This MCP server offers a comprehensive safety suite for AI agents, bringing together essential functionalities like budget enforcement for large language model calls, robust prompt injection scanning, and detailed decision tracing. Designed for seamless integration with MCP-compatible AI assistants such as Claude Code, Cursor, and Zed, it empowers developers to build and deploy more secure, cost-effective, and auditable agents directly within their IDEs. It wraps the powerful `ai-cost-guard`, `ai-injection-guard`, and `ai-decision-tracer` libraries into a single, easy-to-install solution, derived from extensive experience with autonomous AI trading agents.
主要功能
01Prompt Injection Guard: Scan and block malicious or manipulative user inputs.
02LLM Cost Guard: Enforce budgets and track spending for AI model calls.
03Seamless IDE Integration: Works with popular AI coding assistants like Claude Code and Cursor.
04AI Agent Decision Tracer: Log and summarize agent decision-making processes.
05Unified Safety Layer: Combines multiple safety tools into a single MCP server.
060 GitHub stars
