RugSnare

RugSnare is a runtime integrity gateway specifically designed for AI agents. This tool tackles the issue of silent changes in MCP (Model Context Protocol) tool descriptions that users often overlook. By implementing hash pinning, RugSnare ensures that any unauthorized alterations to approved tools are detected, thereby preventing potential exploits such as rug pulls and tool drift after initial approval.

Core Features

• Real-time Monitoring: RugSnare continuously monitors MCP tool descriptions, alerting users to any changes that might indicate unauthorized modifications or malicious intent.
• Integrity Verification: The tool hashes each description, ensuring that even silent changes — typically caused by maintainer updates or compromised packages — are caught, which is vital for maintaining an agent's security.
• CI Integration: RugSnare integrates seamlessly into Continuous Integration workflows, providing automated checks that can prevent builds from succeeding when discrepancies are detected.

Quick Commands

RugSnare is easy to use with straightforward commands. Here are a few examples:

rugsnare init                     # Discover MCP configurations
rugsnare scan --config .mcp.json  # Pin current tool descriptions
rugsnare diff --config .mcp.json  # Check for live changes in descriptions
rugsnare verify <artifact.tgz> --version <v>  # Verify dependency integrity

Live Proxy Functionality

In addition to the base features, RugSnare offers a live proxy mode, enabling protective behaviors during real-time interactions. With commands like rugsnare run, it enforces policy compliance and can quarantine tools that deviate from approved specifications, ensuring agents operate securely throughout their execution.

Replay and Canary Functionality

RugSnare introduces innovative canary features that allow users to replay their actual tool calls against a new version, ensuring compatibility and safety before upgrading to newer software versions. This deters potential breaking changes or other issues before deployment.

Trust and Security Model

• Zero Dependencies: RugSnare is built with security in mind, minimizing the attack surface by having no dependencies.
• Local Operation: All operations occur locally without any telemetry, ensuring that pin stores and event logs remain on the user's machine.
• On-Chain Verification: With an on-chain release log, users can verify their installs against hash values recorded at the time of release.

RugSnare stands as a powerful ally in safeguarding AI agents, rigorously enforcing integrity checks and restoring confidence in the tools that define agent behavior. It is a vital resource for organizations looking to enhance their security posture in AI environments.