Build a Production-Ready MCP Server

Objective

Build a Python class that wraps an MCP server with production-ready features including health checks, metrics collection, and structured logging.

Requirements

Create a class ProductionMCPServer that:

1. Tracks metrics with record_tool_call(tool_name: str, success: bool, duration_ms: float):

   • Stores counts and latencies per tool
   • Returns nothing

2. Provides metrics summary with get_metrics() -> dict:

   • Returns: {"tools": {tool_name: {"calls": int, "successes": int, "errors": int, "avg_latency_ms": float}}}

3. Logs events with log(level: str, message: str, extra: dict = None):

   • Stores logs with timestamp, level, message, and extra data
   • Returns nothing

4. Gets logs with get_logs() -> list:

   • Returns list of log entries

5. Checks health with health_check() -> dict:

   • Returns: {"status": "healthy"|"unhealthy", "checks": {"metrics": bool, "logs": bool}}
   • Status is healthy if all checks pass

Example Usage

server = ProductionMCPServer()

# Record some tool calls
server.record_tool_call("search", success=True, duration_ms=150.0)
server.record_tool_call("search", success=True, duration_ms=200.0)
server.record_tool_call("search", success=False, duration_ms=50.0)

# Get metrics
metrics = server.get_metrics()
# {
#   "tools": {
#     "search": {
#       "calls": 3,
#       "successes": 2,
#       "errors": 1,
#       "avg_latency_ms": 133.33
#     }
#   }
# }

# Log events
server.log("info", "Server started", {"port": 8000})
server.log("error", "Database connection failed")

# Get logs
logs = server.get_logs()
# [{"timestamp": "...", "level": "info", "message": "Server started", "extra": {"port": 8000}}, ...]

# Health check
health = server.health_check()
# {"status": "healthy", "checks": {"metrics": True, "logs": True}}

Hints

• Use datetime.utcnow().isoformat() for timestamps
• Average latency = sum of all latencies / number of calls
• Health check should verify metrics and logs are accessible