todoappapi / audit.py
GrowWithTalha's picture
feat: sync backend changes from SDDRI-Hackathon-2
84c328d
"""Audit logging service for MCP tool invocations.
[Task]: T058
[From]: specs/004-ai-chatbot/tasks.md
This module provides audit logging for all MCP tool invocations to track
usage patterns, detect abuse, and maintain compliance records.
"""
import logging
import json
from datetime import datetime
from typing import Any, Optional
from uuid import UUID
from sqlmodel import Session
from core.database import engine
# Configure audit logger
audit_logger = logging.getLogger("audit")
audit_logger.setLevel(logging.INFO)
# Audit log handler (separate from main logs)
audit_handler = logging.FileHandler("logs/audit.log")
audit_handler.setFormatter(logging.Formatter(
'%(asctime)s | %(levelname)s | %(message)s'
))
audit_logger.addHandler(audit_handler)
def log_tool_invocation(
tool_name: str,
user_id: str | UUID,
args: dict[str, Any],
result: dict[str, Any],
conversation_id: Optional[str | UUID] = None,
execution_time_ms: Optional[float] = None,
error: Optional[str] = None
) -> None:
"""Log an MCP tool invocation for audit purposes.
[From]: specs/004-ai-chatbot/spec.md - NFR-018
Args:
tool_name: Name of the tool that was invoked
user_id: ID of the user who invoked the tool
args: Arguments passed to the tool
result: Result returned by the tool
conversation_id: Optional conversation context
execution_time_ms: Optional execution time in milliseconds
error: Optional error message if invocation failed
"""
log_entry = {
"timestamp": datetime.utcnow().isoformat(),
"tool_name": tool_name,
"user_id": str(user_id),
"conversation_id": str(conversation_id) if conversation_id else None,
"success": error is None,
"error": error,
"execution_time_ms": execution_time_ms,
"args_summary": _summarize_args(tool_name, args),
"result_summary": _summarize_result(result)
}
# Log to file
audit_logger.info(json.dumps(log_entry))
# Also log to database for querying (if needed)
_persist_audit_log(log_entry)
def _summarize_args(tool_name: str, args: dict[str, Any]) -> dict[str, Any]:
"""Create a summary of tool arguments for logging.
[From]: T058 - Add audit logging for all MCP tool invocations
Args:
tool_name: Name of the tool
args: Full arguments dict
Returns:
Summarized arguments (sanitized for sensitive data)
"""
# Don't log full user content for privacy
if "message" in args:
return {"message_length": len(str(args.get("message", "")))}
# For task operations, log relevant info
if tool_name in ["add_task", "update_task", "complete_task", "delete_task"]:
summary = {}
if "task_id" in args:
summary["task_id"] = str(args["task_id"])
if "title" in args:
summary["title"] = args["title"][:50] # Truncate long titles
if "completed" in args:
summary["completed"] = args["completed"]
if "priority" in args:
summary["priority"] = args["priority"]
return summary
# For list_tasks, log filters
if tool_name == "list_tasks":
summary = {}
if "status" in args:
summary["status"] = args["status"]
if "limit" in args:
summary["limit"] = args["limit"]
return summary
# Default: return all args (tool-specific sanitization could be added)
return args
def _summarize_result(result: dict[str, Any]) -> dict[str, Any]:
"""Create a summary of tool result for logging.
[From]: T058 - Add audit logging for all MCP tool invocations
Args:
result: Full result dict from tool
Returns:
Summarized result
"""
if not isinstance(result, dict):
return {"result_type": type(result).__name__}
summary = {}
# Extract key fields
if "success" in result:
summary["success"] = result["success"]
if "error" in result:
summary["error"] = result["error"]
if "task" in result:
task = result["task"]
summary["task_id"] = task.get("id")
summary["task_title"] = task.get("title", "")[:50] if task.get("title") else None
if "tasks" in result:
tasks = result.get("tasks", [])
summary["task_count"] = len(tasks) if isinstance(tasks, list) else 0
if "updated_count" in result:
summary["updated_count"] = result["updated_count"]
if "deleted_count" in result:
summary["deleted_count"] = result["deleted_count"]
if "message" in result:
# Truncate long messages
msg = result["message"]
summary["message"] = msg[:100] + "..." if len(msg) > 100 else msg
return summary
def _persist_audit_log(log_entry: dict) -> None:
"""Persist audit log to database for querying.
[From]: T058 - Add audit logging for all MCP tool invocations
Args:
log_entry: The audit log entry to persist
"""
# Note: This could be extended to write to an audit_logs table
# For now, file-based logging is sufficient
pass
def get_user_activity_summary(
user_id: str | UUID,
limit: int = 100
) -> list[dict[str, Any]]:
"""Get a summary of user activity from audit logs.
[From]: T058 - Add audit logging for all MCP tool invocations
Args:
user_id: User ID to get activity for
limit: Maximum number of entries to return
Returns:
List of audit log entries for the user
"""
# Read audit log file and filter by user_id
try:
with open("logs/audit.log", "r") as f:
user_entries = []
for line in f:
try:
entry = json.loads(line.split(" | ", 2)[-1])
if entry.get("user_id") == str(user_id):
user_entries.append(entry)
if len(user_entries) >= limit:
break
except (json.JSONDecodeError, IndexError):
continue
return user_entries
except FileNotFoundError:
return []
# Decorator for automatic audit logging of MCP tools
def audit_log(tool_name: Optional[str] = None):
"""Decorator to automatically log MCP tool invocations.
[From]: T058 - Add audit logging for all MCP tool invocations
Args:
tool_name: Optional override for tool name (defaults to function name)
Usage:
@audit_log("add_task")
async def add_task(user_id: str, title: str, ...):
...
"""
import functools
import time
def decorator(func):
@functools.wraps(func)
async def wrapper(*args, **kwargs):
name = tool_name or func.__name__
start_time = time.time()
# Extract user_id from args/kwargs
user_id = kwargs.get("user_id") or (args[0] if args else None)
try:
result = await func(*args, **kwargs)
execution_time = (time.time() - start_time) * 1000
log_tool_invocation(
tool_name=name,
user_id=user_id or "unknown",
args=kwargs,
result=result,
execution_time_ms=execution_time
)
return result
except Exception as e:
execution_time = (time.time() - start_time) * 1000
log_tool_invocation(
tool_name=name,
user_id=user_id or "unknown",
args=kwargs,
result={},
execution_time_ms=execution_time,
error=str(e)
)
raise
return wrapper
return decorator
__all__ = [
"log_tool_invocation",
"get_user_activity_summary",
"audit_log",
]