Open-source AI Agent Firewall for MCP Security

Pipelock is an open-source AI agent firewall designed to enhance MCP security by providing agent egress control, data loss prevention (DLP), and protection against prompt injection. It operates as an egress proxy for AI agents, ensuring that all outbound and inbound traffic is monitored and controlled.

Key features:

• Network Scanning - Scans both outbound and inbound requests for security threats.
• Process Containment - Utilizes OS-native kernel primitives for unprivileged process containment.
• Signed Receipts - Generates independently verifiable proof of actions taken by the agent.
• DLP Patterns - Includes 48 built-in patterns to detect sensitive information such as API keys and tokens.
• Injection Protection - Scans for 29 patterns related to prompt injection and state poisoning.
• MCP Proxy - Wraps any MCP server with bidirectional scanning capabilities.
• Kill Switch - Provides an emergency deny-all feature to block all traffic.

Pipelock is particularly useful for developers and organizations that utilize AI agents in environments where security and data integrity are paramount. Its comprehensive scanning and containment features help prevent unauthorized access and data leaks.