Ironclaw
Run untrusted agent shell commands in a gVisor sandbox from any MCP client
- Pricing
- Open source
- Platforms
- self-hosted, CLI, Linux
- Runs
- Locally
- Version
- 0.1.569
- Listing updated
- 3 days ago
About Ironclaw
Ironclaw is an open-source MCP server for security. It runs locally on your machine and is started by your AI client.
The MCP package is a thin client in an OCI image with no host privileges. It exposes a sandbox_exec tool and hands every run to an IronClaw control-plane you operate, which owns the hardened gVisor launch. If the control-plane URL is unset the tool fails closed, and the control-plane's bearer token is also required.
IronClaw itself targets self-hosted autonomous agents that you do not have to trust. According to its README, each sandbox runs with no network, reaches the model only through a host proxy, and cannot change its own configuration. A bundled demo script lets a jailbroken agent try to phone home, read the host filesystem and reach the Docker socket, and prints what was denied. A separate scan action grades a container's isolation from 0 to 100 in CI.
The README warns it is alpha software: flags, on-disk formats and interfaces can change without migration, and not every feature is tested end to end. Docker must be running for the demo.
Install
Run in a terminal
claude mcp add ironclaw --env IRONCLAW_CONTROLPLANE_URL='<IRONCLAW_CONTROLPLANE_URL>' --env IRONCLAW_API_TOKEN='<IRONCLAW_API_TOKEN>' -- docker run -i --rm -e IRONCLAW_CONTROLPLANE_URL -e IRONCLAW_API_TOKEN ghcr.io/ironsecco/ironclaw-mcp:v0.1.569Add to claude_desktop_config.json
{
"mcpServers": {
"ironclaw": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"IRONCLAW_CONTROLPLANE_URL",
"-e",
"IRONCLAW_API_TOKEN",
"ghcr.io/ironsecco/ironclaw-mcp:v0.1.569"
],
"env": {
"IRONCLAW_CONTROLPLANE_URL": "<IRONCLAW_CONTROLPLANE_URL>",
"IRONCLAW_API_TOKEN": "<IRONCLAW_API_TOKEN>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"ironclaw": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"IRONCLAW_CONTROLPLANE_URL",
"-e",
"IRONCLAW_API_TOKEN",
"ghcr.io/ironsecco/ironclaw-mcp:v0.1.569"
],
"env": {
"IRONCLAW_CONTROLPLANE_URL": "<IRONCLAW_CONTROLPLANE_URL>",
"IRONCLAW_API_TOKEN": "<IRONCLAW_API_TOKEN>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"ironclaw": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"IRONCLAW_CONTROLPLANE_URL",
"-e",
"IRONCLAW_API_TOKEN",
"ghcr.io/ironsecco/ironclaw-mcp:v0.1.569"
],
"env": {
"IRONCLAW_CONTROLPLANE_URL": "<IRONCLAW_CONTROLPLANE_URL>",
"IRONCLAW_API_TOKEN": "<IRONCLAW_API_TOKEN>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"ironclaw": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"IRONCLAW_CONTROLPLANE_URL",
"-e",
"IRONCLAW_API_TOKEN",
"ghcr.io/ironsecco/ironclaw-mcp:v0.1.569"
],
"env": {
"IRONCLAW_CONTROLPLANE_URL": "<IRONCLAW_CONTROLPLANE_URL>",
"IRONCLAW_API_TOKEN": "<IRONCLAW_API_TOKEN>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"ironclaw": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"IRONCLAW_CONTROLPLANE_URL",
"-e",
"IRONCLAW_API_TOKEN",
"ghcr.io/ironsecco/ironclaw-mcp:v0.1.569"
],
"env": {
"IRONCLAW_CONTROLPLANE_URL": "<IRONCLAW_CONTROLPLANE_URL>",
"IRONCLAW_API_TOKEN": "<IRONCLAW_API_TOKEN>"
}
}
}
}Configuration
IRONCLAW_CONTROLPLANE_URLBase URL of your running IronClaw control-plane, e.g. http://127.0.0.1:8787. This image is a thin client with no host privilege: it delegates every sandbox_exec run to the control-plane, which owns the hardened gVisor launch. Unset means no backend and the tool fails closed.RequiredIRONCLAW_API_TOKENBearer token for the control-plane API (the value the control-plane was started with).Required
Tools this server provides (1)
Taken from the project documentation; the running server may differ.
- sandbox_exec
- Runs a command in a sandbox through the IronClaw control-plane.
What it does
- Sandboxed command execution
- Runs shell commands from an agent inside a gVisor container.
- Fail-closed backend
- The tool refuses to run when no control-plane is configured.
- Thin client image
- Holds no host privilege and delegates launches to the control-plane.
- Containment demo
- A one-command script shows a jailbroken agent being denied.
- Isolation scoring
- A scan action grades container isolation from 0 to 100 in pull requests.
Good for
- Let a coding agent run unreviewed shell commands safely
- Test how an agent behaves when jailbroken
- Gate merges on a sandbox isolation score
- Self-host agents without giving them host access
Pricing
Open source
Prices change. Check ironclaw.sh before you buy.
Ironclaw FAQ
What does it need to run?
A running IronClaw control-plane URL and its bearer token.
What happens without a control-plane?
The tool fails closed.
Is it production ready?
The README calls it alpha software and warns that interfaces can change.
How is it distributed?
As an OCI image from ghcr.io, run over stdio.
Is Ironclaw free?
Yes. Ironclaw is open source, so you can use and self-host it for free.
How do I add Ironclaw to Claude Code?
Run this command in your terminal: claude mcp add ironclaw --env IRONCLAW_CONTROLPLANE_URL='<IRONCLAW_CONTROLPLANE_URL>' --env IRONCLAW_API_TOKEN='<IRONCLAW_API_TOKEN>' -- docker run -i --rm -e IRONCLAW_CONTROLPLANE_URL -e IRONCLAW_API_TOKEN ghcr.io/ironsecco/ironclaw-mcp:v0.1.569
What are the best alternatives to Ironclaw?
The closest alternatives to Ironclaw are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.
Alternatives to Ironclaw

REA
MCP serverCLI and MCP server that lets coding agents reverse engineer apps and binaries
Runs locally- Open source
- 464K GitHub stars
- Developer tools

Codebase Memory
MCP serverLocal MCP server that builds a code knowledge graph for AI coding agents
Runs locally- Open source
- 48.7K GitHub stars
- Knowledge and memory

GitHub
MCP serverOfficial MCP server connecting AI assistants to GitHub repos, issues and workflows
Not checked yet- Remote
- Open source
- 34.5K GitHub stars
- Developer tools

Serena MCP: the IDE for your agent
MCP serverMCP toolkit giving coding agents symbol-level code retrieval and editing
Runs locally- Open source
- 31.5K GitHub stars
- Developer tools

XERJ
MCP serverLocal search engine for AI agents with cited answers over your code and docs
Runs locally- Open source
- 19.2K GitHub stars
- Knowledge and memory

Ouroboros
MCP serverSpec-first layer that pins requirements and verifies results of AI coding agent work
Runs locally- Open source
- 6.4K GitHub stars
- Developer tools