pipelock

Why we included this project

Pipelock is an open-source firewall for AI agents, sitting between an agent and the tools, MCP servers, or websites it talks to. It inspects mediated HTTP, MCP, A2A, and WebSocket traffic for the failure modes that matter in practice: data exfiltration, SSRF, and prompt injection. Beyond blocking, it emits mediator-signed action receipts, so you get audit evidence of what an agent did, kept outside the agent and verifiable later. That makes it a good fit for security engineers and platform teams who want guardrails on agent egress without taking the agent's word for what happened. It's a focused security component rather than a general agent framework, so plan to run it alongside your existing stack.