fix: 修复MCP Server启动问题

修复内容:
1. 修复docker-compose.yml中DATABASE_URL配置
   - 从postgresql://改为postgresql+asyncpg://
   - 确保使用正确的异步数据库驱动

2. 添加缺失的Prometheus Metrics定义
   - HTTP请求指标 (http_requests_total, http_request_duration)
   - Agent管理指标 (agents_registered_total, agents_queries_total)
   - 工具调用指标 (tool_calls_total, function_tool_calls_total)
   - MCP协议指标 (mcp_requests_total, mcp_request_duration)
   - WebSocket指标 (websocket_connections_total, websocket_connections_active)
   - 系统健康指标 (redis_connections, nats_connections, database_connections, function_registry_size)

问题原因:
- DATABASE_URL格式错误导致SQLAlchemy尝试使用psycopg2而非asyncpg
- Prometheus metrics变量未定义导致启动时NameError

修复结果:
- MCP Server现在可以正常启动并健康运行
- 所有Prometheus metrics正常工作

docker-compose.yml

@@ -111,7 +111,7 @@ services:
     ports:
       - "8002:8000"
     environment:
-      - DATABASE_URL=postgresql://taiji_user:taiji_pass@postgres:5432/taiji_db
+      - DATABASE_URL=postgresql+asyncpg://taiji_user:taiji_pass@postgres:5432/taiji_db
       - REDIS_URL=redis://redis:6379
       - NATS_URL=nats://nats:4222
       - LITELLM_URL=http://litellm-gateway:4000

services/mcp-server/main.py

@@ -77,6 +77,103 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+# Prometheus Metrics定义
+# HTTP请求指标
+http_requests_total = Counter(
+    "http_requests_total",
+    "HTTP请求总数",
+    ["method", "endpoint", "status"]
+)
+http_request_duration = Histogram(
+    "http_request_duration_seconds",
+    "HTTP请求耗时（秒）",
+    ["method", "endpoint"]
+)
+
+# Agent管理指标
+agents_registered_total = Counter(
+    "agents_registered_total",
+    "注册的Agent总数",
+    ["status"]
+)
+agents_queries_total = Counter(
+    "agents_queries_total",
+    "Agent查询总数",
+    ["query_type"]
+)
+agents_active_websockets = Gauge(
+    "agents_active_websockets",
+    "活跃的WebSocket连接数"
+)
+
+# 工具调用指标
+tool_calls_total = Counter(
+    "tool_calls_total",
+    "工具调用总数",
+    ["tool_type", "status"]
+)
+tool_call_duration = Histogram(
+    "tool_call_duration_seconds",
+    "工具调用耗时（秒）",
+    ["tool_type"]
+)
+function_tool_calls_total = Counter(
+    "function_tool_calls_total",
+    "函数工具调用总数",
+    ["function_name", "status"]
+)
+function_tool_call_duration = Histogram(
+    "function_tool_call_duration_seconds",
+    "函数工具调用耗时（秒）",
+    ["function_name"]
+)
+
+# MCP协议指标
+mcp_requests_total = Counter(
+    "mcp_requests_total",
+    "MCP请求总数",
+    ["method", "status"]
+)
+mcp_request_duration = Histogram(
+    "mcp_request_duration_seconds",
+    "MCP请求耗时（秒）",
+    ["method"]
+)
+
+# WebSocket指标
+websocket_connections_total = Counter(
+    "websocket_connections_total",
+    "WebSocket连接总数",
+    ["status"]
+)
+websocket_connections_active = Gauge(
+    "websocket_connections_active",
+    "活跃的WebSocket连接数"
+)
+websocket_messages_total = Counter(
+    "websocket_messages_total",
+    "WebSocket消息总数",
+    ["direction"]
+)
+
+# 系统健康指标
+redis_connections = Gauge(
+    "redis_connections",
+    "Redis连接状态（1=连接，0=断开）"
+)
+nats_connections = Gauge(
+    "nats_connections",
+    "NATS连接状态（1=连接，0=断开）"
+)
+database_connections = Gauge(
+    "database_connections",
+    "数据库连接状态（1=连接，0=断开）"
+)
+function_registry_size = Gauge(
+    "function_registry_size",
+    "函数注册表大小"
+)
+
 # Prometheus Metrics中间件
 @app.middleware("http")
 async def metrics_middleware(request: Request, call_next):