'math__multiply', 'type': 'tool_use' } ], additional_kwargs={}, response_metadata={ 'id': 'msg_011BpoCH4Hu2o4sYGv3pFehk', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'tool_use', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 649, 'output_tokens': 70, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--0fb52d04-0f4c-4ec2-a3b1-09560f41044a-0', tool_calls=[ { 'name': 'math__multiply', 'args': {'a': 7, 'b': 8}, 'id': 'toolu_01NHme3fRFwVjQ6sfLKdqVrh', 'type': 'tool_call' } ], usage_metadata={ 'input_tokens': 649, 'output_tokens': 70, 'total_tokens': 719, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ) ] } } chunk = { 'tools': { 'messages': [ ToolMessage( content='56', name='math__multiply', id='788ba5d8-4e4f-472c-b98e-60af170b5c99', tool_call_id='toolu_01NHme3fRFwVjQ6sfLKdqVrh' ) ] } } chunk = { 'model': { 'messages': [ AIMessage( content='7 × 8 = **56**', additional_kwargs={}, response_metadata={ 'id': 'msg_013uBEweZnqapQBAw4cxPBCK', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'end_turn', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 733, 'output_tokens': 13, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--6b18b6e8-1d6b-4bda-9965-f93eb86d7344-0', usage_metadata={ 'input_tokens': 733, 'output_tokens': 13, 'total_tokens': 746, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ) ] } } ## Multiple MCP Servers Use `MultiServerMCPClient` with `load_mcp_tools` for explicit session control. See `mcp_servers/string_server.py` for the string server implementation. ```python rich.print(Syntax(string_server.read_text(), "python", padding=(1, 2))) ``` from mcp.server.fastmcp import FastMCP mcp = FastMCP("String") @mcp.tool(name="string__uppercase") def uppercase(text: str) -> str: """Convert text to uppercase.""" return text.upper() @mcp.tool(name="string__reverse") def reverse(text: str) -> str: """Reverse a string.""" return text[::-1] if __name__ == "__main__": mcp.run(transport="stdio") ```python from contextlib import AsyncExitStack from langchain_mcp_adapters.client import MultiServerMCPClient multi_client = MultiServerMCPClient( { "math": { "command": "python", "args": [str(math_server.resolve())], "transport": "stdio", }, "string": { "command": "python", "args": [str(string_server.resolve())], "transport": "stdio", }, } ) async def run_multi_server_agent() -> None: """Run agent with multiple MCP servers.""" async with AsyncExitStack() as stack: math_session = await stack.enter_async_context(multi_client.session("math")) string_session = await stack.enter_async_context(multi_client.session("string")) tools = [] tools.extend(await load_mcp_tools(math_session)) tools.extend(await load_mcp_tools(string_session)) rich.print("Available tools:", [t.name for t in tools]) agent: CompiledStateGraph[Any] = create_agent( model=model, tools=tools, system_prompt="You are a helpful assistant with math and string tools.", ) response = await agent.ainvoke( {"messages": [{"role": "user", "content": "Add 10 and 20, then reverse the string 'hello'"}]} ) rich.print("response =", response) await run_multi_server_agent() ``` Available tools: ['math__add', 'math__multiply', 'string__uppercase', 'string__reverse'] response = { 'messages': [ HumanMessage( content="Add 10 and 20, then reverse the string 'hello'", additional_kwargs={}, response_metadata={}, id='88a5e35c-e851-481b-8c6c-84b5c542e8b4' ), AIMessage( content=[ { 'text': "I'll help you add 10 and 20, and reverse the string 'hello'. Since these operations are independent, I can do both at the same time.", 'type': 'text' }, { 'id': 'toolu_01MEvKzTQf2aM1SCdnLq4LCb', 'input': {'a': 10, 'b': 20}, 'name': 'math__add', 'type': 'tool_use' }, { 'id': 'toolu_015QWvHHJiswy4ECSp3AAEet', 'input': {'text': 'hello'}, 'name': 'string__reverse', 'type': 'tool_use' } ], additional_kwargs={}, response_metadata={ 'id': 'msg_012og99rbiKwVjT7Den1LVUW', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'tool_use', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 764, 'output_tokens': 142, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--f133a1cd-4f84-43e3-8af4-29d6778f02b3-0', tool_calls=[ { 'name': 'math__add', 'args': {'a': 10, 'b': 20}, 'id': 'toolu_01MEvKzTQf2aM1SCdnLq4LCb', 'type': 'tool_call' }, { 'name': 'string__reverse', 'args': {'text': 'hello'}, 'id': 'toolu_015QWvHHJiswy4ECSp3AAEet', 'type': 'tool_call' } ], usage_metadata={ 'input_tokens': 764, 'output_tokens': 142, 'total_tokens': 906, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ), ToolMessage( content='30', name='math__add', id='5d24c11a-03ed-4ef9-bf7f-aea639c1f9fd', tool_call_id='toolu_01MEvKzTQf2aM1SCdnLq4LCb' ), ToolMessage( content='olleh', name='string__reverse', id='58bf274e-09c4-4938-bead-1a32759e49be', tool_call_id='toolu_015QWvHHJiswy4ECSp3AAEet' ), AIMessage( content="Perfect! Here are the results:\n- **10 + 20 = 30**\n- **'hello' reversed = 'olleh'**", additional_kwargs={}, response_metadata={ 'id': 'msg_01DMUxaxmYHnEgx5Z9q6rJQi', 'model': 'claude-sonnet-4-5-20250929', 'stop_reason': 'end_turn', 'stop_sequence': None, 'usage': { 'cache_creation': {'ephemeral_1h_input_tokens': 0, 'ephemeral_5m_input_tokens': 0}, 'cache_creation_input_tokens': 0, 'cache_read_input_tokens': 0, 'input_tokens': 972, 'output_tokens': 37, 'server_tool_use': None, 'service_tier': 'standard' }, 'model_name': 'claude-sonnet-4-5-20250929', 'model_provider': 'anthropic' }, id='lc_run--dfb6edd0-0439-467f-9fb4-e0c7fadbc211-0', usage_metadata={ 'input_tokens': 972, 'output_tokens': 37, 'total_tokens': 1009, 'input_token_details': { 'cache_read': 0, 'cache_creation': 0, 'ephemeral_5m_input_tokens': 0, 'ephemeral_1h_input_tokens': 0 } } ) ] }