Post

One for all the models out there!

Thinkst released an MCP Canarytoken—a deception-based detection mechanism that alerts when LLM agents (Claude Code, Copilot CLI, Cursor, VSCode) access and enumerate MCP servers. The token offers two alerting modes: on initial MCP connection or on actual tool invocation, allowing defenders to catch agentic reconnaissance with low noise. A key strength: agents will invoke the token via curl if MCP is organizationally disabled, bypassing attempted restrictions and still triggering alerts. The token integrates into development tool configurations (.claude/mcp.json, .copilot/mcp-config.json, etc.) as a canary to detect unauthorized agent activity in code repositories.

Read original article

This post is licensed under CC BY 4.0 by the author.