Hey Product Hunt! 👋 As autonomous coding agents become part of daily development, granting them direct terminal and filesystem access creates serious risks: buggy scripts, hallucinations, or prompt injections can overwrite source trees, leak root credentials, or escape local workspaces. I built DMCPS (Docker Model Context Protocol Secured) to solve this out of the box with zero host compromise. How it works: • Hardened Isolation: Runs completely non-root in Alpine, drops all Linux kernel capabilities, enforces no-new-privileges, and mounts the container root as read-only. • Dynamic Whitelisting: Agents can only read, write, or run commands in paths explicitly approved via the dashboard. Directory traversal tricks (../) are mathematically blocked and verified via Jest test suites. • Plug-and-Play MCP: Exposes an SSE endpoint ready for Cursor, Claude Desktop, or your own autonomous pipelines. • Fast Deployment: Pre-configured for Docker Compose, Render, and Railway. DMCPS is 100% open-source under the MIT license. Looking forward to your thoughts, feedback, and security test cases!

DMCPS

Secure Docker sandbox for AI agent filesystem & shell access

Secure Docker sandbox for AI agent filesystem & shell access