AgentScan, developed by 7anX, is an AI protocol scanner that maps the emerging AI attack surface for reconnaissance and inventory tasks. The tool performs deep protocol inspection to discover MCP servers, Agent-to-Agent cards, and exposed LLM APIs, reporting available tools, skills, and authentication status. It supports targeted or full protocol scans, 25+ inference framework recognition, and JSON export for integration. Security researchers, penetration testers, and DevSecOps engineers can use it to verify authentication controls and find misconfigured AI services.

What tasks can you actually use it for?

The tool performs deep protocol inspection to discover Model Context Protocol servers, Agent-to-Agent cards, and exposed Large Language Model APIs, producing structured listings of available tools, resources, prompts, and service reachability. It provides a one-command audit to map the AI attack surface and support inventory work, targeting reconnaissance rather than active exploitation so teams can locate exposed model interfaces across internal or public networks.

How reliable are its protocol detections and reported authentication status?

The scanner recognizes more than 25 inference frameworks and open interfaces, increasing detection coverage for varied deployments. It inspects Streamable HTTP and SSE legacy implementations, checks JSON-RPC reachability for A2A components, and runs OAuth 2.0 discovery to report authentication states. Honeypot detection signals are included to flag deceptive endpoints. These outputs provide actionable leads, but they represent reconnaissance findings that require follow-up verification by operators.

What inputs and configurations determine scan coverage?

The scanner accepts IPs, domain names, and CIDR ranges, making it suitable for both internal and external audits. Users can supply masscan results or skip port discovery to validate known targets directly. Proxy options include the following configured via the --proxy argument:

• SOCKS5
• SOCKS4
• HTTPS
• HTTP

Concurrency is adjustable with TCP scanning threads defaulting to 500. The CLI is written in Go and requires Go 1.21+ for source builds; pre-built binaries are available for common platforms.

Is it practical to integrate into security workflows?

The tool emits structured JSON export for downstream processing and integrates with secondary validation pipelines by accepting masscan inputs. It is targeted at security researchers, penetration testers, DevSecOps engineers, and AI developers building MCP servers, aligning with audit and inventory stages rather than exploit testing. The cross-platform CLI and available binaries allow automation in scanner chains and CI-like environments where protocol-aware discovery is required.

Final assessment for security teams

AgentScan is a practical reconnaissance option for security teams assessing AI-facing services, given its emphasis on inventorying rather than exploiting interfaces. Outputs and authentication reports act as starting points, not definitive proof; operators should verify findings with manual tests or dedicated vulnerability assessments before remediation. Teams that feed its structured results into existing validation pipelines can use the scanner as an early detection instrument for AI infrastructure exposure.