GARDU · SECURITY SCAN

The State of MCP Servers

A scan of the top public MCP servers and skills — who ships them, how big they are, and what a security scan finds before your AI agent trusts them.

As of Scanner —
—
Public MCP repos on GitHub (broad, "mcp" in name)
—
Servers tracked
—
Top skills
—
Files security-scanned
✓ PASS 0
! REVIEW 0
✕ FAIL 0

What We Scanned

The 7 top MCP servers (by stars) plus the Anthropic skills repo were cloned and run through the Gardu static scanner. We apply Hexbane rule philosophy across families HB101–HB701: hidden content, injection, secrets, execution & exfiltration, MCP config, and tool manifest.

Limits: This is a pattern/structure scanner that reads text — it does not execute code or follow URLs. PASS = "nothing known found", not "safe". A clean scan means no known bad patterns were detected; it does not guarantee the code is secure.

Repository Files Scanned Findings Verdict

Security Findings

Notable findings from the scan, grouped by repository.

No notable findings recorded yet. Scan status: pending.

Top Public MCP Servers

# Name Category Stars Language Last Push Scan

Top Public Skills

# Name Category Stars Language Last Push

The State

Data & Reproduce

Get the Raw Data

All data is open. Download the JSON files or browse the repository:

Reproduce the Scan

Run the scan yourself on any repository:

npx -y hexbane scan <path>

Source

GitHub API data as of . Scan performed by Gardu/Hexbane static scanner.