Agentโ™ฅ๏ธŽAge
Catalog

io.github.IronSecCo/ironclaw

Official

by IronSecCo ยท Go

Sandboxed shell exec for MCP clients: run untrusted agent commands in a gVisor container.

io.github.IronSecCo/ironclaw MCP Server

The io.github.IronSecCo/ironclaw MCP server provides a sandboxed shell exec capability for MCP clients. It runs untrusted agent commands inside a gVisor container, with the intent that the sandbox cannot phone home, access the host, or rewrite its own rules.

๐Ÿ› ๏ธ Key Features

  • Sandboxed shell execution for MCP clients
  • gVisor container isolation
  • Intended constraints: no phone-home behavior, no host access, no self-rule rewriting
  • Positioned as self-hosted

๐Ÿš€ Use Cases

  • Running AI agent commands when the commands are untrusted
  • Deployments needing stronger isolation for agent-platform workflows

โšก Developer Benefits

  • Supply-chain/security signaling via linked OpenSSF badges in the provided excerpt
  • Self-hosted approach aligned with security and isolation topics

โš ๏ธ Limitations

  • No tool catalog count or additional MCP tool details were provided in the source data
  • Only the sandboxing shell-exec behavior is described
sandboxsecurityshell executioncontainersgvisorisolationagent safety

Topics

agent-platformai-agentsai-assistantclaudegolangmcppersonal-assistantsandboxsecurityself-hostedgvisorllmopen-sourcesupply-chain-security

Related servers

More in Security