Free · in your browser · declared surface only

See what your AI agents can actually reach.

Paste or upload your agent's MCP config (or an npm lockfile) and get back a clear inventory of the models, MCP servers, tools, and scopes it trusts, with any components that carry known-exploited vulnerabilities flagged. We redact secrets before processing and store nothing from your upload. Free, and standards-based (CycloneDX).

Reads a claude_desktop_config.json-style MCP config (the servers, tools and scopes your agent trusts) or an npm lockfile (the declared build surface). Max 2 MB. Everything runs in memory.

No scanner to install. Standards-based CycloneDX output you can keep.

What it does

  • Inventories the declared surface of your agent: the host model, MCP servers, tools, scopes, and data connectors your config trusts.
  • For a lockfile, flags components carrying known-exploited (CISA KEV) or high-likelihood (EPSS) vulnerabilities, as of our latest published vulnerability dataset (not a real-time feed).
  • Surfaces configuration exposure worth a review: auto-approved tools, servers that hold a credential, and the tool-description surface.
  • Returns a clean CycloneDX document you can share with your team or your auditor.

What it does not do

  • It reads the declared surface (what you have configured). It does not discover shadow or runtime-added servers that would need instrumentation inside your environment.
  • It flags where an injection surface exists. It does not run an injection test or prove a path is exploitable. The independent, signed injection test is the paid layer we are building toward.
  • It is a starting map, not a complete picture, and it is not a signed report. Certified, human-signed reports are a separate CyberOrbit engagement.