Trust & Security

Last Updated: July 10, 2026

MetaMesh-UGA operates a strict Zero Trust model for third-party Model Context Protocol (MCP) servers. We treat all indexed servers as non-trusted by default until validated, scanned, and ranked by our automated control plane.

1. Trust Score Formula

Our Trust Engine computes a score between 0.0 and 1.0 for each server daily based on these weighted components:

Uptime (25%): Based on synthetic health checks performed every 5 minutes.
Latency (15%): Normalized average response times. Faster responses increase the score.
Success Rate (25%): Percentage of successful versus failed RPC calls.
Popularity (15%): Historical usage volume across all tenant agents.
Security Score (20%): Outbound CVE scan results and malware detection checks.

2. Security Scanning & Sandbox Isolation

Every server is scanned for dependencies, CVE vulnerabilities, and malware. We run a normalized security score:

Security Score = 1.0
- 0.3 for each Critical CVE
- 0.2 for each High CVE
- 0.1 for excessive permission capability request
- 0.0 if malware or spyware signature is detected

Servers with a security score below 0.5 are placed in quarantine and execution is blocked by default.

3. Egress Control & SSRF Protection

Our Data Plane applies DNS pinning and egress proxy filtering to prevent Server-Side Request Forgery (SSRF). Any tool attempt to request metadata endpoints (e.g., `169.254.169.254`), localhost, or private networks (RFC1918) is blocked instantly and triggers a security alert.

4. Contact & Security Team

Our vulnerability disclosure policy is published at /.well-known/security.txt. Report any security issues to [email protected].