Skip to main content
MCP server CX Codex

Cowork — Claude Code MCP server: what it does, how to install

Cowork is a Claude Code MCP server that routes tasks across multiple agents using fallback chains during API rate limits. It keeps task context intact when switching execution to backup models.

MIT ★ 6 updated 2026-11-15

Open on GitHub

What Cowork does

Cowork runs a self-hosted control plane and a dark web interface on port 6868 to unify Claude Code, Codex, Gemini CLI, and local runtimes into one task inbox. When a provider returns a quota error, the dispatcher transfers execution to the next configured model in sequence.

State management relies entirely on plain JSON and Markdown files on disk without SQL or Redis migrations. The dashboard tracks CPU and GPU metrics, manages scheduled calendar routines, and includes 285 preconfigured specialist personas for routing across 19 technical domains.

What it helps with

  • reroute running tasks to backup models when API limits hit
  • monitor multi-agent execution and host thermals on port 6868
  • attach remote machines into an agent cluster using a single Node script
  • persist task history and artifacts without database migrations

Example prompts for the agent

  • Send this refactoring task to the Cowork queue
  • Check the status of pending jobs in Cowork
  • Reroute the stalled task to the fallback model in Cowork

How to install Cowork

Install

The MCP registry has no launch command for it: see the repository README

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about Cowork

How to install Cowork in Claude Code?

Clone the repository with git clone https://github.com/slashman413/cowork && cd cowork, run deploy/install-server.sh to launch the server, and execute deploy/install-skill.sh to configure the MCP connection.

What does Cowork do?

It provides a shared dispatch inbox for coding agents, automatically handing off workloads to alternative models when an API reaches rate limits.

Does Cowork require a database?

The server runs without PostgreSQL, Redis, or SQLite. It stores all tasks, execution history, and artifacts directly on disk in JSON and Markdown files.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open