MCP Deception

Expose decoy tools that reveal prompt-injection attempts against AI agents.

MCP service files require tools and do not use commands. Each tool has a name, description, parameters, optional annotations, and a JSON handler. The Streamable HTTP endpoint is /mcp on the configured address.

Keep the decoy tool outside any real tool execution boundary. Record invocations as security events, alert on unexpected use, and test with an isolated agent. See the MCP tripwire recipe.

Edit on GitHub

TELNET Deception

Emulate terminal-based devices and legacy systems with authentication and command responses.

Service Configuration

Reference for core and per-service YAML fields in API version v1.