ADR
ADR is an enterprise-grade AI agent security system providing observability, security benchmarking, and threat detection.
- Type
- MCP
- Transport
- http
- Open source
- Yes
- GitHub Stars
- ★ 1.2k
- Source
- mcp-github
- Repository
- github.com/uber/ADR
Overview
ADR is an enterprise-grade AI agent security system that helps organizations protect AI agents used by employees and customers. It safeguards AI agents through four complementary capabilities: observing agent activities, assessing defensive measures, detecting threats, and preventing unsafe behaviors. ADR has been deployed in Uber's production environment and was presented in a paper at MLSys 2026. The system is suitable for enterprises needing to protect internal automation tools and customer support agents.
Capabilities
- ▪Observe agent activities
- ▪Assess defensive measures
- ▪Detect threats
- ▪Prevent unsafe behaviors
Use cases
Setup
git clone https://github.com/uber/ADR; cd ADR/Detection; uv sync; export ANTHROPIC_API_KEY="..." OPENAI_API_KEY="..."
This information was compiled by AI from public sources and may contain inaccuracies — please refer to the source.
FAQ
What are ADR's main features?
ADR provides observability, security benchmarking, and threat detection to protect enterprise AI agents.
How do I get started with ADR Detection?
Clone the repository, navigate to the Detection directory, sync dependencies, and set up API keys.
Related skills
xiaohongshu-mcp
Xiaohongshu MCP enables AI assistants to directly access and operate Xiaohongshu data.
internet-court-skill
Provides a trust layer for agent-to-agent transactions, including natural language authorization, ERC-7710 delegation permissions, x402 payments, escrow, and dispute resolution.
lenny-skills
86 product management skills extracted from Lenny's podcast, applicable to Claude Code and AI agents.
Day1Global-Skills
Provides investment analysis capabilities for AI agents, covering tech stock earnings, value investing, market sentiment, macro liquidity, and Bitcoin cycle analysis.
paperasse
AI agent skills tailored for the French bureaucratic system, covering accounting, notarization, taxation, and more.
Product-Manager-Skills
A practical, battle-tested product management skill framework designed for AI agents such as Claude Code, Cowork, and Codex.