Agentic Radar

Tools & Infrastructure 💻 Python ⚖️ Apache-2.0 🔴 No recent commits
1.0k stars

Agentic Radar is a security scanner for agentic workflows that analyzes code to identify tools, MCP servers, and vulnerabilities, generating an HTML report. It helps developers and security professionals understand and secure AI agent systems. It supports multiple frameworks like OpenAI Agents, CrewAI, n8n, LangGraph, and Autogen.

Agentic Radar demo
🖼️ Screenshot from the project README

✨ Key features

  • Generates comprehensive HTML security reports for agentic systems.
  • Visualizes agentic workflow graphs and identifies external/custom tools.
  • Detects MCP servers used by agents in the workflow.
  • Maps vulnerabilities to OWASP Top 10 LLM and Agentic AI frameworks.
  • Optionally hardens system prompts using LLM for best practices.
  • Runtime testing for prompt injection, PII leakage, and harmful content.

🎯 Use cases

  • Scan codebases to generate security reports for agentic workflows.
  • Identify and map vulnerabilities in AI agent systems to OWASP frameworks.
  • Test agents at runtime for prompt injection and other adversarial inputs.
  • Integrate security scanning into CI/CD pipelines for continuous monitoring.
  • Harden system prompts to improve agent instructions and security.

📦 Installation

🧰 Requirements: Python (pip) installed; optional OPENAI_API_KEY for prompt hardening and testing features.

pip install agentic-radar
Check that it is installed
agentic-radar --version

For CrewAI extras:

pip install "agentic-radar[crewai]"

For OpenAI Agents extras:

pip install "agentic-radar[openai-agents]"

🚀 Usage

Scan a LangGraph project
agentic-radar scan langgraph -i path/to/langgraph/example/folder -o report.html
Test an OpenAI Agents workflow (requires OPENAI_API_KEY)
agentic-radar test openai-agents "path/to/openai-agents/example.py"

⚠️ Good to know

Some features are framework-specific; for example, runtime testing is only supported for OpenAI Agents currently, and CrewAI extras require Python >=3.10 and <3.13.

❓ FAQ

Is my source code shared or is everything local?

The main static analysis features run completely locally, so your code is not shared. Optional advanced features like prompt hardening may send system prompts to an LLM for analysis.

Which frameworks are supported for scanning?

Scanning is supported for OpenAI Agents, CrewAI, n8n, LangGraph, and Autogen. MCP detection is available for all these frameworks.

How do I run custom vulnerability tests?

Create a YAML file with test definitions and use the --config option with the test command. You can disable default tests by setting include_default_tests to false.

What are the prerequisites for using Agentic Radar?

You only need Python with pip installed. For testing and prompt hardening features, you need an OPENAI_API_KEY or AZURE_OPENAI_API_KEY set as an environment variable.

📊 Repository

Stars★ 1,046
Forks🍴 145
Open issues🐛 15
Last commit🕒 Nov 27, 2025
Created📅 Feb 2025
Language💻 Python
License⚖️ Apache-2.0
Websitesplx.ai

🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.