Skip to main content
Langchain is a comprehensive framework for developing applications powered by language models. Its agent capabilities are particularly well-suited for using tools. Universal MCP provides adapters to make its tools easily consumable by Langchain agents, especially those built with LangGraph.

​
Core Steps

  1. Initialize your LLM: Choose and configure the language model you want your agent to use (e.g., ChatOpenAI).
  2. Initialize ToolManager: This is the central registry for your MCP tools. 
    from universal_mcp.tools import ToolManager
tool_manager = ToolManager()
  3. Load Applications and Register Tools:
    • Load your desired Application instances (e.g., using app_from_slug or direct instantiation).
    • Register their tools with the ToolManager using tool_manager.register_tools_from_app(your_app_instance).
    • Add any custom standalone tools using tool_manager.add_tool(your_function).
  4. Convert Tools to Langchain Format: The ToolManager can list tools in a format compatible with Langchain. 
    from universal_mcp.tools.adapters import ToolFormat
langchain_tools = tool_manager.list_tools(format=ToolFormat.LANGCHAIN)
    This uses the convert_tool_to_langchain_tool adapter internally, which wraps your MCP tool’s run method within a Langchain StructuredTool.
  5. Create a Langchain Agent: Use a Langchain agent constructor, like create_react_agent from langgraph.prebuilt, passing the llm and the langchain_tools. 
    from langchain_openai import ChatOpenAI # Or your preferred LLM
from langgraph.prebuilt import create_react_agent

llm = ChatOpenAI(model="gpt-4o-mini") # Example
agent_executor = create_react_agent(
    llm,
    tools=langchain_tools,
    prompt="You are a helpful assistant that can use tools." # Customize as needed
)
    The ReAct (Reasoning and Acting) agent is a common choice that works well with tools.
  6. Invoke the Agent: Call the agent with user input. 
    async def run_agent(prompt_text):
    result = await agent_executor.ainvoke(
        input={"messages": [{"role": "user", "content": prompt_text}]}
    )
    print(result["messages"][-1].content)
This pattern of initializing ToolManager, registering tools (either from applications or custom functions), converting them, and then passing them to a Langchain agent is fundamental. The subsequent example guides will build upon this by using tools from specific MCP applications.