Codex Octopus

Name: Codex Octopus
Author: xiaolai

@xiaolai

About Codex Octopus

One brain, many arms. Spawn multiple specialized Codex agents as MCP servers via the OpenAI Codex SDK. Each instance gets its own model, sandbox, effort, and personality.

#codex #openai #multi-agent #mcp-server #codex-sdk

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "codex": {
      "command": "npx",
      "args": [
        "codex-octopus"
      ],
      "env": {
        "CODEX_SANDBOX_MODE": "workspace-write",
        "CODEX_APPROVAL_POLICY": "never"
      }
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Codex Octopus?

Codex Octopus is an MCP server that wraps the OpenAI Codex SDK, enabling you to run multiple specialized Codex agents from any MCP client. Each agent can have its own model, sandbox, effort, and personality.

How to use Codex Octopus?

Configure it in your MCP client with the command npx codex-octopus and set environment variables like CODEX_SANDBOX_MODE and CODEX_APPROVAL_POLICY. You can also use the factory wizard to generate .mcp.json configs interactively.

Key features of Codex Octopus

Run several agents with different configurations simultaneously
Override model, sandbox, effort, approval policy, and cwd per invocation
Follow up on previous conversations via the _reply tool
Generate .mcp.json configs interactively with the factory wizard
Sandbox and approval modes only tighten per invocation for security

Use cases of Codex Octopus

A strict code reviewer operating in a read-only sandbox
A test writer with workspace‑write access to generate tests
A cheap quick helper running on minimal effort
A deep thinker using maximum effort for complex tasks

FAQ from Codex Octopus

What does Codex Octopus do that a single Codex instance cannot?

It lets you run multiple specialized Codex agents simultaneously, each with its own model, sandbox, effort, and personality, all from any MCP client.

How do I configure agents with different behaviors?

You can set environment variables like CODEX_SANDBOX_MODE and CODEX_APPROVAL_POLICY, or use per-invocation overrides for model, sandbox, effort, and more.