# AgentScan

A command-line security scanner that discovers exposed MCP servers, A2A agent cards, and open LLM inference APIs across a network in a single scan.

AgentScan is a reconnaissance scanner purpose-built for AI infrastructure, going a layer beyond a generic port scanner: it determines whether a discovered HTTP service is an MCP server, an A2A agent, or an open LLM inference API, and reports the tools, agent capabilities, model lists, and authentication state it finds. It recognizes MCP servers over both Streamable HTTP and the legacy HTTP-plus-SSE transport, including OAuth 2.0 discovery for authenticated servers and signals for honeypots, A2A agent cards and their skills and interfaces, and about twenty-five inference frameworks such as Ollama, vLLM, SGLang, TGI, LM Studio, and LocalAI.

It scans a domain, an IP, or a whole subnet, or verifies a list of known host and port targets while skipping its own port-scanning step, and can be combined with an external port scanner such as masscan by feeding its open-port output back into AgentScan for protocol identification.

It is written in Go, distributed as source or precompiled Linux binaries, and its documentation states it is intended only for scanning targets a user is authorized to test.

Website: https://github.com/7anX/AgentScan