LangChain

LangChain supports MCP servers through the langchain-community/tools/mcp package. Use it to blend Xweather data with other tools, retrievers, or chains.

from langchain_community.chat_models import ChatAnthropic
from langchain_community.tools.mcp import JsonMCPTool
from langchain.agents import AgentExecutor, create_openai_functions_agent
 
weather_tool = JsonMCPTool(
    name="xweather",
    server_url="https://mcp.api.xweather.com/mcp",
    authorization_header=f"Bearer {os.environ['XWEATHER_API_KEY']}",
    allowed_tools=[
        "xweather_get_current_weather",
        "xweather_get_weather_forecast",
        "xweather_get_weather_impacts",
    ],
)
 
llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")
 
agent = create_openai_functions_agent(
    llm=llm,
    tools=[weather_tool],
    prompt="You are a weather analyst who uses Xweather tools when asked."
)
 
executor = AgentExecutor(agent=agent, tools=[weather_tool], verbose=True)
 
response = executor.invoke({"input": "Draft an operations brief for ORD tonight."})
print(response["output"])

Tips

Combine the MCP tool with a vector retriever for playbooks or escalation policies.
Set allowed_tools to the exact functions needed for each chain to avoid unnecessary tool calls.
Use callbacks or event tracing to capture tool parameters for observability.

Vercel AI SDK Tools