Agentless workstation security that inventories and controls what AI coding agents install

Safety, registered as Safety CLI Cybersecurity, was founded in Vancouver and is known in the Python world for the open-source Safety CLI scanner and the Safety vulnerability database. Its current argument is that the tooling most organizations already run was designed before coding agents existed. Endpoint detection watches processes and operating system events but not which packages, MCP servers, or IDE extensions are present. Device management tracks patches and policy but not code-layer dependencies. Software composition analysis scans what reaches the CI pipeline, not what an agent pulled onto a laptop an hour ago. Safety targets that gap, deploying through MDM with no agent to install and no developer action required.

The product line runs from visibility to prevention. An inventory covers packages across more than ten ecosystems, every MCP server, IDE extension, and AI tool in use, and whether a developer is signed into a personal or an enterprise AI account, with alerts on configuration drift. Risk detection draws on a proprietary database for Python, Java, and JavaScript that combines public feeds with model-assisted analysis and malicious package scanning; the company says this yields four times the vulnerability coverage of public sources alone. Prevention comes from a package firewall that wraps installers and an MCP server that Claude Code and similar agents consult before installing a dependency. Investors include First Round Capital, K5 Global, Kearny Jackson, and Mythos Ventures.