Three agents working right now

Half your to‑do list
can do itself.

Done is one board for the work you keep and the work you hand off. Assign a task to an AI agent, it does the job and reports back — and nothing reaches done without your sign-off.

YoursAgentWaiting on you
Today · 4 of 24
Q3 investor update
Agent finished a draft · needs your sign-off
Review
Refactor the digest email templates
Running
AR
Decide the Teams price
Yours — judgement call
Today
Migrate workspace_members index
Shipped by agent · you approved 2h ago
/mcpclaude · chatgpt · cursor
agent shipped “changelog 0.9.4”? agent asked “ship to prod or staging?” claimed “backfill tags table” 41 tasks closed overnight you approved 6 results plan proposed for “store screenshots”

One board.
Two kinds of doer.

Every task says who has it: you, an agent, or you again for review. No second tool, no separate agent dashboard, no wondering where a piece of work went.

Yours

Keep the judgement calls

Pricing, hiring, the email you'd never let a machine send. Capture in plain language — the date, priority and project parse out of one line.

call sarah re: seed fri !urgent ~30m
Theirs

Hand off the rest

The agent claims the task, proposes how it will work, does it, and posts the result into the task's own thread. Blocked work stays blocked.

Plan approved11:02
Working · 14 tool calls11:09
Back to you

The last word is yours

Plans to green-light, questions to answer, finished work to sign off — one queue, cleared in a couple of minutes from the keyboard.

ApproveAsk for changes
The handoff

Watch a task leave, work, and come back.

It travels over your account’s own MCP endpoint — not a shared key, not a third party’s queue — and returns to the row it left, with a thread of everything that happened.

01
You assign
One click, or type @agent as you capture.
02
It proposes a plan
On anything substantial, it says how before it starts.
03
It works, and narrates
Progress, tool calls and artifacts land in the thread.
04
You close it
Or send it back with a note. Nothing self-signs.
While you sleep

Wake up to work
already done.

Point an agent at your queue on a schedule and it works unattended. You set the caps — tasks, minutes, tool calls per night. Questions collect quietly and you clear them over coffee.

41
tasks closed
3
questions parked
0
budget overruns
23:0002:0005:0007:30
23:14Regenerated the changelog for 0.9.4auto-ok
01:02Triaged 18 inbound bug reportsauto-ok
03:37Asked: “Refund the annual plan in full?”parked
05:48Drafted the Q3 investor updatefor review
06:10Stopped — nightly budget reachedcapped
The approval gate

Nothing ships
unreviewed.

An agent can propose a completion. It cannot mark your task done — not through the API, not through a flag it sets itself. The permission lives in the credential, so there is no route around it.

  • Agents can't raise their own budgets
  • Agents can't assign work to themselves
  • Tokens narrow to read-only, or one project
  • Deletes sit in Trash for 30 days
Claude proposed completion
Q3 investor update · 6 min ago
PENDING
Pulled the 41 shipped items from the changelog, grouped them into four themes, and wrote the update at 620 words. Numbers cross-checked against Stats. Two claims I could not verify are flagged inline.
Under 700 words
Every metric sourced
Reviewed by a human
Approve & closeAsk for changes
MCP-native

Bring your own agent. Any of them.

Every account gets its own MCP endpoint over OAuth or a bearer token. Paste one URL and your client is connected — scoped to your data, revocable in a click.

ClaudeChatGPTAny MCP clientSlackGoogle Calendar
$ claude mcp add --transport http done https://trydone.app/mcp

Everything else a serious list needs.

Auto-decompose

One vague task becomes a clean set of subtasks in a single action — by you, or by an agent on your behalf.

Command palette

Anything, anywhere, without the mouse — and it searches what happened, not just titles.

K

Focus mode

One task, full screen, a timer. Everything else disappears.

Dependencies

Cycle-checked “blocked by” chains keep agents from jumping the queue.

Goals & projects

A North Star tier above projects, so a week’s tasks ladder up to something.

You vs. agent, measured

Stats show the completion trend, your streak, and honestly how much of what shipped came from an agent.

Plan my day, around the meetings you actually have

Your open work, time-blocked into the gaps your calendar leaves. When a meeting moves, it re-plans and tells you what changed — “your 2:00 PM moved to 3:00 PM, two blocks shifted” — instead of quietly showing you a day you no longer have.

Close the day

What shipped, who closed it, what rolls over — and one line on how it went.

Say it, don’t type it

Ramble five things at once; they arrive as five separate tasks to keep, edit or drop.

AlsoDescribe a view in a sentenceRead an agent’s draft in the taskRecurring tasksTags & smart viewsEmail digestTrash · 30-day undoShared workspacesHome-screen widgetsiPhone · iPad · Android · MacCSV & Markdown export

Priced like a tool, not a seat tax.

Every core feature is on the free tier. You pay when your agents get busy.

Free
$0
forever · email only, no card
Every core feature
100 agent calls / month
One workspace
Start free
MOST PICKED
Pro
$5/mo
for people who actually delegate
Unlimited agent calls
Overnight runs & budgets
Calendar sync · digests
Go Pro
Teams
$20/mo
the whole team, not per seat
Shared workspaces
Per-project agent tokens
Everyone sees every run
Add your team
FAQ

Questions about the AI to-do app

What is Done?

Done is an AI to-do app and task manager. You capture tasks in plain language, and you can assign the ones you don't want to do yourself to an AI agent — Claude, ChatGPT, or any MCP-compatible model — that reads the task's context, does the work, and reports back in its activity feed.

How is an AI to-do app different from a normal to-do list?

A normal to-do list only tracks work. Done also helps finish it: quick capture parses dates, priority, and projects as you type; auto-decompose breaks a big task into subtasks; and any task can be handed to an AI agent — Claude, ChatGPT, or another model — instead of sitting in your backlog.

Can I assign tasks to an AI agent?

Yes. Every task can be assigned to an AI agent. Because Done is MCP-native, whichever model you connect — Claude, ChatGPT, or another — picks it up, works on it, and posts progress and results straight into the task's activity feed.

Which AI models can I connect?

Any MCP-compatible client. Done is MCP-native: every account gets its own remote MCP endpoint over OAuth or a bearer token, so Claude, ChatGPT, Cursor, and custom agents all connect securely and stay scoped to your data alone. See the docs for setup.

What can the AI actually do?

It can claim a task assigned to it, work through the queue by priority, auto-decompose a task into subtasks, and comment results back. Recurring and dependency rules keep it from jumping ahead of blocked work.

What happens when the AI gets stuck or needs a decision?

It parks the question and moves on to the next task instead of stalling. The question waits for you in the Control Room with the agent's own recommendation and reasoning, so answering is usually one click — and you can answer straight from a Slack DM. Answering hands the task back to the agent with your answer attached; it never approves or completes anything on its own.

Can I let an AI agent work my tasks overnight?

Yes. Point any MCP client at Done on a schedule and it works the queue unattended. You set caps on how far one night may go — tasks, minutes and tool calls — and choose whether questions ping you as they're asked or collect quietly so you clear them in one pass in the morning.

How much does Done cost?

Done has a free tier with every core feature and 100 AI agent (MCP) calls a month — no credit card. Pro is $5/month for unlimited agent calls, and Teams is $20/month to add teammates and share workspaces.

Give something on your list away tonight.

Sign up with an email, connect your agent, and see what an empty board feels like.