Trust · Capability · Risk

MCP server security inspector

Every MCP server triaged for filesystem reach, shell exec, network calls, secrets, browser automation. Trust scored 0–100 with transparent signal weights.

Start with a featured server

How the score works

  1. 1. Capabilities

    8 capability kinds matched against server metadata: filesystem read/write, shell, process spawn, DB write, network, browser, secrets.

  2. 2. Signals

    6 signals: Loadout-verified, official author, GitHub stars, public source, stability tag, capability surface. Each carries a fixed weight.

  3. 3. Score & grade

    Sum of signals → 0–100. Mapped to A / B / C / D / F. Higher is safer to install with default scope.

Security inspector — FAQ

  • What does the security inspector detect?

    It runs an in-browser heuristic over each catalogued MCP server: filesystem reach, shell exec, network calls, secrets handling, browser automation. Each capability gets a risk level (low / medium / high) and contributes to a trust score (0–100).

  • Is this a real security audit?

    No — it is a first-pass triage. The score is built from heuristics over server metadata and install commands, not full source code analysis. Use it as a starting point for your own review, not the final answer.

  • How is the trust score computed?

    Six signals contribute: verification status, official authorship, community traction (GitHub stars), public source availability, stability tag, and capability surface (more high-risk capabilities lower the score). Each signal carries a fixed weight.

  • Can I audit a server that is not in your catalog?

    Submit it through /submit so we ingest it. A direct "paste-a-repo-URL" mode that fetches GitHub data live is on the roadmap.