Home

veil-mcp

An agent-native shell, delivered as an MCP server. It sits between an LLM agent and the shell and turns command effects into data: quiet structured results, addressable detail pulled on demand, and a real safety net for risky commands.

quiet-by-default · addressable · lazy-detail · structured-replaces-text

claude mcp add veil -- npx -y veil-mcp
npx -y veil-mcp init      # drop the "prefer sh_run" nudge into a project's CLAUDE.md

Why

A terminal dumps everything into scrollback; an agent swallows it all, regexes fragile text, round-trips for state, and has no undo. veil returns typed effects, stores full output addressably (so it is never re-emitted into context), and offers opt-in kernel confinement and copy-on-write checkpoints.

The token economy is a consequence of returning structure, not the bet — the structure and safety stay valuable even as context windows grow.

Pages

• Installation — npm, GitHub, from source, agent config
• Tools Reference — sh_run, sh_detail, sh_logs/sh_kill, checkpoints
• Configuration — every VEIL_* environment variable
• Sandbox and Trace — kernel confinement (sandbox enforcement) and FS tracing (structured syscall trace)
• FAQ — common questions and honest limits
• Roadmap — what's done and what's planned

Links

• 📦 npm: https://www.npmjs.com/package/veil-mcp
• 🧱 Source + ARCHITECTURE.md: https://github.com/vkmtx/veil-mcp
• 💬 Discussions: https://github.com/vkmtx/veil-mcp/discussions
• 🌱 Good first issues: https://github.com/vkmtx/veil-mcp/labels/good%20first%20issue