OpenWork Review: Open-Source Claude Cowork Alternative

Key highlights:

• Agent-agnostic MCP — one OpenWork MCP endpoint exposes your skills, plugins, and connected services to Claude Code, Codex, Cursor, OpenCode, or any MCP client, so you don't have to live in the desktop app.

• Share-first design — create a skill or workflow once, then share it across your own machines, teammates, or friends.

• OpenWork Den — an admin/control-plane layer for teams: provision inference, manage access per model provider, publish skills through marketplaces, and set desktop policies.

• opencode under the hood — inherits opencode's provider flexibility (Anthropic, OpenAI, local models via Ollama/llama.cpp, and more).

• Two MCP tools — `search_capabilities` finds what you can use, `execute_capability` runs it. That's the whole surface.

To understand OpenWork, you first need to understand what it's cloning.

Claude Cowork (released by Anthropic as a research preview in January 2026, later GA) is a desktop agent that grew out of Claude Code. Where Claude Code lives in the terminal for developers, Cowork brings the same agentic loop — plan, act, iterate — to a friendlier desktop app aimed at non-coders. It reads and writes files inside folders you explicitly grant, automates local apps, and runs multi-step tasks in a sandboxed VM. OpenAI's Codex desktop occupies similar ground.

The catch, for a lot of people, is obvious: these are vendor-cloud products. Your workflows, your file context, and your model choice are all mediated by one company's stack.

The MCP integration is the feature I'd actually reach for. Instead of running the OpenWork desktop app, you register its remote MCP server and pull your capabilities into the agent you already live in.

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

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

For solo users, the desktop app plus MCP is the whole story. For organizations, OpenWork Den is the control plane, and it's where the product ambition shows:

- Provision inference at scale and control which members and teams can use each model provider.

- Invite teammates, create teams, and manage access from one place.

This is a condensed version. The full article includes architecture diagrams, code examples, comparison tables, and practical use cases.

👉 Read the complete analysis on andrew.ooo

Topics: openwork, claude-cowork, opencode, ai-agents, desktop-agent