We independently audited 6,762 Model Context Protocol servers against an OWASP-aligned trust rubric. Here's what the ecosystem actually looks like.
Live figures · last computed 2026-06-03 · methodology below
What "D or F" means — it's mostly rot, not vulnerabilities. The single biggest driver of low grades is unreachability: 13% of registry-listed servers don't respond at all (dead or unresponsive), and many that do are auth-protected or missing transparency signals, so they can't be vetted from outside. Confirmed security issues are comparatively rare — about 1% of audited servers exposed an actual problem like a credential-exfiltration surface or plaintext transport. This measures how vettable the ecosystem is, not that most servers are compromised.
How all 6,762 audited servers grade out, A through F.
Share of a 120-server sample exhibiting each issue. Unreachability dominates; genuine security findings (plaintext transport, prompt-injection, secret-exfiltration) affect roughly 1%.
Categorized by what their tools do. Click a category for its own ranked leaderboard.
+ 2,664 uncategorized — mostly unreachable or auth-protected servers whose tools couldn't be inspected from outside. Developer tooling is by far the largest identifiable category; consumer-facing categories are thin.
The top scorers in the ecosystem right now. Grades are free and identical whether or not the operator pays — independence is the point.
| # | Server | Grade | Score |
|---|---|---|---|
| 1 | arxiv.caseyjhand.com | A+ | 97 |
| 2 | mcp.influship.com | A | 96 |
| 3 | npm:mcp-accessibility-scanner | A | 96 |
| 4 | www.cannonstudio.app | A | 96 |
| 5 | api.lad.lviv.ua | A | 95 |
| 6 | hemmabo-mcp-server.vercel.app | A | 95 |
| 7 | mcp.nausika.app | A | 95 |
| 8 | npm:402-mcp | A | 95 |
| 9 | npm:@10iii/air-mcp-server | A | 95 |
| 10 | npm:@402index/mcp-server | A | 95 |
Each server is scored 0–100 against an OWASP-aligned rubric covering authentication & transport security, tool-annotation honesty, transparency (e.g. RFC 9728 OAuth resource metadata), and behavioral signals observed from real proxied traffic. Letter grades map A+ through F. The distribution and averages above are computed across the full set of 6,762 graded servers; the weakness breakdown is computed from a rolling sample of 120 full audit reports. Figures are recomputed continuously as new servers are graded and existing ones are re-checked for drift. This is an independent assessment — wmcp.sh is not affiliated with the servers listed, and grades are never influenced by payment.