Lume
Open Lume
For developers

Your task list, but your agents do the work.

Lume is a to-do app that doubles as an agent work queue. Assign a task to Claude Code or Codex, let it run, and review the result. Connect over MCP, script it with a REST API — and keep the calm, native design.

Open Lume free →How agent tasks work
# pull the next task you assigned to Codex
$ lume next --assignee codex
→ "Add tests for the date parser"

$ codex run "$task"
  …working… opened PR #214

$ lume update "$task" --review
Ready for review — your call
The difference

A to-do item is a unit of executable work

Every task app lets an AI read your list. Lume lets you hand a task to a coding agent that actually does it — then returns it for review.

Assign to Claude Code & Codex

Delegate scoped work — a refactor, tests, a draft — to an agent the way you’d assign it to a teammate. It works in the background and reports progress.

MCP server

Connect Lume to Claude, Cursor, or Codex over MCP so your assistant can query open tasks, add new ones, and update status from your editor.

REST API

Everything the app does, the API does. Poll for assigned tasks, post results, build your own loop — from a CI job or a Mac mini.

Your backlog, queryable

Stop hand-building a task store for your agents

Markdown plans rot. Queue tables are yet another thing to maintain. Lume gives agents a structured, dependency-aware list they can ask for the next ready task — without you running a parallel system.

  • Query open tasks by project, assignee, or status
  • Per-user MCP tokens — agents only see your list
  • Progress and results land back in the task timeline
  • Agents set “ready for review”, never “done”
# overnight loop on a Mac mini
while true; do
  task=$(lume next --assignee codex)
  [ -z "$task" ] && sleep 60 && continue
  codex run "$task"
  lume update "$task" --review
done

Read the full setup in running Claude & Codex on a loop.

Lume for developers — FAQ

What makes Lume a task manager for developers?+

Beyond a clean to-do app, Lume exposes your tasks over an MCP server and a REST API and lets you assign tasks to coding agents like Claude Code and OpenAI Codex. Your backlog becomes a queue agents can pull from and act on.

Can I run agents against my tasks automatically?+

Yes. Use the REST API or MCP server to poll for tasks assigned to an agent from a CI job, a cron, or a Mac mini, run Claude Code or Codex on each, then mark them for review. There’s a full walkthrough on the blog.

Will an agent ever close a task on its own?+

No. Agents can only move a task to “ready for review”. Completion is always a human action, and every change is undoable — so it’s safe to leave a loop running.

Does Lume have an API and MCP server?+

Both. The REST API covers everything the app does, and the first-party MCP server connects Lume to Claude, Cursor, and Codex. They’re available on the Pro plan.

Queue it. Let the agent run it. Review it.

Free to start. Assign your first task to Claude Code or Codex today.

Open Lume free →