assay
★ 13
vs
ADR
★ 2.1k
assay vs ADR
assay: Assay provides an open, recomputable evidence profile for privileged AI agent tool actions, enforcing policies deterministically and fail-closed. It records decisions, observations, and unproven claims into tamper-evident bundles for offline verification and security auditing.; ADR: ADR (Agentic Detection and Response) is an open-source framework designed to secure AI agents utilizing the Model Context Protocol (MCP). It comprises an ADR Sensor for telemetry collection and an ADR Detector, which serves as a dual-agent benchmark and detection system.
01
TL;DR
assaySecurity & Safety · ★ 13
Generating small, reviewable CI artifacts for bounded auditability from existing evaluation outputs, runtime decisions, or MCP tool-call tests.
ADRSecurity & Safety · ★ 2.1k
When to choose:
- You research detection for MCP-based agents and want to reproduce the MLSys 2026 ADR paper.
- You want a benchmark: ADR-Bench has 133 MCP servers and 303 tasks with baselines.
- You want to collect telemetry from Claude Code, Cursor and Codex into one schema with the sensor.
When to skip:
- You want a packaged product to deploy: this is research code for a paper.
- You want the ADR Explorer red-teaming engine: the README says it is not included.
- You cannot pay for Anthropic and OpenAI API calls: the default detector needs both.
02
Side-by-Side Comparison
Field
assayADRCategory
Security & Safety
Security & Safety
Stars
★ 13
★ 2.1k
License
MIT
Apache-2.0
Install Command
curl -fsSL https://getassay.dev/install.sh | sh
git clone https://github.com/uber/ADR && cd ADR/Detection && uv sync
Updated
1d ago
4d ago
Open Source
Yes
Yes
Tags
AI Agent Security, Tool Call Enforcement, Evidence Auditing
AI Agent Security, Telemetry, Detection System
03
Features
assay01Deterministic, fail-closed policy enforcement for tool calls, including optional kernel-level enforcement on Linux.
02Generates offline-verifiable, tamper-evident evidence bundles for audit and replay, integrated with CI without a hosted backend.
03Classifies claims by their trust basis (verified, self_reported, inferred, absent) to ensure honesty and prevent claims from exceeding observed facts.
04Policy gate (`assay mcp wrap`) for pre-execution allow/deny decisions on privileged tool calls, with clear reasons.
05Integrates with CI/CD pipelines via GitHub Actions and SARIF, enabling policy gates on pull requests and security-tab integration.
ADR01ADR Sensor for collecting and normalizing agent telemetry.
02ADR Detector, a dual-agent detection system with 133 MCP servers and 303 benchmark tasks.
03Support for the Model Context Protocol (MCP) for securing AI agents.
04Provides a step-by-step workflow to reproduce benchmark detection and paper figures.
04
Use Cases
assay↳Generating small, reviewable CI artifacts for bounded auditability from existing evaluation outputs, runtime decisions, or MCP tool-call tests.
↳Processing Promptfoo JSONL output from CI evaluations to produce evidence receipts and verified bundles.
↳Auditing and enforcing policies on MCP (Multi-Modal Command Protocol) tool calls, providing an allow/deny audit trail and observed-behavior evidence.
↳Implementing GitHub PR gates based on Trust Basis diffs and policy status, with SARIF/JUnit-ready output.
↳Generating inventory receipts and verified bundles from CycloneDX ML-BOM model components for model inventory and provenance.
ADR↳Securing AI agents that use the Model Context Protocol (MCP).
↳Benchmarking and evaluating the security performance of agentic AI systems.
↳Collecting and analyzing telemetry data from various AI agent platforms.
↳Red-teaming and defensive security research for agentic AI.
05
Best For
assay—
ADRTrending
FAQ
FAQ
What is the difference between assay and ADR?
Both assay and ADR are in the Security & Safety category. assay has 13 stars, while ADR has 2.1k stars.
Which is better, assay or ADR?
The best choice depends on your use case. Choose assay if Generating small, reviewable CI artifacts for bounded auditability from existing evaluation outputs, runtime decisions, or MCP tool-call tests., and ADR if Securing AI agents that use the Model Context Protocol (MCP)..
Is assay free or open source?
Yes, assay is open source on GitHub (MIT).
Is ADR free or open source?
Yes, ADR is open source on GitHub (Apache-2.0).
Cloud Infrastructure
AdDeploy on DigitalOcean
Host your MCP servers, Docker containers, and AI agents with $200 free credit over 60 days.
→