OpenWork

// Overview

OpenWork is a desktop application for macOS, Windows and Linux that stores the skills, MCP connections and connected services you use with AI agents in one place, and then serves them back to whichever agent you happen to be working in. Its own README positions it as an open-source alternative to Claude Cowork and Codex. The desktop app is the dedicated workspace, but it is optional: the same capabilities reach any MCP-compatible client through a single remote server URL.

That server is deliberately small. It exposes exactly two tools — `search_capabilities`, which finds what you are allowed to use, and `execute_capability`, which runs it — so an agent does not need to know about the individual skills, plugins or Google Workspace and Microsoft 365 integrations sitting behind them. Adding the MCP opens a browser for sign-in and organisation selection, after which anything published to you is callable from Codex, Claude Code, OpenCode or any other MCP client.

For teams there is OpenWork Den, the control plane: it provisions inference across model providers, manages members and teams, sets desktop policies and version pinning, and publishes skills and plugins through marketplaces that can be assigned to an organisation, a team or one person. It can also import Agent Plugins or Anthropic-compatible plugins and re-expose their skills and remote MCP servers. The repository uses a directory-split licence — everything outside `ee/` is MIT, while the Den code under `ee/` is source-available and converts to MIT two years after each release.

// What it does

• One remote MCP server (`search_capabilities` + `execute_capability`) that brings your skills and connections into any MCP-compatible agent
• Cross-platform Electron desktop app for macOS, Windows and Linux, or headless use straight from an existing agent
• Shares skills, plugins, MCP connections and Google Workspace / Microsoft 365 capabilities between tools, teammates and machines
• OpenWork Den control plane: teams, access control, desktop policies, model-provider provisioning and version pinning
• Marketplaces for publishing skills and plugins, with per-organisation, per-team or per-person assignment
• Imports Agent Plugins and Anthropic-compatible plugins and re-exposes their skills and remote MCP servers

// Getting started

The quickest path is to point an agent you already use at the hosted OpenWork MCP server; the desktop app is a separate, optional download.

codex mcp add openwork --url https://api.openworklabs.com/mcp/agent

claude mcp add --transport http openwork https://api.openworklabs.com/mcp/agent

{
  "mcp": {
    "openwork": {
      "type": "remote",
      "enabled": true,
      "url": "https://api.openworklabs.com/mcp/agent",
      "oauth": {}
    }
  }
}

git clone https://github.com/different-ai/openwork.git
cd openwork
corepack enable
pnpm install
pnpm dev

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it when you switch between Claude Code, Cursor and Codex and want the same skills and MCP connections in all of them
• Reach for it when a team needs one place to publish approved capabilities and control who can call which model provider
• Reach for it when you want a dedicated desktop workspace for agent work rather than driving everything from a terminal
• Reach for it when you want to reuse Anthropic-compatible plugins from agents that cannot load them natively

// How OpenWork compares

OpenWork alongside other open-source assistants & chatbots tools AI/TLDR tracks, ranked by GitHub stars.