ironcurtain

Overview

IronCurtain provides a secure runtime environment for autonomous AI agents, where security policies are derived from human-readable constitutions. It enforces policies through an MCP-based architecture, allowing or denying tool calls based on plain-language security rules. The system assumes the LLM may be compromised and enforces security boundaries without trusting the model's intent.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• git_status

  Check git status of a repository

  Try:Show me the git statusTry:Check if there are uncommitted changes
• git_diff

  Show differences between files in git

  Try:Show me the changes in the last commitTry:Compare this branch with main
• git_push

  Push commits to a remote git repository

  Try:Push my changes to originTry:Upload the commits to GitHub
• git_clone

  Clone a git repository

  Try:Clone the react repositoryTry:Get a copy of the project from GitHub
• git_pull

  Pull changes from a remote repository

  Try:Update my local repositoryTry:Get the latest changes from origin
• git_fetch

  Fetch changes from a remote repository

  Try:Check for remote changesTry:See what updates are available

Note: Tool names inferred from policy examples and documentation. MCP servers like filesystem and git are mentioned but specific tools not exhaustively listed.

Comparable tools

Installation

# Install globally
npm install -g @provos/ironcurtain

# Setup API keys
export ANTHROPIC_API_KEY=sk-ant-...

# Run setup wizard
ironcurtain setup

# Start with terminal multiplexer (recommended)
ironcurtain mux

FAQ

How does IronCurtain enforce security?

It enforces security through a policy engine that checks every MCP tool call against compiled rules from a natural language constitution, allowing, denying, or escalating actions.

What isolation mechanisms are used?

IronCurtain uses V8 sandboxes for code execution, Docker containers for agents with no network access, and semantic interposition through MCP servers.

On Hacker News

Recent discussion from the developer community.

• Show HN: IronCurtain: A secure* runtime for AI agent loops▲ 1💬 1
  Story by nielsprovos · 2026-03-03

Compare ironcurtain with