← Back to Discover

Agentic Radar

Open-source command-line scanner that maps agentic AI workflows, lists their tools and MCP servers, flags known risks, and can run adversarial tests against OpenAI Agents workflows.

Testing & QACommand-line tools
Visit website ↗

What it does

Agentic Radar is a Python command-line tool for people building multi-agent systems. You point it at the code of an agentic workflow and it produces an HTML report that draws the workflow as a graph, lists the external and custom tools the agents use, and lists any MCP servers they connect to. The report also maps those tools to known vulnerabilities and ties findings to security frameworks, including the OWASP Top 10 for LLM applications and OWASP's agentic AI threat guidance. Scanning covers LangGraph, CrewAI, n8n, OpenAI Agents and Autogen projects. The main static analysis runs locally, so your source code is not shared. Two optional features need an OpenAI or Azure OpenAI key. Prompt hardening rewrites detected system prompts into more structured ones. A separate test command runs simulated attacks such as prompt injection and personal data leakage against an OpenAI Agents workflow, and custom test cases can be added in a YAML file. A sample GitHub Actions workflow lets a team run scans on every push and keep the report as a build artifact.

Capabilities

  • Workflow graph showing how agents and tools connect
  • Inventory of external and custom tools plus detected MCP servers
  • Vulnerability mapping to OWASP LLM and agentic AI guidance
  • Shareable HTML report of all findings
  • Runtime adversarial tests for OpenAI Agents workflows
  • Optional prompt hardening and a CI workflow example

When to use it

  • Reviewing the security of a LangGraph, CrewAI, n8n, OpenAI Agents or Autogen project before release
  • Finding out which tools and MCP servers your agents can reach
  • Adding an automated agent security check to a CI pipeline
  • Red-teaming an OpenAI Agents workflow for prompt injection and data leakage