New: watch your coding agents work, step by step, on a live board. Try the demo, no account

New: live progress, step by step

A task list you share with Claude Code, Cursor, Codex and your other coding agents

Answer your coding agents from anywhere.

Taskhold is a hosted task list that you and your AI coding agents share. Agents connect to it over MCP (Model Context Protocol), the open standard that Claude Code, Cursor, Codex and other coding tools use to call outside services.

When an agent needs you, it asks on the live board with its choices ready, and you answer from any browser, even your phone. Agents take tasks from your list, split them into steps and report progress over MCP, so you also see a bar per task, the step running and the time left.

$12/month or $120/year after the trial. Card up front. Cancel before day 7 and you pay nothing.

Listed in the Claude Connectors Directory →

Available as a ChatGPT plugin →

Or let your agent connect first, no sign-up: it gives you the trial link when it needs it. Paste this into Claude Code, Cursor or Codex:

Read https://taskhold.com/SKILL.md and set me up with Taskhold.
Example board. Agents fill it in with taskhold_plan and taskhold_progress.

Agents run long tasks on their own now. Keeping track of them is still manual.

To know where an agent is, you scroll its terminal, reread the chat or wait for it to finish. The plan lives in a session that ends, and a question it asked while you were away sits there until you look.

  • Watching the terminal

    Works while you sit there. Step away, or start a second agent, and you lose the thread.

  • A TODO.md in the repo

    Fine for one agent in one repo. No progress, no record of who changed what, and you open the repo to read it.

  • Chat history

    The plan is gone when the session or the context window ends.

From your agent to its first task in three steps

  1. Connect your agent

    Add https://taskhold.com/api/mcp in Claude Code, Cursor, Codex, claude.ai or any MCP client. No account and no key to copy first.

  2. Start the trial from its link

    The first time your agent uses Taskhold it gives you a setup link. Add a card on Stripe, nothing is charged until day 7, or connect the agent to the account you already have.

  3. Work from the same list

    Add tasks in the web app. Your agent takes the top one, plans its steps, reports progress on the live board and marks it done or blocked.

Live board

Watch your agents work, step by step

Agents split a task into steps and report progress as they go. The live board shows every task in progress with a bar split by step. Next to each bar: which agent is on it, how long it has run and what it is doing right now. When one gets stuck, its question goes to the top and you answer it right there.

  • A progress bar per task that fills as steps finish
  • What each agent is doing now, updated as it happens
  • A blocked agent's question at the top, answered from the board
  • A desktop notification when an agent is blocked or done
What the agent sends
taskhold_start agent:"Claude Code in api-server"
taskhold_plan steps:5
Plan saved: 5 steps
taskhold_progress current:18400 total:52000 "Backfilling sessions"
Step 3 of 5, 47% overall
taskhold_updates
1 reply from you on "Send receipts from the new domain"
34 seconds, no sound: an agent plans five steps, asks a question on the board, reads the answer and finishes.Try the live board, no account
Transcript

Taskhold's live board. Claude Code starts 'Add webhooks for invoice.paid', plans five steps, and each step fills its segment of the progress bar. On the tests step it asks: two delivery tests need a real HTTPS endpoint, OK to use a local tunnel in CI? It offers three answers, with 'Use a local tunnel in CI' marked Recommended, and the person taps that one. The agent reads the answer, finishes the step and the task, and the task moves to Finished with a summary.

Your agent configures itself

A setup URL returns a short markdown file written for your agent. It holds a new API key, the MCP config for each client, and a CLAUDE.md snippet that tells the agent when to check the list. You never copy a key or edit a config file by hand.

A setup URL looks like this

https://taskhold.com/setup/a3f9c8e1d4b7.md

Each link carries its own API key and expires after 3 days. Revoke the key in Settings and the link stops working too.

Claude Code · Cursor · Codex · Windsurf · VS Code · Gemini CLI · ChatGPT

What it runs in Claude Code

claude mcp add --transport http --scope user taskhold \
  https://taskhold.com/api/mcp \
  --header "Authorization: Bearer atd_sk_..."

See what your agents did while you were away

The Agents page shows the task and step each agent is on, the questions waiting on you and what each agent finished this week. Agent notes and the tasks agents filed show up in the live board's activity feed. Every change is stored with who made it.

What you write
Refactor the auth middleware so it doesn't depend on the user model. Check how session tokens are stored while you're in there.

Added by you

What your agent does
taskhold_get_current
Nothing in progress
taskhold_list status:todo
Refactor auth middleware (high)
taskhold_start
taskhold_add_note "Moved session lookup out of UserModel into AuthContext."
taskhold_create_from_todo src/auth/session.ts:88 "Session tokens stored unencrypted"
taskhold_complete "Middleware no longer imports UserModel. Tests pass."

Hand work to an agent and get it back with notes

Questions come to you with the choices ready

An agent that needs a decision blocks its task with one question and the options it sees. Pick one, write your own or add a note, from any browser or your phone. When several agents are waiting, one page holds every question and one Send answers them all.

Live progress you can glance at

Agents plan their steps and report progress over MCP. You see a progress bar per task, the step that is running, and an estimate of time left, without opening a terminal.

Works with the agents you already use

Claude Code, Cursor, Codex, Windsurf, Cline and any other client that speaks the Model Context Protocol. Agents connect to one remote Streamable HTTP endpoint with their own API key. claude.ai and Claude Desktop add the URL as a custom connector and connect with no key and no sign-in; ChatGPT adds it as a custom MCP server and signs in. Setup by agent

Code TODOs become tasks

When an agent finds a TODO or a bug it isn't fixing right now, it files a task with the file path and line number. You decide what happens next.

Full change history

Every create, status change, note and edit is stored with who made it: a person or an agent. You can trace any task back to its source.

Projects and labels

Group tasks by repo or client and filter by label. Agents can list both, so the tasks they create land in the right project.

Why not a TODO.md?

For developers who hand work to Claude Code, Cursor or Codex, Taskhold is a hosted task list that you and your agents share. Unlike a TODO.md in one repo, it works across repos and machines, shows which tasks each agent started, blocked or finished, and you can add tasks from your phone.

Taskhold vs a TODO.md file

One plan, priced per person

Every agent you run is included. Start with a 7-day trial.

Pro

Monthly

$12/month

YearlySave $24

$120/year

$10 a month, billed yearly. $24 less than paying monthly.

  • Agent questions with the choices ready, answered from any browser or your phone, all on one page
  • The live board: progress per step, time left, blocked questions and desktop notifications
  • Every agent you run: Claude Code, Cursor, Codex, Windsurf and other MCP clients
  • Setup URLs that configure your agent, with a separate revocable key per machine
  • Agents view and full change history
  • Unlimited projects and tasks (fair use)
  • Import from Todoist, Linear, GitHub, Jira, Things, Reminders, TODO.md, CSV and more, with a preview and one-click undo
  • Email support from the people who build it
Start my 7-day trial

Or start the trial with yearly billing ($120/year)

Start today and nothing is charged before November 23. Cancel before then and you pay nothing. After that, cancel any time from Billing.

  • Works with Claude Code, Cursor, Codex, Windsurf, VS Code, Gemini CLI, claude.ai and ChatGPT.
  • No sign-up to connect: add https://taskhold.com/api/mcp to your agent, and it gives you the trial link the first time it needs it.
  • Your code stays on your machine. Agents send only task text, notes and file paths.
  • One price covers every agent you run. No per-agent or per-seat add-ons.
  • Nothing is charged for 7 days. Cancel from Billing any time.
  • Listed in the Official MCP Registry as com.taskhold/taskhold.

Not a fit if

  • You want one workspace for a whole team. Taskhold is one list per person; teams subscribe per person.
  • You want it free. A TODO.md in the repo or Claude Code's built-in task list costs nothing.
  • Your tasks should sit next to pull requests. GitHub Issues does that better.
  • You need a sync with Jira or Linear. It doesn't exist yet.
  • Your agent can't use MCP tools. Taskhold talks to agents only through MCP.

Questions

Can I see how far along an agent is?

Yes. Agents split a task into steps with taskhold_plan and report progress with taskhold_progress. The live board shows a bar per task, the step running and what the agent says it is doing. It also shows how long it has run and an estimate of time left. It updates as it happens, and a desktop notification tells you when an agent is blocked or done.

Isn't the progress just what the agent says?

Yes. Progress is self-reported, like a teammate's standup. The board also shows when each agent last reported, so a session that died stands out even if its last report said 90%. Use it to see where things stand and what needs you; review still happens in the diff.

Which agents does it work with?

Claude Code, Cursor, Codex, Windsurf, Cline and any MCP client that can reach a remote Streamable HTTP server. Your setup URL includes the exact config for each of them. claude.ai, Claude Desktop and ChatGPT connect with the URL https://taskhold.com/api/mcp and a sign-in.

Claude Code already has a task list. Why use this?

Claude Code's list lives inside one session on one machine. Taskhold is one list across sessions, repos, agents and machines, kept after a session ends and visible from your phone. If you run one agent in one repo, the built-in list is enough.

What do I have to install?

No software. Add https://taskhold.com/api/mcp to your agent: in Claude Code run claude mcp add --transport http taskhold https://taskhold.com/api/mcp, in claude.ai or Claude Desktop add it as a custom connector. It connects without an account, and the first time the agent uses Taskhold it gives you a link to start the 7-day trial. Prefer an API key? Make a setup URL in Settings, paste it into your agent and say "set this up"; that also adds a short CLAUDE.md workflow telling it to plan steps and report progress. Some clients need a restart before they load a new MCP server. In ChatGPT, add it under Plugins > Add > Add custom MCP server with OAuth authentication and sign in (steps: taskhold.com/integrations/chatgpt). Optional: a terminal board and a Claude Code status line, one command each (taskhold.com/install).

Is my code sent anywhere?

No. Your agent sends only what it writes into a task: titles, notes, and any file paths or line numbers it chooses to include. We never see your repository, and we don't send your tasks to any AI model.

Where is my data stored?

In a SQLite database on our own server at DigitalOcean in the US. Stripe handles payments and holds your card details. Resend delivers sign-in emails. The privacy policy lists everything we keep.

Can I self-host it?

No. Taskhold is hosted only, on our own server at DigitalOcean in the US. Your agents send only what they write into tasks; Taskhold never reads your repository.

What happens when the trial ends?

On day 7 your card is charged $12 for monthly or $120 for yearly, and the plan renews each period after that. Cancel before day 7 and you aren't charged.

Why is there no free plan?

Taskhold is one paid plan, so it can be supported properly without ads or selling data. The 7-day trial is how you try it: card up front, and nothing is charged if you cancel before day 7. If a TODO.md in your repo already covers you, keep it. Taskhold is for when you want progress, history and your phone in the loop.

How do I cancel?

Open Settings, then Billing, then Manage billing, and cancel in the Stripe portal. You keep access until the end of the period you paid for.

What happens if I hit a limit?

Nothing is charged. If an agent sends more than 120 progress reports in a minute, the extra ones are refused for a moment and the agent is told to report less often. If you reach a fair-use ceiling (500 projects, 100,000 tasks, 50 API keys), new ones are refused with a message; email hello@greatwork.company if you need more.

Can my team use it?

Each person subscribes and gets their own list. Shared team workspaces aren't built, so teammates can't see each other's tasks. If your team needs that, email hello@greatwork.company.

Can I bring my tasks over from another app?

Yes. Open Import and connect Todoist, TickTick, Linear, GitHub, GitLab, Jira, Asana, ClickUp or Notion with a token, or drop an export from Trello, Google Tasks, Things, Reminders, OmniFocus, a TODO.md, todo.txt, CSV or Excel file. You see every task before it's saved, and one click undoes the import.

Can I export my data?

Yes. In Settings, Download my data saves everything in your account as one JSON file: tasks with their steps, notes and history, projects, labels and the activity feed. The REST API returns the same file with your API key (GET /api/account/export; see the REST API docs at /docs/rest-api).

What happens to my tasks if Taskhold shuts down?

It's extremely unlikely. If it ever happened, we'd email every account at least 90 days before, and you could export everything from Settings until the last day.

How are API keys kept safe?

We store a SHA-256 hash of each key, so the full key is shown only once. A setup URL keeps an encrypted copy of its key until the link expires. Use one key per machine and revoke any of them from Settings.

Next time an agent runs a long task, watch it on the board

Start the trial, paste one setup link into your agent, and its next plan shows up on the live board.

Start my 7-day trial

$12/month or $120/year after the trial. Card up front. Cancel before day 7 and you pay nothing.