ELYRAON MCP / Agent Permission Auditor

Most MCP servers and agent skills tell you what they say they need. This auditor shows what the package appears able to do, and where the two differ.

Point it at a local directory containing an MCP server, agent skill or tool package. It runs a non-executing static analysis of the Python source and reads the configuration manifests (MCP JSON and SKILL.md), then compares declared permissions against what the code appears to access. The result is a structured JSON report with the declared permissions, the statically observed behavior, inferred findings, evidence and any errors.

What it does

• Permission inventory lists declared, statically observed and inferred permissions for a local package.
• AST inspection analyzes Python source for calls and imports that imply resource access.
• Config parsing reads MCP JSON and SKILL.md declarations to compare against the code.
• Discrepancy reporting flags mismatches between what a package declares and what its code appears to do.
• Structured JSON output gives a schema-based report you can review or feed into other tooling.

What it is not

It is an auditor, not an enforcer: it reports findings and does not block, sandbox or modify anything. It is not a penetration test, a malware scanner or a security certification. It does not execute the target package, so inferred capability is not the same as confirmed execution, and an empty report does not prove a package is safe. See Known Limitations.