trustpact 0.1.0

Behavioral trust scanner for MCP servers and AI agents

trustpact

Behavioral trust scanner for MCP servers and AI agents.

Install

pip install trustpact

Usage

# Scan a server from the Smithery registry
trustpact scan "slack"

# Scan a local server spec (JSON)
trustpact scan server.json

# JSON output for CI/CD integration
trustpact scan server.json --json

# Show AEGIS scoring methodology
trustpact info

What It Does

TrustPact scans MCP server tool definitions for manipulation patterns and calculates a behavioral trust score using the AEGIS 5-dimensional model:

• Trust Signals (35%) — metadata, documentation, authentication
• Manipulation Risk (25%) — hidden instructions, poisoning patterns
• Protection Level (15%) — auth, scope, licensing
• Vulnerability Index (15%) — critical exposure surface
• Context Modifier (10%) — runtime context signals

Attack Classes Detected

Class	Description
SIREN	Hidden instruction injection
PHANTOM	Identity spoofing
HYDRA	Coordinated Sybil attacks
MIRAGE	Capability misrepresentation
LEECH	Data/credential exfiltration
CHIMERA	Code injection, safety bypass

Trust Tiers

Tier	Score	Meaning
SOVEREIGN	95+	Highest trust
SENTINEL	85+	Proven track record
MASTER	65+	Reliable
ADEPT	40+	Limited history
FELLOW	0+	New or unverified

License

Proprietary — ARQON GmbH (i.G.)

Links

• trustpact.ai
• Patent Provisional 63/928,604