TrustScan

Security scanner for MCP servers and AI skills. Detects invisible Unicode prompt-injection, dangerous code patterns (MCP001–006), hardcoded secrets, typosquat package names. Trust scoring for the agent economy — scan before you trust.

At a glance

4
check classes
6
dangerous patterns (MCP001–006)
v0.1.0
version
Free
scanner

How it works

unicodeInvisible / steganographic Unicode characters hidden in tool descriptions and strings (prompt-injection vector).
patternsDangerous code patterns — eval/exec, shell execution, raw sockets, unsafe deserialization, obfuscated base64 (rules MCP001–MCP006).
secretsHardcoded credentials and secret material scanned out of source files.
typosquatPackage name compared against popular MCP servers via Levenshtein edit distance to catch typosquat naming.

MCP integration

Registry: io.github.entradox/trust-scan · Remote: https://trust-scan-production.up.railway.app/mcp/ 

{
  "mcpServers": {
    "trust-scan": {
      "url": "https://trust-scan-production.up.railway.app/mcp/"
    }
  }
}

Exposes two tools: trust_scan_server (directory or file, with typosquat check) and trust_scan_file (single file).

Getting started

curl -X POST https://trust-scan-production.up.railway.app/v1/scan \
  -H "Content-Type: application/json" \
  -d '{"path": "/path/to/mcp-server"}'
curl https://trust-scan-production.up.railway.app/health