mcp 工具sse调用失败，报错 ToolMessage(content="Error: NotImplementedError('StructuredTool does not support sync invocation.') #173
Description
python3.12
langchain-mcp-adapters 0.1.1
完整代码如下：
import asyncio
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from langgraph.graph import MessagesState, StateGraph
from langgraph.constants import START, END
from langchain_openai import ChatOpenAI
先创建llm
import os
from langchain.chat_models import init_chat_model
API_KEY = "sk-123"
BASE_URL = "https://api.deepseek.com"
os.environ["OPENAI_API_KEY"] = API_KEY
os.environ["OPENAI_API_BASE"] = BASE_URL
llm = init_chat_model("openai:deepseek-chat")
model = llm
class ComplicatedState(MessagesState):
new_field: str
async def make_graph():
client = MultiServerMCPClient(
{
"bing-cn-mcp-server": {
"url": "https://mcp.api-inference.modelscope.cn/sse/3a7bdd8299e141",
"transport": "sse",
}
}
)
# since get_tools is not an async, it could be called directly
tools = await client.get_tools()
print(tools)
sub_graph_workflow = StateGraph(ComplicatedState)
sub_graph_workflow.add_node("executor", create_react_agent(model, tools))
sub_graph_workflow.add_edge(START, "executor")
sub_graph_workflow.add_edge("executor", END)
sub_graph = sub_graph_workflow.compile()
def call_sub_graph(state: MessagesState):
# this is just a demonstration, the graph could be larger
# there will be more invoke and more def instead of async def
return sub_graph.invoke(ComplicatedState(messages=state["messages"], new_field="something"))
main_graph_workflow = StateGraph(MessagesState)
main_graph_workflow.add_node("sub_graph", call_sub_graph)
main_graph_workflow.add_edge(START, "sub_graph")
main_graph_workflow.add_edge("sub_graph", END)
main_graph = main_graph_workflow.compile()
return main_graph
async def main():
graph = await make_graph() # top level async here
# result = await graph.ainvoke({"messages": "What is the weather like in New York?"}) # it seems ainvoke is necessary for the later tool invocation
# print(result)
bing_response=graph.stream({"messages": "小米发布的芯片叫什么?"}, stream_mode="updates")
for chunk in bing_response:
print(chunk)
Change the transport to 'sse' and start the mcp server first:
https://github.com/modelcontextprotocol/quickstart-resources/tree/main/weather-server-python
if name == "main":
asyncio.run(main())
报错：
ToolMessage(content="Error: NotImplementedError('StructuredTool does not support sync invocation.')\n Please fix your
mistakes.", name='bing_search', id='95b1e19b-9af7-4ba5-815f-041359fa8a76',