\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df，\u6a21\u578b\u4e0a\u4e0b\u6587\u534f\u8bae（Model Context Protocol，\u7b80\u79f0 MCP）\u4f5c\u4e3a\u4e00\u79cd\u6807\u51c6\u5316\u7684\u534f\u8bae，\u4e3a\u5927\u578b\u8bed\u8a00\u6a21\u578b（LLM）\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u4e0a\u4e0b\u6587\u548c\u5de5\u5177\u652f\u6301\u3002\u800c FastMCP \u4f5c\u4e3a\u6784\u5efa MCP \u670d\u52a1\u5668\u548c\u5ba2\u6237\u7aef\u7684 Python \u6846\u67b6，\u4ee5\u5176\u7b80\u6d01\u7684 API \u8bbe\u8ba1\u3001\u9ad8\u6548\u7684\u5f00\u53d1\u4f53\u9a8c\u4ee5\u53ca\u5f3a\u5927\u7684\u6269\u5c55\u80fd\u529b，\u6b63\u9010\u6e10\u6210\u4e3a\u5f00\u53d1\u8005\u4eec\u7684\u9996\u9009\u5de5\u5177\u3002<\/p>\n
FastMCP \u662f\u4e00\u4e2a\u7528\u4e8e\u6784\u5efa MCP \u670d\u52a1\u5668\u548c\u5ba2\u6237\u7aef\u7684 Python \u6846\u67b6，\u5176\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u8005\u80fd\u591f\u4ee5\u66f4\u5c11\u7684\u4ee3\u7801\u3001\u66f4\u9ad8\u7684\u6548\u7387\u6784\u5efa\u51fa\u529f\u80fd\u5f3a\u5927\u7684 MCP \u5e94\u7528\u3002\u76f8\u6bd4\u5b98\u65b9\u7684 MCP Python SDK，FastMCP \u5728\u591a\u4e2a\u65b9\u9762\u8fdb\u884c\u4e86\u4f18\u5316\u548c\u6539\u8fdb，\u63d0\u4f9b\u4e86\u66f4\u7b80\u6d01\u7684 API \u8bbe\u8ba1\u3001\u66f4\u9ad8\u6548\u7684\u5f00\u53d1\u4f53\u9a8c\u4ee5\u53ca\u66f4\u5f3a\u5927\u7684\u6269\u5c55\u80fd\u529b\u3002<\/p>\n
FastMCP \u7684\u6838\u5fc3\u4f18\u52bf\u4e3b\u8981\u4f53\u73b0\u5728\u4ee5\u4e0b\u51e0\u4e2a\u65b9\u9762：<\/p>\n
FastMCP \u91c7\u7528\u4e86 Pythonic \u7684\u8bbe\u8ba1\u98ce\u683c，\u901a\u8fc7\u88c5\u9970\u5668（Decorator）\u5373\u53ef\u8f7b\u677e\u5b9a\u4e49\u5de5\u5177（Tools）\u3001\u8d44\u6e90（Resources）\u548c\u63d0\u793a（Prompts）\u3002\u8fd9\u79cd\u8bbe\u8ba1\u65b9\u5f0f\u6781\u5927\u5730\u51cf\u5c11\u4e86\u5f00\u53d1\u8fc7\u7a0b\u4e2d\u7684\u6837\u677f\u4ee3\u7801，\u8ba9\u5f00\u53d1\u8005\u80fd\u591f\u66f4\u52a0\u4e13\u6ce8\u4e8e\u4e1a\u52a1\u903b\u8f91\u7684\u5b9e\u73b0\u3002<\/p>\n
FastMCP \u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u5185\u7f6e\u529f\u80fd，\u5982\u670d\u52a1\u5668\u7ec4\u5408\u3001\u8fdc\u7a0b\u670d\u52a1\u5668\u4ee3\u7406\u3001OpenAPI\/FastAPI \u96c6\u6210\u7b49，\u8fd9\u4e9b\u529f\u80fd\u4e0d\u4ec5\u63d0\u9ad8\u4e86\u5f00\u53d1\u6548\u7387，\u8fd8\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u66f4\u591a\u7684\u7075\u6d3b\u6027\u3002\u6b64\u5916，FastMCP \u8fd8\u652f\u6301\u591a\u79cd\u5ba2\u6237\u7aef\/\u670d\u52a1\u7aef\u4f20\u8f93\u6a21\u5f0f，\u5305\u62ec Stdio\u3001SSE \u548c\u5185\u5b58\u4f20\u8f93，\u80fd\u591f\u6ee1\u8db3\u4e0d\u540c\u573a\u666f\u4e0b\u7684\u5f00\u53d1\u9700\u6c42\u3002<\/p>\n
FastMCP \u7684\u8bbe\u8ba1\u5177\u6709\u9ad8\u5ea6\u7684\u53ef\u6269\u5c55\u6027，\u5f00\u53d1\u8005\u53ef\u4ee5\u6839\u636e\u81ea\u5df1\u7684\u9700\u6c42\u8f7b\u677e\u6dfb\u52a0\u65b0\u7684\u529f\u80fd\u548c\u6a21\u5757\u3002\u65e0\u8bba\u662f\u6784\u5efa\u590d\u6742\u7684 AI \u5e94\u7528，\u8fd8\u662f\u8fdb\u884c\u7b80\u5355\u7684\u6570\u636e\u5904\u7406，FastMCP \u90fd\u80fd\u591f\u63d0\u4f9b\u5f3a\u5927\u7684\u652f\u6301\u3002<\/p>\n
\u5feb\u901f\u5f00\u53d1：\u63d0\u4f9b\u9ad8\u5c42\u7ea7\u63a5\u53e3，\u5f00\u53d1\u8005\u4e0d\u9700\u8981\u5904\u7406\u590d\u6742\u7684\u534f\u8bae\u7ec6\u8282\u548c\u670d\u52a1\u5668\u7ba1\u7406，\u51cf\u5c11\u4e86\u7f16\u5199\u4ee3\u7801\u7684\u5de5\u4f5c\u91cf\u3002<\/p>\n<\/li>\n
\u7b80\u5355\u6613\u7528：\u4ec5\u9700\u5c11\u91cf\u7684\u6837\u677f\u4ee3\u7801\u5373\u53ef\u6784\u5efa MCP \u670d\u52a1\u5668，\u5f00\u53d1\u8005\u53ef\u4ee5\u901a\u8fc7\u88c5\u9970\u5668\u7684\u65b9\u5f0f，\u8f7b\u677e\u5b9a\u4e49\u5de5\u5177\u3001\u8d44\u6e90\u548c\u63d0\u793a，\u5927\u5927\u51cf\u5c11\u4e86\u5197\u4f59\u4ee3\u7801\u3002<\/p>\n<\/li>\n
Pythonic \u98ce\u683c：\u5145\u5206\u5229\u7528\u4e86 Python \u7684\u8bed\u8a00\u7279\u6027，\u8ba9\u4ee3\u7801\u7f16\u5199\u66f4\u52a0\u7b26\u5408 Python \u7684\u7f16\u7a0b\u4e60\u60ef\u3002<\/p>\n<\/li>\n
\u529f\u80fd\u5b8c\u6574：\u63d0\u4f9b\u5bf9 MCP \u6838\u5fc3\u89c4\u8303\u7684\u5b8c\u6574\u5b9e\u73b0，\u76ee\u524d\u6838\u5fc3\u529f\u80fd\u5df2\u7ecf\u80fd\u591f\u6ee1\u8db3\u5927\u591a\u6570\u5f00\u53d1\u9700\u6c42，\u5e76\u4ecd\u5904\u4e8e\u6d3b\u8dc3\u5f00\u53d1\u4e2d\u3002<\/p>\n<\/li>\n<\/ul>\n
\u5de5\u5177（Tools）：\u5b9a\u4e49 LLM \u53ef\u6267\u884c\u7684\u64cd\u4f5c（\u5982\u8ba1\u7b97\u3001API \u8c03\u7528），\u652f\u6301\u540c\u6b65\u548c\u5f02\u6b65\u51fd\u6570\u3002<\/p>\n<\/li>\n
\u8d44\u6e90（Resources）：\u63d0\u4f9b\u7ed3\u6784\u5316\u6570\u636e\u8bbf\u95ee（\u5982\u914d\u7f6e\u4fe1\u606f\u3001\u7528\u6237\u6570\u636e），\u652f\u6301\u52a8\u6001\u8def\u5f84\u53c2\u6570\u3002<\/p>\n<\/li>\n
\u4e0a\u4e0b\u6587（Context）：\u5185\u7f6e\u65e5\u5fd7\u8bb0\u5f55\u3001\u8fdb\u5ea6\u62a5\u544a\u3001\u8d44\u6e90\u8bfb\u53d6\u7b49\u80fd\u529b，\u589e\u5f3a\u4ea4\u4e92\u63a7\u5236\u3002<\/p>\n<\/li>\n
\u63d0\u793a（Prompts）：\u53ef\u590d\u7528\u4ea4\u4e92\u6a21\u677f，\u6307\u5bfc LLM \u66f4\u9ad8\u6548\u5730\u4f7f\u7528\u5de5\u5177\u3002<\/p>\n<\/li>\n<\/ul>\n
\u5feb\u901f\u5f00\u53d1：\u901a\u8fc7\u9ad8\u7ea7\u63a5\u53e3，\u51cf\u5c11\u4ee3\u7801\u91cf，\u52a0\u901f\u5f00\u53d1\u3002<\/p>\n<\/li>\n
\u7b80\u6d01\u6613\u7528：\u4f7f\u7528\u6700\u5c11\u7684\u6837\u677f\u4ee3\u7801\u6784\u5efa MCP \u670d\u52a1\u5668\u3002<\/p>\n<\/li>\n
Pythonic：\u7b26\u5408 Python \u5f00\u53d1\u8005\u7684\u4e60\u60ef\u3002<\/p>\n<\/li>\n
\u5b8c\u6574\u5b9e\u73b0：\u5168\u9762\u652f\u6301 MCP \u6838\u5fc3\u89c4\u8303\u3002<\/p>\n<\/li>\n<\/ul>\n
ai\u00a0\u52a9\u624b\u5de5\u5177\u96c6：\u4e3a AI \u52a9\u624b\u63d0\u4f9b\u672c\u5730\u529f\u80fd\u6269\u5c55\u3002<\/p>\n<\/li>\n
\u6570\u636e\u5e93\u4ea4\u4e92：\u5b89\u5168\u5730\u66b4\u9732\u6570\u636e\u5e93\u67e5\u8be2\u529f\u80fd\u3002<\/p>\n<\/li>\n
\u6587\u4ef6\u5904\u7406：\u8bfb\u53d6\u548c\u5904\u7406\u672c\u5730\u6587\u4ef6\u3002<\/p>\n<\/li>\n
API \u96c6\u6210：\u5c06\u73b0\u6709 API \u5305\u88c5\u4e3a AI \u53ef\u7528\u7684\u5de5\u5177\u3002<\/p>\n<\/li>\n
\u667a\u80fd\u5ba2\u670d：\u53ef\u4ee5\u4f7f\u7528 FastMCP \u6784\u5efa\u670d\u52a1\u5668，\u4e3a\u5ba2\u670d\u4eba\u5458\u63d0\u4f9b\u5feb\u901f\u67e5\u8be2\u77e5\u8bc6\u5e93\u3001\u8c03\u7528\u6570\u636e\u5206\u6790\u5de5\u5177\u7b49\u529f\u80fd<\/p>\n<\/li>\n
\u6570\u636e\u5206\u6790：\u901a\u8fc7 FastMCP \u5c06\u5404\u79cd\u6570\u636e\u5206\u6790\u5de5\u5177\u96c6\u6210\u5230 LLM \u4e2d，\u8ba9\u7528\u6237\u80fd\u591f\u66f4\u65b9\u4fbf\u5730\u8fdb\u884c\u6570\u636e\u5904\u7406\u548c\u5206\u6790<\/p>\n<\/li>\n<\/ul>\n
FastMCP 1.0 \u7684\u6838\u5fc3\u6982\u5ff5\u5df2\u7ecf\u88ab\u7eb3\u5165\u5b98\u65b9\u7684 MCP Python SDK，\u800c\u5f53\u524d\u7684 FastMCP 2.0 \u5219\u662f\u5728\u6b64\u57fa\u7840\u4e0a\u8fdb\u884c\u7684\u8fdb\u4e00\u6b65\u6269\u5c55\u548c\u4f18\u5316\u3002FastMCP 2.0 \u4e0d\u4ec5\u7ee7\u627f\u4e86 1.0 \u7248\u672c\u7684\u57fa\u7840\u529f\u80fd，\u8fd8\u5f15\u5165\u4e86\u5b8c\u6574\u7684\u5ba2\u6237\u7aef\u652f\u6301\u3001\u670d\u52a1\u5668\u7ec4\u5408\u3001OpenAPI\/FastAPI \u96c6\u6210\u3001\u8fdc\u7a0b\u670d\u52a1\u5668\u4ee3\u7406\u4ee5\u53ca\u5185\u7f6e\u6d4b\u8bd5\u5de5\u5177\u7b49\u65b0\u529f\u80fd，\u663e\u7740\u63d0\u5347\u4e86\u5f00\u53d1\u6548\u7387\u548c\u5e94\u7528\u7684\u7075\u6d3b\u6027\u3002<\/p>\n
\u4e3a\u4e86\u66f4\u597d\u5730\u7406\u89e3 FastMCP \u7684\u5f3a\u5927\u529f\u80fd\u548c\u7b80\u6d01\u7684\u5f00\u53d1\u65b9\u5f0f，\u6211\u4eec\u4ee5\u4e00\u4e2a\u57fa\u4e8e FastMCP \u7684\u6570\u5b66\u8fd0\u7b97\u667a\u80fd\u95ee\u7b54\u5e94\u7528\u4e3a\u4f8b，\u8be6\u7ec6\u4ecb\u7ecd\u5176\u5f00\u53d1\u8fc7\u7a0b\u3002<\/p>\n
\u9996\u5148，\u9700\u8981\u5b89\u88c5 FastMCP \u6846\u67b6\u3002\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u5373\u53ef\u5b8c\u6210\u5b89\u88c5：<\/p>\n
uv pip install fastmcp<\/p>\n
\u670d\u52a1\u7aef\u7684\u5b9e\u73b0\u975e\u5e38\u7b80\u5355，\u53ea\u9700\u8981\u5b9a\u4e49\u51e0\u4e2a\u57fa\u672c\u7684\u6570\u5b66\u8fd0\u7b97\u5de5\u5177\u5373\u53ef\u3002\u4ee5\u4e0b\u662f\u670d\u52a1\u7aef\u7684\u4ee3\u7801\u793a\u4f8b：<\/p>\n
from fastmcp import FastMCP<\/p>\n
mcp = FastMCP(name="MyAssistantServer")<\/p>\n
@mcp.tool()
\ndef add(a: float, b: float) -> float:
\n """\u52a0\u6cd5\u8fd0\u7b97<\/p>\n
\u53c2\u6570:
\n a: \u7b2c\u4e00\u4e2a\u6570\u5b57
\n b: \u7b2c\u4e8c\u4e2a\u6570\u5b57<\/p>\n
\u8fd4\u56de:
\n \u4e24\u6570\u4e4b\u548c
\n """
\n return a + b<\/p>\n
@mcp.tool()
\ndef subtract(a: float, b: float) -> float:
\n """\u51cf\u6cd5\u8fd0\u7b97<\/p>\n
\u53c2\u6570:
\n a: \u7b2c\u4e00\u4e2a\u6570\u5b57
\n b: \u7b2c\u4e8c\u4e2a\u6570\u5b57<\/p>\n
\u8fd4\u56de:
\n \u4e24\u6570\u4e4b\u5dee (a – b)
\n """
\n return a – b<\/p>\n
@mcp.tool()
\ndef multiply(a: float, b: float) -> float:
\n """\u4e58\u6cd5\u8fd0\u7b97<\/p>\n
\u53c2\u6570:
\n a: \u7b2c\u4e00\u4e2a\u6570\u5b57
\n b: \u7b2c\u4e8c\u4e2a\u6570\u5b57<\/p>\n
\u8fd4\u56de:
\n \u4e24\u6570\u4e4b\u79ef
\n """
\n return a * b<\/p>\n
@mcp.tool()
\ndef divide(a: float, b: float) -> float:
\n """\u9664\u6cd5\u8fd0\u7b97<\/p>\n
\u53c2\u6570:
\n a: \u88ab\u9664\u6570
\n b: \u9664\u6570<\/p>\n
\u8fd4\u56de:
\n \u4e24\u6570\u4e4b\u5546 (a \/ b)<\/p>\n
\u5f02\u5e38:
\n ValueError: \u5f53\u9664\u6570\u4e3a\u96f6\u65f6
\n """
\n if b == 0:
\n raise ValueError("\u9664\u6570\u4e0d\u80fd\u4e3a\u96f6")
\n return a \/ b<\/p>\n
if __name__ == "__main__":
\n mcp.run(transport='sse', host="127.0.0.1", port=8001)<\/p>\n
\u5728\u4e0a\u8ff0\u4ee3\u7801\u4e2d，\u6211\u4eec\u5b9a\u4e49\u4e86\u56db\u4e2a\u57fa\u672c\u7684\u6570\u5b66\u8fd0\u7b97\u5de5\u5177：\u52a0\u6cd5\u3001\u51cf\u6cd5\u3001\u4e58\u6cd5\u548c\u9664\u6cd5\u3002\u901a\u8fc7\u88c5\u9970\u5668\u00a0@mcp.tool()，\u8fd9\u4e9b\u51fd\u6570\u88ab\u6ce8\u518c\u4e3a MCP \u670d\u52a1\u5668\u7684\u5de5\u5177，\u53ef\u4f9b\u5ba2\u6237\u7aef\u8c03\u7528\u3002\u6700\u540e，\u901a\u8fc7\u8c03\u7528\u00a0mcp.run()\u00a0\u65b9\u6cd5\u542f\u52a8\u670d\u52a1\u5668，\u5e76\u6307\u5b9a\u4f7f\u7528 SSE \u4f20\u8f93\u6a21\u5f0f\u3002<\/p>\n
\u5ba2\u6237\u7aef\u7684\u5b9e\u73b0\u540c\u6837\u7b80\u5355\u3002\u901a\u8fc7\u4e00\u884c\u4ee3\u7801\u5373\u53ef\u521b\u5efa\u4e00\u4e2a MCP \u5ba2\u6237\u7aef，\u5e76\u8fde\u63a5\u5230\u670d\u52a1\u7aef\u3002\u4ee5\u4e0b\u662f\u5ba2\u6237\u7aef\u7684\u4ee3\u7801\u793a\u4f8b：<\/p>\n
from fastmcp import Client
\nimport asyncio<\/p>\n
async def main():
\n # \u6d4b\u8bd5 mcp \u5ba2\u6237\u7aef\u7684\u529f\u80fd
\n async with Client("http:\/\/127.0.0.1:8001\/sse") as mcp_client:
\n tools = await mcp_client.list_tools()
\n print(f"Available tools: {tools}")
\n result = await mcp_client.call_tool("add", {"a": 5, "b": 3})
\n print(f"Result: {result[0].text}")<\/p>\n
if __name__ == "__main__":
\n asyncio.run(main())<\/p>\n
\u5728\u4e0a\u8ff0\u4ee3\u7801\u4e2d，\u6211\u4eec\u901a\u8fc7\u00a0Client\u00a0\u7c7b\u521b\u5efa\u4e86\u4e00\u4e2a MCP \u5ba2\u6237\u7aef，\u5e76\u8fde\u63a5\u5230\u670d\u52a1\u7aef\u3002\u901a\u8fc7\u8c03\u7528\u00a0list_tools()\u00a0\u65b9\u6cd5，\u6211\u4eec\u53ef\u4ee5\u83b7\u53d6\u670d\u52a1\u7aef\u63d0\u4f9b\u7684\u6240\u6709\u5de5\u5177\u5217\u8868\u3002\u7136\u540e，\u901a\u8fc7\u8c03\u7528\u00a0call_tool()\u00a0\u65b9\u6cd5，\u6211\u4eec\u53ef\u4ee5\u8c03\u7528\u670d\u52a1\u7aef\u7684\u5de5\u5177\u5e76\u83b7\u53d6\u7ed3\u679c\u3002<\/p>\n
\u57fa\u4e8e FastMCP，\u6211\u4eec\u8fd8\u53ef\u4ee5\u6784\u5efa\u4e00\u4e2a\u6570\u5b66\u8fd0\u7b97\u667a\u80fd\u95ee\u7b54\u5e94\u7528\u3002\u8be5\u5e94\u7528\u901a\u8fc7\u4e0e\u5927\u8bed\u8a00\u6a21\u578b（LLM）\u8fdb\u884c\u4ea4\u4e92，\u6839\u636e\u7528\u6237\u7684\u8f93\u5165\u8c03\u7528\u76f8\u5e94\u7684\u5de5\u5177\u8fdb\u884c\u8ba1\u7b97，\u5e76\u8fd4\u56de\u7ed3\u679c\u3002\u4ee5\u4e0b\u662f\u8be5\u5e94\u7528\u7684\u4ee3\u7801\u793a\u4f8b：<\/p>\n
import asyncio
\nimport json
\nimport logging
\nimport os
\nfrom typing import List, Dict<\/p>\n
from fastmcp import Client
\nfrom openai import OpenAI<\/p>\n
class LLMClient:
\n """LLM\u5ba2\u6237\u7aef，\u8d1f\u8d23\u4e0e\u5927\u8bed\u8a00\u6a21\u578bAPI\u901a\u4fe1"""<\/p>\n
def __init__(self, model_name: str, url: str, api_key: str) -> None:
\n self.model_name: str = model_name
\n self.url: str = url
\n self.client = OpenAI(api_key=api_key, base_url=url)<\/p>\n
def get_response(self, messages: List[Dict[str, str]]) -> str:
\n """\u53d1\u9001\u6d88\u606f\u7ed9LLM\u5e76\u83b7\u53d6\u54cd\u5e94"""
\n response = self.client.chat.completions.create(
\n model=self.model_name,
\n messages=messages,
\n stream=False
\n )
\n return response.choices[0].message.content<\/p>\n
class ChatSession:
\n """\u804a\u5929\u4f1a\u8bdd，\u5904\u7406\u7528\u6237\u8f93\u5165\u548cLLM\u54cd\u5e94，\u5e76\u4e0eMCP\u5de5\u5177\u4ea4\u4e92"""<\/p>\n
def __init__(self, llm_client: LLMClient, mcp_client: Client) -> None:
\n self.mcp_client: Client = mcp_client
\n self.llm_client: LLMClient = llm_client<\/p>\n
async def process_llm_response(self, llm_response: str) -> str:
\n """\u5904\u7406LLM\u54cd\u5e94，\u89e3\u6790\u5de5\u5177\u8c03\u7528\u5e76\u6267\u884c"""
\n try:
\n # \u5c1d\u8bd5\u79fb\u9664\u53ef\u80fd\u7684markdown\u683c\u5f0f
\n if llm_response.startswith('```json'):
\n llm_response = llm_response.strip('```json').strip('```').strip()
\n tool_call = json.loads(llm_response)
\n if "tool" in tool_call and "arguments" in tool_call:
\n # \u68c0\u67e5\u5de5\u5177\u662f\u5426\u53ef\u7528
\n tools = await self.mcp_client.list_tools()
\n if any(tool.name == tool_call["tool"] for tool in tools):
\n try:
\n # \u6267\u884c\u5de5\u5177\u8c03\u7528
\n result = await self.mcp_client.call_tool(
\n tool_call["tool"], tool_call["arguments"]
\n )<\/p>\n
return f"Tool execution result: {result}"
\n except Exception as e:
\n error_msg = f"Error executing tool: {str(e)}"
\n logging.error(error_msg)
\n return error_msg
\n return f"No server found with tool: {tool_call['tool']}"
\n return llm_response
\n except json.JSONDecodeError:
\n # \u5982\u679c\u4e0d\u662fJSON\u683c\u5f0f，\u76f4\u63a5\u8fd4\u56de\u539f\u59cb\u54cd\u5e94
\n return llm_response<\/p>\n
async def start(self, system_message: str) -> None:
\n """\u542f\u52a8\u804a\u5929\u4f1a\u8bdd\u7684\u4e3b\u5faa\u73af"""
\n messages = [{"role": "system", "content": system_message}]
\n while True:
\n try:
\n # \u83b7\u53d6\u7528\u6237\u8f93\u5165
\n user_input = input("\u7528\u6237: ").strip().lower()
\n if user_input in ["quit", "exit", "\u9000\u51fa"]:
\n print('AI\u52a9\u624b\u9000\u51fa')
\n break
\n messages.append({"role": "user", "content": user_input})<\/p>\n
# \u83b7\u53d6LLM\u7684\u521d\u59cb\u54cd\u5e94
\n llm_response = self.llm_client.get_response(messages)
\n print("\u52a9\u624b: ", llm_response)<\/p>\n
# \u5904\u7406\u53ef\u80fd\u7684\u5de5\u5177\u8c03\u7528
\n result = await self.process_llm_response(llm_response)<\/p>\n
# \u5982\u679c\u5904\u7406\u7ed3\u679c\u4e0e\u539f\u59cb\u54cd\u5e94\u4e0d\u540c，\u8bf4\u660e\u6267\u884c\u4e86\u5de5\u5177\u8c03\u7528，\u9700\u8981\u8fdb\u4e00\u6b65\u5904\u7406
\n while result != llm_response:
\n messages.append({"role": "assistant", "content": llm_response})
\n messages.append({"role": "system", "content": result})<\/p>\n
# \u5c06\u5de5\u5177\u6267\u884c\u7ed3\u679c\u53d1\u9001\u56deLLM\u83b7\u53d6\u65b0\u54cd\u5e94
\n llm_response = self.llm_client.get_response(messages)
\n result = await self.process_llm_response(llm_response)
\n print("\u52a9\u624b: ", llm_response)<\/p>\n
messages.append({"role": "assistant", "content": llm_response})<\/p>\n
except KeyboardInterrupt:
\n print('AI\u52a9\u624b\u9000\u51fa')
\n break<\/p>\n
async def main():
\n async with Client("http:\/\/127.0.0.1:8001\/sse") as mcp_client:
\n # \u521d\u59cb\u5316LLM\u5ba2\u6237\u7aef，\u4f7f\u7528\u901a\u4e49\u5343\u95ee\u6a21\u578b
\n llm_client = LLMClient(model_name='qwen-plus-latest', api_key=os.getenv('DASHSCOPE_API_KEY'),
\n url='https:\/\/dashscope.aliyuncs.com\/compatible-mode\/v1')<\/p>\n
# \u83b7\u53d6\u53ef\u7528\u5de5\u5177\u5217\u8868\u5e76\u683c\u5f0f\u5316\u4e3a\u7cfb\u7edf\u63d0\u793a\u7684\u4e00\u90e8\u5206
\n tools = await mcp_client.list_tools()
\n dict_list = [tool.__dict__ for tool in tools]
\n tools_description = json.dumps(dict_list, ensure_ascii=False)<\/p>\n
# \u7cfb\u7edf\u63d0\u793a，\u6307\u5bfcLLM\u5982\u4f55\u4f7f\u7528\u5de5\u5177\u548c\u8fd4\u56de\u54cd\u5e94
\n system_message = f'''
\n \u4f60\u662f\u4e00\u4e2a\u667a\u80fd\u52a9\u624b，\u4e25\u683c\u9075\u5faa\u4ee5\u4e0b\u534f\u8bae\u8fd4\u56de\u54cd\u5e94：<\/p>\n
\u53ef\u7528\u5de5\u5177：{tools_description}<\/p>\n
\u54cd\u5e94\u89c4\u5219：
\n 1\u3001\u5f53\u9700\u8981\u8ba1\u7b97\u65f6，\u8fd4\u56de\u4e25\u683c\u7b26\u5408\u4ee5\u4e0b\u683c\u5f0f\u7684\u7eaf\u51c0JSON：
\n {{
\n "tool": "tool-name",
\n "arguments": {{
\n "argument-name": "value"
\n }}
\n }}
\n 2\u3001\u7981\u6b62\u5305\u542b\u4ee5\u4e0b\u5185\u5bb9：
\n – Markdown\u6807\u8bb0（\u5982```json）
\n – \u81ea\u7136\u8bed\u8a00\u89e3\u91ca（\u5982"\u7ed3\u679c："）
\n – \u683c\u5f0f\u5316\u6570\u503c（\u5fc5\u987b\u4fdd\u6301\u539f\u59cb\u7cbe\u5ea6）
\n – \u5355\u4f4d\u7b26\u53f7（\u5982\u5143\u3001kg）<\/p>\n
\u6821\u9a8c\u6d41\u7a0b：
\n \u2713 \u53c2\u6570\u6570\u91cf\u4e0e\u5de5\u5177\u5b9a\u4e49\u4e00\u81f4
\n \u2713 \u6570\u503c\u7c7b\u578b\u4e3anumber
\n \u2713 JSON\u683c\u5f0f\u6709\u6548\u6027\u68c0\u67e5<\/p>\n
\u6b63\u786e\u793a\u4f8b：
\n \u7528\u6237：\u5355\u4ef788.5\u4e70235\u4e2a\u591a\u5c11\u94b1？
\n \u54cd\u5e94：{{"tool":"multiply","arguments":{{"a":88.5,"b":235}}}}<\/p>\n
\u9519\u8bef\u793a\u4f8b：
\n \u7528\u6237：\u603b\u91d1\u989d\u662f\u591a\u5c11？
\n \u9519\u8bef\u54cd\u5e94：\u603b\u4ef7500\u5143 \u2192 \u542b\u81ea\u7136\u8bed\u8a00
\n \u9519\u8bef\u54cd\u5e94：```json{{…}}```\u2192 \u542bMarkdown<\/p>\n
3\u3001\u5728\u6536\u5230\u5de5\u5177\u7684\u54cd\u5e94\u540e：
\n – \u5c06\u539f\u59cb\u6570\u636e\u8f6c\u5316\u4e3a\u81ea\u7136\u3001\u5bf9\u8bdd\u5f0f\u7684\u56de\u5e94
\n – \u4fdd\u6301\u56de\u590d\u7b80\u6d01\u4f46\u4fe1\u606f\u4e30\u5bcc
\n – \u805a\u7126\u4e8e\u6700\u76f8\u5173\u7684\u4fe1\u606f
\n – \u4f7f\u7528\u7528\u6237\u95ee\u9898\u4e2d\u7684\u9002\u5f53\u4e0a\u4e0b\u6587
\n – \u907f\u514d\u7b80\u5355\u91cd\u590d\u4f7f\u7528\u539f\u59cb\u6570\u636e
\n '''
\n # \u542f\u52a8\u804a\u5929\u4f1a\u8bdd
\n chat_session = ChatSession(llm_client=llm_client, mcp_client=mcp_client)
\n await chat_session.start(system_message=system_message)<\/p>\n
if __name__ == "__
\n__main__":
\n asyncio.run(main())<\/p>\n
\u8fd0\u884c\u670d\u52a1\u7aef\u4ee3\u7801：<\/p>\n
python fast_mcp_server.py<\/p>\n
\u8fd0\u884c\u5ba2\u6237\u7aef\u4ee3\u7801：<\/p>\n
python fast_mcp_client.py<\/p>\n
\u5728\u5ba2\u6237\u7aef\u4e2d\u8f93\u5165\u6570\u5b66\u95ee\u9898，\u4f8b\u5982：<\/p>\n
\u7528\u6237: \u73b0\u5728\u8981\u8d2d\u4e70\u4e00\u6279\u8d27，\u5355\u4ef7\u662f 1034.32423，\u6570\u91cf\u662f 235326\u3002\u5546\u5bb6\u540e\u6765\u53c8\u8bf4，\u53ef\u4ee5\u5728\u8fd9\u4e2a\u57fa\u7840\u4e0a，\u625395\u6298，\u6298\u540e\u603b\u4ef7\u662f\u591a\u5c11？<\/p>\n
\u5ba2\u6237\u7aef\u4f1a\u8c03\u7528\u670d\u52a1\u7aef\u7684\u00a0multiply\u00a0\u5de5\u5177\u8fdb\u884c\u8ba1\u7b97，\u5e76\u8fd4\u56de\u7ed3\u679c：<\/p>\n
\u52a9\u624b: {
\n "tool": "multiply",
\n "arguments": {
\n "a": 1034.32423,
\n "b": 235326
\n }
\n}
\n\u52a9\u624b: {
\n "tool": "multiply",
\n "arguments": {
\n "a": 243403383.74898,
\n "b": 0.95
\n }
\n}
\n\u52a9\u624b: \u6298\u540e\u603b\u4ef7\u662f231233214.56\u3002<\/p>\n
\u9664\u4e86\u4e0a\u8ff0\u63d0\u5230\u7684\u529f\u80fd\u5916，FastMCP \u8fd8\u63d0\u4f9b\u4e86\u8bb8\u591a\u5176\u4ed6\u5f3a\u5927\u7684\u529f\u80fd，\u4f8b\u5982：<\/p>\n
\u901a\u8fc7\u00a0mcp.mount()\u00a0\u6216\u00a0mcp.import_server()\u00a0\u65b9\u6cd5，\u53ef\u4ee5\u5c06\u591a\u4e2a\u00a0FastMCP\u00a0\u5b9e\u4f8b\u7ec4\u5408\u5230\u4e00\u4e2a\u7236\u670d\u52a1\u5668\u4e2d，\u4ece\u800c\u6784\u5efa\u51fa\u6a21\u5757\u5316\u7684\u5e94\u7528\u7a0b\u5e8f\u3002\u8fd9\u79cd\u65b9\u5f0f\u4e0d\u4ec5\u63d0\u9ad8\u4e86\u4ee3\u7801\u7684\u53ef\u7ef4\u62a4\u6027，\u8fd8\u65b9\u4fbf\u4e86\u529f\u80fd\u7684\u6269\u5c55\u3002<\/p>\n
FastMCP \u63d0\u4f9b\u4e86\u4ece\u73b0\u6709\u7684 OpenAPI \u89c4\u8303\u6216 FastAPI \u5e94\u7528\u7a0b\u5e8f\u751f\u6210\u00a0FastMCP\u00a0\u670d\u52a1\u5668\u7684\u529f\u80fd\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f，\u5f00\u53d1\u8005\u53ef\u4ee5\u8f7b\u677e\u5730\u5c06\u73b0\u6709\u7684 Web API \u96c6\u6210\u5230 MCP \u751f\u6001\u7cfb\u7edf\u4e2d，\u8fdb\u4e00\u6b65\u6269\u5c55\u4e86 MCP \u7684\u5e94\u7528\u573a\u666f\u3002<\/p>\n
\u901a\u8fc7\u00a0FastMCP.as_proxy()\u00a0\u65b9\u6cd5，\u53ef\u4ee5\u521b\u5efa\u4e00\u4e2a\u4ee3\u7406\u670d\u52a1\u5668，\u8be5\u670d\u52a1\u5668\u53ef\u4ee5\u4f5c\u4e3a\u672c\u5730\u6216\u8fdc\u7a0b MCP \u670d\u52a1\u5668\u7684\u4e2d\u95f4\u5c42\u3002\u8fd9\u79cd\u65b9\u5f0f\u7279\u522b\u9002\u7528\u4e8e\u6865\u63a5\u4e0d\u540c\u7684\u4f20\u8f93\u534f\u8bae（\u4f8b\u5982，\u5c06\u8fdc\u7a0b SSE \u670d\u52a1\u5668\u4ee3\u7406\u5230\u672c\u5730 Stdio \u5ba2\u6237\u7aef）\u6216\u4e3a\u4e0d\u53d7\u63a7\u5236\u7684\u670d\u52a1\u5668\u6dfb\u52a0\u903b\u8f91\u5c42\u3002<\/p>\n
FastMCP \u63d0\u4f9b\u4e86\u5f3a\u5927\u7684\u5185\u7f6e\u6d4b\u8bd5\u5de5\u5177，\u652f\u6301\u901a\u8fc7\u5185\u5b58\u4f20\u8f93\u76f4\u63a5\u8fde\u63a5\u5230\u00a0FastMCP\u00a0\u670d\u52a1\u5668\u5b9e\u4f8b，\u4ece\u800c\u5728\u6d4b\u8bd5\u8fc7\u7a0b\u4e2d\u65e0\u9700\u8fdb\u884c\u8fdb\u7a0b\u7ba1\u7406\u548c\u7f51\u7edc\u8c03\u7528\u3002\u8fd9\u79cd\u65b9\u5f0f\u6781\u5927\u5730\u63d0\u9ad8\u4e86\u6d4b\u8bd5\u6548\u7387，\u964d\u4f4e\u4e86\u6d4b\u8bd5\u6210\u672c\u3002<\/p>\n
FastMCP \u4f5c\u4e3a\u4e00\u4e2a\u9ad8\u6548\u3001\u7b80\u6d01\u4e14\u529f\u80fd\u5f3a\u5927\u7684 Python \u6846\u67b6，\u4e3a\u6784\u5efa MCP \u670d\u52a1\u5668\u548c\u5ba2\u6237\u7aef\u63d0\u4f9b\u4e86\u6781\u5927\u7684\u4fbf\u5229\u3002\u901a\u8fc7\u5176\u7b80\u6d01\u7684 API \u8bbe\u8ba1\u3001\u9ad8\u6548\u7684\u5f00\u53d1\u4f53\u9a8c\u548c\u5f3a\u5927\u7684\u6269\u5c55\u80fd\u529b，\u5f00\u53d1\u8005\u53ef\u4ee5\u5feb\u901f\u6784\u5efa\u51fa\u529f\u80fd\u4e30\u5bcc\u7684 MCP \u5e94\u7528\u3002\u65e0\u8bba\u662f\u7b80\u5355\u7684\u5de5\u5177\u5f00\u53d1，\u8fd8\u662f\u590d\u6742\u7684 AI \u5e94\u7528\u6784\u5efa，FastMCP \u90fd\u80fd\u591f\u6ee1\u8db3\u5f00\u53d1\u8005\u7684\u9700\u6c42\u3002<\/p><\/p>\n","protected":false},"excerpt":{"rendered":"
\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb901\u6b21\uff0c\u70b9\u8d5e8\u6b21\uff0c\u6536\u85cf20\u6b21\u3002FastMCP \u662f\u4e00\u4e2a\u7528\u4e8e\u6784\u5efa MCP \u670d\u52a1\u5668\u548c\u5ba2\u6237\u7aef\u7684 Python \u6846\u67b6\uff0c\u5176\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u8005\u80fd\u591f\u4ee5\u66f4\u5c11\u7684\u4ee3\u7801\u3001\u66f4\u9ad8\u7684\u6548\u7387\u6784\u5efa\u51fa\u529f\u80fd\u5f3a\u5927\u7684 MCP \u5e94\u7528\u3002\u76f8\u6bd4\u5b98\u65b9\u7684 MCP Python SDK\uff0cFastMCP \u5728\u591a\u4e2a\u65b9\u9762\u8fdb\u884c\u4e86\u4f18\u5316\u548c\u6539\u8fdb\uff0c\u63d0\u4f9b\u4e86\u66f4\u7b80\u6d01\u7684 API \u8bbe\u8ba1\u3001\u66f4\u9ad8\u6548\u7684\u5f00\u53d1\u4f53\u9a8c\u4ee5\u53ca\u66f4\u5f3a\u5927\u7684\u6269\u5c55\u80fd\u529b_fastmcp openai<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[53,81,227,43],"topic":[],"class_list":["post-46470","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-python","tag-227","tag-43"],"yoast_head":"\n