Upload 4 files

app.py

@@ -8,11 +8,13 @@ from huggingface_hub import InferenceClient
 MCP_SERVICES = {
     "financial": {
         "name": "SEC Financial Reports",
-        "url": "https://jc321-easyreportdatemcp.hf.space/mcp"
+        "url": "https://jc321-easyreportdatemcp.hf.space/mcp",
+        "type": "fastmcp"  # 标准 FastMCP (HTTP JSON-RPC)
     },
     "market": {
         "name": "Market & Stock Data (Finnhub)",
-        "url": "https://jc321-marketandstockmcp.hf.space/mcp"
+        "url": "https://jc321-marketandstockmcp.hf.space",
+        "type": "gradio"  # Gradio API
     }
 }
 
@@ -110,14 +112,14 @@ MCP_TOOLS = [
     }
 ]
 
-# 工具路由：工具名 -> 服务 URL
+# 工具路由:工具名 -> 服务配置
 TOOL_ROUTING = {
-    "advanced_search_company": MCP_SERVICES["financial"]["url"],
-    "get_latest_financial_data": MCP_SERVICES["financial"]["url"],
-    "extract_financial_metrics": MCP_SERVICES["financial"]["url"],
-    "get_quote": MCP_SERVICES["market"]["url"],
-    "get_market_news": MCP_SERVICES["market"]["url"],
-    "get_company_news": MCP_SERVICES["market"]["url"],
+    "advanced_search_company": MCP_SERVICES["financial"],
+    "get_latest_financial_data": MCP_SERVICES["financial"],
+    "extract_financial_metrics": MCP_SERVICES["financial"],
+    "get_quote": MCP_SERVICES["market"],
+    "get_market_news": MCP_SERVICES["market"],
+    "get_company_news": MCP_SERVICES["market"],
 }
 
 # ========== 初始化 LLM 客户端 ==========
@@ -145,45 +147,179 @@ You have access to TWO data sources:
 
 Automatically use the right tools and provide clear, data-driven insights."""
 
-# ========== 核心函数：调用 MCP 工具 ==========
+# ========== 核心函数:调用 MCP 工具 ==========
 def call_mcp_tool(tool_name, arguments):
-    """调用 MCP 工具（统一使用标准 MCP JSON-RPC 协议）"""
-    mcp_url = TOOL_ROUTING.get(tool_name)
-    if not mcp_url:
+    """调用 MCP 工具(支持 FastMCP 和 Gradio 两种协议)"""
+    service_config = TOOL_ROUTING.get(tool_name)
+    if not service_config:
         return {"error": f"Unknown tool: {tool_name}"}
     
+    service_type = service_config["type"]
+    service_url = service_config["url"]
+    
+    print(f"\n[DEBUG] Calling tool: {tool_name}")
+    print(f"[DEBUG] Service type: {service_type}, URL: {service_url}")
+    
     try:
-        response = requests.post(
-            mcp_url,
-            json={
-                "jsonrpc": "2.0",
-                "method": "tools/call",
-                "params": {"name": tool_name, "arguments": arguments},
-                "id": 1
-            },
-            headers={"Content-Type": "application/json"},
-            timeout=60
-        )
-        
-        if response.status_code == 200:
-            data = response.json()
-            # 解包 JSON-RPC 响应，只返回 result 字段给模型
-            if isinstance(data, dict) and "result" in data:
-                return data["result"]
-            return data
+        if service_type == "fastmcp":
+            # FastMCP: 标准 MCP JSON-RPC 协议
+            return _call_fastmcp(service_url, tool_name, arguments)
+        elif service_type == "gradio":
+            # Gradio: Gradio API 协议
+            return _call_gradio_api(service_url, tool_name, arguments)
         else:
-            return {"error": f"HTTP {response.status_code}", "detail": response.text[:200]}
+            return {"error": f"Unknown service type: {service_type}"}
+    
     except Exception as e:
-        return {"error": str(e)}
+        error_msg = f"Exception: {str(e)}"
+        print(f"[DEBUG] {error_msg}")
+        return {"error": error_msg}
+
+
+def _call_fastmcp(service_url, tool_name, arguments):
+    """\u8c03\u7528 FastMCP \u670d\u52a1 (\u6807\u51c6 MCP JSON-RPC \u534f\u8bae)"""
+    response = requests.post(
+        service_url,
+        json={
+            "jsonrpc": "2.0",
+            "method": "tools/call",
+            "params": {"name": tool_name, "arguments": arguments},
+            "id": 1
+        },
+        headers={"Content-Type": "application/json"},
+        timeout=60
+    )
+    
+    if response.status_code == 200:
+        data = response.json()
+        print(f"[DEBUG] FastMCP raw response: {json.dumps(data, ensure_ascii=False)[:500]}")
+        
+        # 解包 JSON-RPC 响应
+        if isinstance(data, dict) and "result" in data:
+            result = data["result"]
+            
+            # MCP 协议格式: {"content": [{"type": "text", "text": "..."}]}
+            if isinstance(result, dict) and "content" in result:
+                content = result["content"]
+                
+                # 提取第一个 content item 的 text
+                if isinstance(content, list) and len(content) > 0:
+                    first_item = content[0]
+                    if isinstance(first_item, dict) and "text" in first_item:
+                        text_data = first_item["text"]
+                        
+                        # text 可能是 JSON 字符串,尝试解析
+                        try:
+                            parsed_data = json.loads(text_data)
+                            print(f"[DEBUG] Parsed data: {json.dumps(parsed_data, ensure_ascii=False)[:300]}")
+                            return parsed_data
+                        except (json.JSONDecodeError, TypeError):
+                            # 如果不是 JSON,直接返回文本
+                            print(f"[DEBUG] Returning text as-is")
+                            return {"text": text_data}
+            
+            # 如果不是 content 格式,直接返回 result
+            print(f"[DEBUG] Returning result directly")
+            return result
+        
+        # 如果没有 result 字段,返回整个响应
+        print(f"[DEBUG] No result field, returning full response")
+        return data
+    else:
+        error_msg = f"HTTP {response.status_code}: {response.text[:200]}"
+        print(f"[DEBUG] {error_msg}")
+        return {"error": error_msg}
+
+
+def _call_gradio_api(service_url, tool_name, arguments):
+    """\u8c03\u7528 Gradio API (Gradio \u5185\u7f6e MCP \u670d\u52a1)"""
+    # Gradio API 工具名映射
+    gradio_tool_map = {
+        "get_quote": "test_quote_tool",
+        "get_market_news": "test_market_news_tool",
+        "get_company_news": "test_company_news_tool"
+    }
+    
+    gradio_fn_name = gradio_tool_map.get(tool_name)
+    if not gradio_fn_name:
+        return {"error": f"No Gradio mapping for tool: {tool_name}"}
+    
+    # 步骤1: 提交调用请求
+    call_url = f"{service_url}/call/{gradio_fn_name}"
+    
+    # 构造 Gradio API 参数格式
+    if tool_name == "get_quote":
+        data_params = [arguments.get("symbol", "")]
+    elif tool_name == "get_market_news":
+        data_params = [arguments.get("category", "general")]
+    elif tool_name == "get_company_news":
+        data_params = [
+            arguments.get("symbol", ""),
+            arguments.get("from_date", ""),
+            arguments.get("to_date", "")
+        ]
+    else:
+        data_params = []
+    
+    print(f"[DEBUG] Gradio call URL: {call_url}")
+    print(f"[DEBUG] Gradio data params: {data_params}")
+    
+    # 提交请求
+    response = requests.post(
+        call_url,
+        json={"data": data_params},
+        headers={"Content-Type": "application/json"},
+        timeout=10
+    )
+    
+    if response.status_code != 200:
+        return {"error": f"Gradio call failed: HTTP {response.status_code}"}
+    
+    call_data = response.json()
+    event_id = call_data.get("event_id")
+    
+    if not event_id:
+        return {"error": "No event_id returned from Gradio"}
+    
+    print(f"[DEBUG] Got event_id: {event_id}")
+    
+    # 步骤2: 轮询获取结果 (SSE 流)
+    result_url = f"{call_url}/{event_id}"
+    result_response = requests.get(result_url, stream=True, timeout=30)
+    
+    if result_response.status_code != 200:
+        return {"error": f"Failed to get result: HTTP {result_response.status_code}"}
+    
+    # 解析 SSE 流
+    result_text = ""
+    for line in result_response.iter_lines():
+        if line:
+            line_str = line.decode('utf-8')
+            if line_str.startswith('data: '):
+                data_part = line_str[6:]  # 移除 'data: ' 前缀
+                try:
+                    result_data = json.loads(data_part)
+                    if isinstance(result_data, list) and len(result_data) > 0:
+                        result_text = result_data[0]
+                        print(f"[DEBUG] Gradio result: {result_text[:200]}")
+                        break
+                except json.JSONDecodeError:
+                    continue
+    
+    if not result_text:
+        return {"error": "No result received from Gradio"}
+    
+    # 返回统一格式
+    return {"text": result_text, "_source": "gradio_api"}
 
-# ========== 核心函数：AI 助手 ==========
+# ========== 核心函数:AI 助手 ==========
 def chatbot_response(message, history):
-    """AI 助手主函数"""
+    """AI 助手主函数(流式输出)"""
     try:
         # 构建消息历史
         messages = [{"role": "system", "content": SYSTEM_PROMPT}]
         
-        # 添加对话历史（最近5轮）
+        # 添加对话历史(最近5轮)
         if history:
             for item in history[-5:]:
                 if isinstance(item, dict):
@@ -195,19 +331,22 @@ def chatbot_response(message, history):
         
         messages.append({"role": "user", "content": message})
         
-        # LLM 调用循环（支持多轮工具调用）
+        # 工具调用日志和响应前缀
         tool_calls_log = []
+        response_prefix = ""
         max_iterations = 5
         
+        # LLM 调用循环(支持多轮工具调用)
         for iteration in range(max_iterations):
-            # 调用 LLM
+            # 调用 LLM(非流式,用于工具调用判断)
             response = client.chat_completion(
                 messages=messages,
                 model="Qwen/Qwen2.5-72B-Instruct:novita",
                 tools=MCP_TOOLS,
                 max_tokens=3000,
                 temperature=0.7,
-                tool_choice="auto"
+                tool_choice="auto",
+                stream=False
             )
             
             choice = response.choices[0]
@@ -236,29 +375,59 @@ def chatbot_response(message, history):
                 
                 continue  # 继续下一轮
             else:
-                # 无工具调用，返回最终答案
-                response_text = choice.message.content
+                # 无工具调用,准备流式输出最终答案
                 break
         
-        # 构建最终响应
-        final_response = ""
+        # 构建响应前缀(模型信息+工具调用)
+        response_prefix += "<div style='padding: 10px; background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); border-radius: 8px; margin-bottom: 12px; box-shadow: 0 2px 8px rgba(0,0,0,0.1);'>\n"
+        response_prefix += "<div style='color: white; font-size: 0.95em; font-weight: 500;'>🤖 <strong>Model:</strong> Qwen/Qwen2.5-72B-Instruct:novita</div>\n"
+        response_prefix += "</div>\n\n"
         
-        # 显示模型信息
-        final_response += f"<div style='padding: 8px; background: #e3f2fd; border-left: 3px solid #2196f3; margin-bottom: 10px; font-size: 0.9em;'>🤖 <strong>Model:</strong> Qwen/Qwen2.5-72B-Instruct:novita</div>\n\n"
-        
-        # 显示工具调用日志
+        # 显示工具调用日志(精致卡片样式)
         if tool_calls_log:
-            final_response += "**🛠️ MCP Tools Used:**\n\n"
+            response_prefix += "<div style='background: #f8f9fa; border-radius: 8px; padding: 12px; margin-bottom: 16px; border: 1px solid #e9ecef;'>\n"
+            response_prefix += "<div style='font-weight: 600; color: #495057; margin-bottom: 10px; font-size: 1em;'>🛠️ MCP Tools Used</div>\n"
+            
             for i, tool_call in enumerate(tool_calls_log, 1):
-                final_response += f"{i}. `{tool_call['name']}` - {json.dumps(tool_call['arguments'])}\n"
-            final_response += "\n---\n\n"
+                # 工具图标映射
+                tool_icons = {
+                    "advanced_search_company": "🔍",
+                    "get_latest_financial_data": "📊",
+                    "extract_financial_metrics": "📈",
+                    "get_quote": "💹",
+                    "get_market_news": "📰",
+                    "get_company_news": "📢"
+                }
+                icon = tool_icons.get(tool_call['name'], "⚙️")
+                
+                response_prefix += f"<div style='background: white; padding: 8px 12px; margin: 6px 0; border-radius: 6px; border-left: 3px solid #28a745; font-size: 0.9em;'>\n"
+                response_prefix += f"<span style='color: #28a745; font-weight: 600;'>{icon} {tool_call['name']}</span>\n"
+                response_prefix += f"<div style='color: #6c757d; margin-top: 4px; font-family: monospace; font-size: 0.85em;'>{json.dumps(tool_call['arguments'], ensure_ascii=False)}</div>\n"
+                response_prefix += "</div>\n"
+            
+            response_prefix += "</div>\n\n"
+            response_prefix += "<div style='border-top: 2px solid #dee2e6; margin: 16px 0;'></div>\n\n"
         
-        final_response += response_text
+        # 流式输出最终答案
+        yield response_prefix
+        
+        # 流式调用 LLM
+        stream = client.chat_completion(
+            messages=messages,
+            model="Qwen/Qwen2.5-72B-Instruct:novita",
+            max_tokens=3000,
+            temperature=0.7,
+            stream=True
+        )
         
-        return final_response
+        accumulated_text = ""
+        for chunk in stream:
+            if chunk.choices[0].delta.content:
+                accumulated_text += chunk.choices[0].delta.content
+                yield response_prefix + accumulated_text
         
     except Exception as e:
-        return f"❌ Error: {str(e)}"
+        yield f"❌ Error: {str(e)}"
 
 # ========== Gradio 界面 ==========
 with gr.Blocks(title="Financial & Market AI Assistant") as demo: