← All use cases
Done for developers

A task queue Claude Code and Codex can work from.

Connect Done to Claude Code or Codex with one command. Your coding agent takes its next task, works it, leaves receipts and waits for your approval, while the rest of your project lives on the same list.

Done is an MCP task manager for indie hackers and developers who use Claude Code or Codex. It gives your coding agent a real queue with priorities and dependencies, keeps your non-code tasks (launch posts, support replies, pricing changes) on the same list, and makes agent work wait for your approval before it counts as done.

The jobs, and how Done handles them

  1. Connect your coding agent in one command

    claude mcp add --transport http --scope user done https://trydone.app/mcp adds Done to Claude Code, and codex mcp add done --url https://trydone.app/mcp adds it to Codex. Both sign in with OAuth. For headless machines and CI, use a per-account bearer token instead, optionally read-only or limited to one project.

  2. Give the agent a queue, not a pile of TODO comments

    next_task picks and claims the highest-priority task that's ready in one step, skipping tasks blocked by an unfinished dependency and tasks another agent holds. decompose_task splits a big task into subtasks, and changes_since lets a new session read only what changed since the last one instead of the whole board.

  3. Review on evidence, not on a summary

    When the agent is finished it calls propose_completion and the task waits for you. Runs can leave receipts (the commit, the CI run, the test tally) and deliverables that render in the task, and get_evidence puts them next to the run trace. A tool-call or minute budget on a task makes start_run refuse once it's spent, so a stuck session stops instead of spinning.

  4. Keep the product work next to the code

    The launch post, the support reply and the pricing change sit on the same board as the refactor. Quick add, recurring tasks and Plan my day handle your share; the agent handles its own.

A week with Done

  1. Monday

    You write five tasks for this week's release: two bug fixes, a migration, the changelog and a launch post. The migration is blocked by one of the fixes. You assign the code tasks to Claude Code.

  2. Tuesday

    Claude Code calls next_task, gets the first bug fix, and proposes completion with the commit and the test run attached. You read the diff and approve.

  3. Wednesday

    With its blocker done, the migration becomes workable and next_task hands it out. You set a 40-tool-call budget on it so a stuck run stops on its own.

  4. Thursday

    Codex, connected on your laptop, drafts the changelog into its task as a Markdown deliverable. You change two lines and approve it.

  5. Friday

    You write the launch post yourself, add Monday's support replies with quick add, and check the weekly review: four of five tasks shipped, one moves to next week.

Which plan fits

Pro · $5 a month

Coding agents are chatty: one session can make dozens of MCP calls as it reads the queue, claims, comments and proposes completion. Free's 100 calls a month is enough to set it up and try it on a few tasks; Pro at $5 a month removes the cap and adds scheduled routines. Teams only matters if a collaborator needs the board.

Compare plans →

When Done isn't the right fit

If your project is a team codebase whose work already lives in GitHub issues and pull requests, keep it there, or use Linear, which lets you assign issues directly to coding agents such as Codex and Cursor. Done is for the developer who wants one queue for code and non-code work, and an approval step that isn't only a merged pull request.

Questions

How do I add Done to Claude Code?

Run claude mcp add --transport http --scope user done https://trydone.app/mcp, then sign in to Done when Claude Code asks. The Claude Code setup page also covers a token-based setup for headless machines.

Does Done work with Codex?

Yes. Run codex mcp add done --url https://trydone.app/mcp, then codex mcp login done. The Codex CLI, the IDE extension and the ChatGPT desktop app share that configuration.

Can two coding agents work the same board without colliding?

Yes. next_task claims atomically and skips tasks another agent holds a lease on. With an agent roster, each credential works only its own lane, so Claude Code and Codex can each have their own queue.

Will my agent close tasks without me?

Not by default. Tasks start at Ask first, so the agent proposes completion and you approve it. You can let a single task close on its own.

Done for other kinds of work

Start with one task you'd rather not do

Sign up with just an email, connect Claude or ChatGPT, and assign it your first task. Free to start, no credit card.

Open Done