Skip to content
MCP server

Runeward

Governed sandbox for AI agents with policy checks, approvals and signed audit logs

Runs locallyActively maintainedStatus and history

The repository was last updated 6 days ago and is published under Apache-2.0.

Install
Pricing
Open source
Platforms
CLI, macOS, Linux
Runs
Locally
GitHub stars
7
Last commit
6 days ago
Language
Rust

Similar MCP servers

About Runeward

Runeward is an open-source MCP server for developer tools. It runs locally on your machine and is started by your AI client.

Rather than a single tool, Rune Box (published as Runeward) is a governance layer around whatever agent you already run. Each shell, code, file, network or browser action is checked against a policy file and is allowed, denied or sent for human approval before it executes inside a Docker, OrbStack, Podman or Kubernetes sandbox.

It adds budgets for wall-clock time, exec, egress, tokens, cost and retry loops, a deny-by-default hostname network policy, and an append-only, hash-chained, Ed25519-signed audit trail that can be exported and verified independently. Parent and child runs keep lineage, and a child sandbox cannot widen its parent's policy.

The CLI is written in Rust and is run with runebox mcp to expose governed tools to an MCP-capable IDE. Python and TypeScript SDKs include adapters for LangChain, CrewAI, LlamaIndex, OpenAI Agents, Strands and the Vercel AI SDK. runebox quickstart demonstrates an allowed command, a denied destructive one and audit verification.

Install

Run in a terminal

claude mcp add runeward -- docker run -i --rm ghcr.io/runewardd/runeward:v0.1.4

Add to claude_desktop_config.json

{
  "mcpServers": {
    "runeward": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/runewardd/runeward:v0.1.4"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "runeward": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/runewardd/runeward:v0.1.4"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "runeward": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/runewardd/runeward:v0.1.4"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "runeward": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/runewardd/runeward:v0.1.4"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "runeward": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/runewardd/runeward:v0.1.4"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • RUNEWARD_CONFIG_DIRDirectory to resolve profile TOML files from (optional; defaults to the standard search path).
  • RUNEWARD_STATE_DIRDirectory where the tamper-evident audit ledger is written (optional).

Tools this server provides (1)

Taken from the project documentation; the running server may differ.

runebox_publish_conversation
Publish user, assistant, tool and system turns so teammates can follow a redacted live chat in the dashboard.

What it does

Policy per action
Allow, deny or require approval for each shell, file, network or browser action.
Isolated sandboxes
Run work in Docker, OrbStack, Podman or Kubernetes.
Run budgets
Cap time, exec, egress, tokens, cost and retry loops.
Signed audit trail
Hash-chained, Ed25519-signed events with exportable evidence.
Framework adapters
SDKs and adapters for common agent frameworks.
Dashboard and REST API
`runebox serve` provides a web dashboard and governed API.

Good for

  • Sandbox an autonomous coding agent
  • Require approval for destructive commands
  • Cap spend and egress per agent run
  • Export verifiable evidence of agent actions
  • Govern a group of parallel worker agents

Pricing

Open source

Apache-2.0

Runeward FAQ

Which container engines are supported?

Docker, OrbStack or Podman locally, and Kubernetes with stricter network enforcement.

Does it replace my agent framework?

No. It wraps an existing agent or multi-agent framework.

How do I try it?

Install the CLI (Homebrew or script) and run `runebox quickstart`.

What is the license?

Apache-2.0.

Is Runeward free?

Yes. Runeward is open source, so you can use and self-host it for free.

How do I add Runeward to Claude Code?

Run this command in your terminal: claude mcp add runeward -- docker run -i --rm ghcr.io/runewardd/runeward:v0.1.4

What are the best alternatives to Runeward?

The closest alternatives to Runeward are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers