How we grade security on Agent Market

Of the 44,816 visible MCP servers in the catalog, 44,816 have a freshness grade. 0 of 171,396 visible listings of all types have a security grade, so every security badge you see on this site today is a dash, not a grade.

Freshness grades are live. They come from the most recent release or commit the catalog can see. Each letter from A to F stands for an age band listed on the trust page.

The trust policy says money never buys rank or a grade. Organic rank and grades are computed the same way for every listing, and any sponsored placement would be labeled and kept apart from organic results.

The trust page says the security grade will combine dependency vulnerability scans, secret detection, install-hook and obfuscation checks, prompt-injection and tool-poisoning checks on SKILL.md files and tool descriptions, and a record of the permissions each version asks for.

The trust page describes security grades as starting in November 2026. It calls grades automated assessments, not guarantees.

Security grades are not published yet. The trust page says that until a listing is scanned it shows a dash, not a grade. It also says the full security method will be published on that page before the first grade appears.

Separately from any grade, the MCP security best practices page warns that local servers run with the user's privileges and can execute arbitrary code, and that clients should show the exact command before a one-click local install.

Security grades are not published yet. Search ranking is described as a score built from security (30 percent), adoption (25), freshness (20), reviews (15) and documentation (10). Until scans run, a publisher whose domain the MCP Registry verified gets a small, temporary edge in place of the security part.

Abandoned projects rank lower over time, and publishers with very large numbers of near-identical listings are ranked lower while duplicate detection is built.

Reports of a problem with a listing go through the Report a problem link on the listing or the submit form.

Security reports get a human reply within 24 hours, and a listing confirmed as malicious comes down immediately.