Blog

Keeping AI agents in bounds.

Notes for engineers who ship agents: what goes wrong when agents call tools, and how to check actions before they run.

MCP Guard

Check your agent's next tool call before it runs.

A verdict (allow, ask or block) and nine calibrated scores in one call. It reduces risk; it does not replace your permissions, allowlists and backups. 1,000 free requests, then $0.20 per 1,000 checks.