GenPark Prompt Injection Canary Leak Sentinel

Generates HMAC-signed zero-width invisible canary tokens and audits model output for leaked canaries, as a local Python MCP server.

What you can do

• Create HMAC-signed canary tokens that are invisible zero-width characters.
• Audit model output responses to detect whether canary tokens leaked.
• Support LLM guardrail checks around prompt injection and data leakage.
• Run locally as a Python standard-library script with no third-party dependencies.

When to use it

Useful when you want to test or monitor an LLM pipeline for prompt-injection-driven leakage by planting invisible canary tokens and checking whether they surface in responses.

Requirements

A local Python 3 runtime; the README configures it as a stdio server launched with python mcp_server.py. No environment variables, headers or authentication are declared.